fix(tier): honor force for lifecycle references

fix: provide cross-repo token for auto-testing checkout (#6831 )
The test workflows checkout the private rustfs/auto-testing repository, but the default GITHUB_TOKEN only has access to rustfs/rustfs, so every checkout failed with 'repository ... not found' (nightly runs on 2026-08-28). Pass secrets.PF_TESTING_GH_TOKEN (the existing cross-repo PAT already used by the performance workflow) to the auto-testing checkout steps in all three workflows.
2026-08-29 12:41:17 +08:00 · 2026-08-29 12:00:19 +08:00 · 2026-08-29 03:25:05 +00:00 · 2026-08-28 17:34:04 +00:00 · 2026-08-28 17:31:10 +00:00 · 2026-08-28 17:12:30 +00:00
655 changed files with 94434 additions and 39931 deletions
@@ -1,12 +1,12 @@
 ---
 name: rustfs-release-publish
-description: "Run the end-to-end RustFS console gate, version bump, preview validation, and final-tag publication pipeline. Use only when the user explicitly asks to release or publish a RustFS version (发版/发布)."
+description: "Run the end-to-end RustFS console gate, version bump, preview validation, human confirmation, and final-tag publication pipeline. Use only when the user explicitly asks to release or publish a RustFS version (发版/发布)."
 ---
 # RustFS Release Publish (preview-validated pipeline)

 This skill orchestrates a full release. It wraps `rustfs-release-version-bump` (which only edits version files and opens the PR) with a mandatory preview-tag validation loop before the final tag is published.

-Core design: **version files never carry a `-preview.N` suffix**. The preview suffix exists only in tag names. A preview tag creates a visible GitHub Release marked Prerelease and uploads versioned assets, but it never becomes GitHub Latest and never updates `*-latest`, `latest.json`, R2, Docker, or Helm channels. This works because the binary self-reports the git tag it was built from (`build::TAG` via shadow_rs, see `rustfs/src/config/cli.rs` `SHORT_VERSION`), and `build.yml` derives artifact names and preview classification from the tag name — Cargo.toml's version is only a no-tag fallback. Therefore the preview tag and the final tag can (and MUST) point at the exact same commit: what you validated is byte-for-byte the source that ships.
+Core design: **version files never carry a `-preview.N` suffix**. The preview suffix exists only in tag names. A preview tag creates a visible GitHub Release marked Prerelease and uploads versioned assets, but it never becomes GitHub Latest and never updates `*-latest`, `latest.json`, R2, Docker, or Helm channels. That Release is temporary: `build.yml` deletes it automatically once the final tag's Release is published, so the Releases page ends up carrying deliverables only while the `-preview.N` tags stay behind as the traceability record. This works because the binary self-reports the git tag it was built from (`build::TAG` via shadow_rs, see `rustfs/src/config/cli.rs` `SHORT_VERSION`), and `build.yml` derives artifact names and preview classification from the tag name — Cargo.toml's version is only a no-tag fallback. Therefore the preview tag and the final tag can (and MUST) point at the exact same commit: what you validated is byte-for-byte the source that ships.

 Pipeline shape:

@@ -17,7 +17,9 @@ check console main against its latest Release
  -> tag <preview-tag> at that commit -> CI green
  -> verify preview Release assets -> run binary locally + console checks
  -> validate with latest rc client
+  -> report preview acceptance results -> STOP for explicit human confirmation
  -> tag <target> at the SAME commit (zero delta) -> re-verify CI/release
+  -> CI deletes the <target>-preview.N Releases (tags kept)
 ```

 On validation failure: fix lands on main via normal PR (version files are already at `<target>`, no new bump PR), then tag `<preview-tag N+1>` at the new main commit and restart from Phase 2.
@@ -50,16 +52,20 @@ Rules:
 - Use `<target>-preview.N` for every target, e.g. `1.0.0-beta.10-preview.3` or `1.1.0-preview.1`.
 - The canonical suffix is exactly `-preview.<digits>`. `build.yml` recognizes it before alpha/beta/rc classification and routes it to the preview-only path; any other tag containing `-preview` fails closed instead of being treated as a release.
 - A preview Release MUST be published with `isPrerelease=true` and `isLatest=false`. Any `*-latest` preview asset or preview-triggered `latest.json`, R2, Docker, or Helm publication is a pipeline failure.
+- Preview Releases are cleaned up by the `cleanup-preview-releases` job after `publish-release` succeeds for the deliverable tag. It deletes every Release whose tag is exactly `<target>-preview.<digits>` and never passes `--cleanup-tag`, so the tags survive.

 ## Hard rules

 - Version files (Cargo.toml, Cargo.lock, README, flake.nix, Chart.yaml, rustfs.spec) are bumped ONCE, directly to `<target>`. Never write a `-preview.N` suffix into any version file. If `rustfs-release-version-bump` is ever asked for a `-preview` version, that is a pipeline bug — stop.
- Preview Release assets are versioned and intentionally visible on the Releases page. Do not label them Latest or use them to update any latest distribution channel.
+- Preview Release assets are versioned and intentionally visible on the Releases page for the duration of validation. Do not label them Latest or use them to update any latest distribution channel.
+- Never delete a preview Release by hand before Phase 6 finishes — Phase 4 downloads its assets and the final Release notes are generated while it still exists. Cleanup is CI's job; only step in manually (`gh release delete "<preview-tag>" --yes`, never `--cleanup-tag`) if `cleanup-preview-releases` failed.
 - Tags have no `v` prefix. Always annotated: `git tag -a <tag> -m "Release <tag>"`.
 - The final tag MUST point at exactly `PREVIEW_HASH` — the commit the validated preview tag points at. Never tag current `main` HEAD (commits merged after validation are unvalidated), and never create an extra version-bump commit between preview and final.
- When a previous deliverable exists, GitHub Release notes for the preview and final tags MUST use it as their shared comparison baseline: the most recently published non-preview Release before the target. Internal `-preview.N` Releases are explicitly excluded from that selection, even when they point at the same commit as the final tag. If no previous deliverable exists, omit `previous_tag_name` and record that GitHub's default baseline fallback was used.
+- When a previous deliverable exists, GitHub Release notes for the preview and final tags MUST use it as their shared comparison baseline: the most recently published non-preview Release before the target. Internal `-preview.N` Releases are explicitly excluded from that selection, even when they point at the same commit as the final tag — cleanup runs after the notes are generated, so the preview Release is still present and would otherwise be picked as the baseline. If no previous deliverable exists, omit `previous_tag_name` and record that GitHub's default baseline fallback was used.
 - Generated Release notes carry a workflow-management marker so retries can repair them. Before manually curating a generated body, remove that marker; unmarked non-placeholder notes are preserved by later workflow runs.
 - Phases run in order; a failure in any phase blocks everything after it. After the fix lands on main, restart from Phase 2 with the next preview iteration against the new `origin/main` hash — do not resume mid-pipeline against a stale hash.
+- Completing preview acceptance does not authorize the final tag. After Phases 3–5 pass, report the acceptance evidence and stop until the user explicitly confirms continuation. The original release request, an earlier confirmation, silence, or an automated follow-up does not satisfy this gate.
+- Confirmation is scoped to the reported `<target>`, `<preview-tag>`, and `PREVIEW_HASH`. A failed or repeated acceptance cycle, including any new preview iteration, invalidates prior confirmation and requires a new one.
 - If the release is abandoned after Phase 1 merged, main's version files claim a version that was never tagged. Either revert the bump PR or leave it to be overwritten by the next release — but tell the user explicitly and record the decision.
 - User-facing status updates in Chinese; commits, PR titles/bodies, and tag messages in English. No hard-wrapping in commit messages, PR bodies, or documentation prose — one logical line per sentence/paragraph, let soft wrap handle display.

@@ -207,6 +213,12 @@ rc alias remove preview

 - Any FAIL blocks the release. Afterwards stop the server and delete the scratch data directory.

+### Manual confirmation gate
+
+After every Phase 3–5 check passes, report the target, preview tag, `PREVIEW_HASH`, preview Release URL, console result, and rc matrix, then explicitly ask the user whether to publish the final tag. End the turn without creating or pushing `<target>`.
+
+Continue to Phase 6 only after a new user reply explicitly confirms the reported target, preview tag, and commit. A clear affirmative reply to that exact report, such as `确认继续`, is sufficient; if the reply is ambiguous or any reported value changed, ask again.
+
 ## Phase 6 — Publish the final tag on the validated commit

 No second version bump, no release branch. The final tag goes on the exact commit the preview validated:
@@ -221,6 +233,7 @@ git push origin "<target>"
 - CI rebuilds from the same source; the only changed input is the tag name, so the binary now self-reports `<target>`.
 - Verify the final tag's complete publication path: all matrix and release jobs green; `gh release view "<target>"` shows the full versioned and `-latest` asset set plus checksums, SBOM, and provenance; Docker and Helm workflows succeed; `latest.json` points to `<target>`. A stable target must have `isPrerelease=false` and `isLatest=true`. An alpha/beta/rc target must have `isPrerelease=true`; GitHub does not permit prereleases to be Latest, but the project `latest.json` still advances to the final non-preview target.
 - Verify the final Release body contains `## What's Changed` and a Full Changelog link. When `PREVIOUS_DELIVERABLE` exists, the link MUST be `https://github.com/rustfs/rustfs/compare/<PREVIOUS_DELIVERABLE>...<target>` and the baseline MUST equal the preview Release baseline; for example, both `1.0.0-beta.12-preview.1` and `1.0.0-beta.12` compare from `1.0.0-beta.11`.
+- Verify the preview cleanup: `cleanup-preview-releases` must succeed, `gh release view "<preview-tag>"` must then report `release not found` for every preview iteration of this target, and `git rev-parse "<preview-tag>^{commit}"` must still resolve to `PREVIEW_HASH` (the tag is kept). If the job failed, delete the leftover Releases manually with `gh release delete "<preview-tag>" --yes` and report it.
 - Optionally spot-check `./rustfs --version` from a final-tag artifact — it must report `<target>`.

 ## Output contract
@@ -229,5 +242,6 @@ Always report:

 - Console gate result: previous/latest Console tags, whether merged changes required a release, `CONSOLE_HASH`, and Console run/Release URLs when a release was published.
 - Target version, preview tag(s) used, `PREVIEW_HASH` (which both tags point at).
- Per-phase result (PASS/FAIL/BLOCKED) with key evidence: preview and final Release URLs, preview `isPrerelease`/`isLatest` state, final latest-channel state, console check results, and the rc command matrix.
+- Manual confirmation gate status (`WAITING_FOR_CONFIRMATION` or `CONFIRMED`) and its exact target, preview tag, and `PREVIEW_HASH`.
+- Per-phase result (PASS/FAIL/BLOCKED) with key evidence: preview and final Release URLs, preview `isPrerelease`/`isLatest` state, final latest-channel state, console check results, the rc command matrix, and the preview-Release cleanup result (deleted Releases plus surviving tags).
 - Any deviation from this pipeline and why the user approved it.
@@ -1,2 +1,2 @@
-sha256-darwin=ae412c27e2e43f3fda3e0c21ef616b48a6403354faab8db8d4154202e3c4eec5
-sha256-linux=ef3be856bd3257c2c369428f66a48ad073dad8187229fe95840a600a81edf22b
+sha256-darwin=d6aa36cfaae2c4d8590482c7e47138c5965b335b34a75f50d11ffc3366e9021e
+sha256-linux=96db8060fce98addda4f69092d297ca236bec4892d820617a26a261eedac61b0
@@ -1 +1 @@
-sha256=655a3f3c1d042e694339d15caba7580518320322d1bac0f09450b37e6c09e2e7
+sha256=8d5517f5f2fc32d561782dfccd51b7f746f5e25b2835e37e100c883f7f18777d
@@ -1 +1 @@
-sha256=ec27cde6ce6400723c4b372bfbd2ac61709c744294e4810af765e8a808d8e31d
+sha256=294350518743cac8d7c41880a2835216e4b697908d7b0b1bc92b62816d94c59d
@@ -45,6 +45,11 @@ logging-guardrails-check: ## Check logging guardrails for redaction and noise re
 	@echo "🪵 Checking logging guardrails..."
 	./scripts/check_logging_guardrails.sh

+.PHONY: error-other-ratchet-check
+error-other-ratchet-check: ## Check the ecstore ::other(format!) quorum-bucketing ratchet stays shrink-only
+	@echo "🪣 Checking error other(format!) ratchet..."
+	./scripts/check_error_other_format_ratchet.sh
+
 .PHONY: tokio-io-uring-check
 tokio-io-uring-check: ## Check tokio io-uring runtime feature stays removed
 	@echo "🚫 Checking tokio io-uring feature guard..."
@@ -75,6 +80,11 @@ embedded-secrets-check: ## Check no private key material or credential literal i
 	@echo "🔑 Checking embedded secret material guard..."
 	./scripts/check_embedded_secrets.sh

+.PHONY: offline-enrollment-e2e-check
+offline-enrollment-e2e-check: core-deps ## Build and exercise the dedicated offline enrollment E2E root
+	@echo "🔐 Checking the offline enrollment E2E root boundary..."
+	./scripts/check_offline_enrollment_e2e.sh
+
 .PHONY: test-wiring-check
 test-wiring-check: ## Check tests stay registered and selected by their intended runners
 	@echo "🧪 Checking test wiring..."
@@ -19,13 +19,13 @@ planning-docs-check: ## Check that no planning-type documents are committed
 	./scripts/check_no_planning_docs.sh

 .PHONY: pre-commit
-pre-commit: fmt-check unsafe-code-check architecture-migration-check logging-guardrails-check tokio-io-uring-check extension-schema-check body-cache-whitelist-check s3s-footprint-check fips-wording-check embedded-secrets-check test-wiring-check doc-paths-check planning-docs-check quick-check ## Run fast pre-commit checks without clippy/full tests
+pre-commit: fmt-check unsafe-code-check architecture-migration-check logging-guardrails-check error-other-ratchet-check tokio-io-uring-check extension-schema-check body-cache-whitelist-check s3s-footprint-check fips-wording-check embedded-secrets-check test-wiring-check doc-paths-check planning-docs-check quick-check ## Run fast pre-commit checks without clippy/full tests
 	@echo "✅ All pre-commit checks passed!"

 .PHONY: pre-pr
-pre-pr: fmt-check unsafe-code-check architecture-migration-check logging-guardrails-check tokio-io-uring-check extension-schema-check body-cache-whitelist-check s3s-footprint-check fips-wording-check embedded-secrets-check test-wiring-check doc-paths-check planning-docs-check log-analyzer-rules-check clippy-check test ## Run full pre-PR checks with clippy and tests
+pre-pr: fmt-check unsafe-code-check architecture-migration-check logging-guardrails-check error-other-ratchet-check tokio-io-uring-check extension-schema-check body-cache-whitelist-check s3s-footprint-check fips-wording-check embedded-secrets-check test-wiring-check doc-paths-check planning-docs-check log-analyzer-rules-check offline-enrollment-e2e-check clippy-check test ## Run full pre-PR checks with clippy and tests
 	@echo "✅ All pre-PR checks passed!"

 .PHONY: dev-check
-dev-check: fmt-check unsafe-code-check architecture-migration-check logging-guardrails-check tokio-io-uring-check extension-schema-check body-cache-whitelist-check s3s-footprint-check fips-wording-check embedded-secrets-check test-wiring-check doc-paths-check planning-docs-check quick-check ## Run fast local development checks
+dev-check: fmt-check unsafe-code-check architecture-migration-check logging-guardrails-check error-other-ratchet-check tokio-io-uring-check extension-schema-check body-cache-whitelist-check s3s-footprint-check fips-wording-check embedded-secrets-check test-wiring-check doc-paths-check planning-docs-check quick-check ## Run fast local development checks
 	@echo "✅ Fast development checks passed!"
@@ -34,6 +34,7 @@ script-tests: ## Run shell script tests
 	./scripts/test_exact_1mib_handoff_abba.sh
 	./scripts/test_pinned_paired_abba_bench.sh
 	./scripts/test_manual_transition_runbooks.sh
+	./scripts/test_fuzz_runner.sh
 	./scripts/check_embedded_secrets.sh --self-test
 	python3 ./scripts/check_test_wiring.py --self-test
 	python3 ./scripts/check_security_coverage.py --self-test
@@ -1,5 +1,7 @@
 # nextest configuration for RustFS.
 #
+experimental = ["setup-scripts"]
+
 # Serialize the ecstore tests that share the process-wide disk registry or
 # exercise a multi-disk commit handoff across nextest process boundaries.
 #
@@ -44,7 +46,27 @@ e2e-reliability = { max-threads = 1 }
 e2e-inline-boundaries = { max-threads = 1 }
 e2e-cluster-nightly = { max-threads = 1 }

+# These exact regression scenarios build deep async storage futures that exceed
+# libtest's 2 MiB spawned-thread stack on Linux. Give only their test processes
+# the same 32 MiB stack already used by the crate's dedicated large-stack tests.
+[scripts.setup.ecstore-large-stack]
+command = ['sh', '-c', 'echo RUST_MIN_STACK=33554432 >> "$NEXTEST_ENV"']
+
+# The serial ILM selection builds the same deep storage futures in both the
+# lifecycle transition module and scanner integration binary. Different tests
+# in each have overflowed first across otherwise unrelated CI runs.
+[scripts.setup.lifecycle-large-stack]
+command = ['sh', '-c', 'echo RUST_MIN_STACK=33554432 >> "$NEXTEST_ENV"']
+
 # --- default profile (local): serialize the flaky groups, never retry --------
+[[profile.default.scripts]]
+filter = 'package(rustfs-ecstore) & test(/^(bucket::lifecycle::bucket_lifecycle_ops::tests::manual_transition_worker_result_recovery_marks_unknown_for_corrupt_marker|services::rebalance::entry::tests::real_rebalance_run_fence_loss_blocks_multipart_publication|store::init::tests::(decommission_entry_(allows_free_version_consumed_before_source_lock|rejects_subquorum_free_version_conflict_and_retains_source|skips_cleanup_only_marker_when_free_version_is_present)|prepared_tier_delete_recovery_(checks_later_pool_then_commits_after_source_removal|finds_directory_source_on_encoded_set|retains_journal_on_source_metadata_error)|tier_mutation_peer_handler_applies_prepare_commit_and_abort_idempotently|transition_response_loss_persists_unknown_outcome_for_provider_recovery|transition_transaction_recovery_(drops_record_after_confirmed_local_commit|keeps_cleanup_pending_local_commit)))$/)'
+setup = 'ecstore-large-stack'
+
+[[profile.default.scripts]]
+filter = 'binary(lifecycle_integration_test) | (package(rustfs) & test(/^app::lifecycle_transition_api_test::/))'
+setup = 'lifecycle-large-stack'
+
 [[profile.default.overrides]]
 filter = 'package(rustfs-ecstore) & (test(concurrent_resend_same_part_commits_one_generation) | test(concurrent_config_writes_from_separate_nodes_do_not_lose_writes) | test(/^store::bucket::tests::bucket_delete_(mark_delete|purge_removes|default_s3_delete)/))'
 test-group = 'ecstore-serial-flaky'
@@ -113,9 +135,10 @@ filter = 'package(e2e_test) & test(/^inline_fast_path_cluster_test::/)'
 test-group = 'e2e-inline-boundaries'

 # Vault KMS tests share the fixed dev-server port 8200. serial_test's #[serial]
-# does not cross nextest process boundaries, so keep these tests in one group.
+# does not cross nextest process boundaries, so keep every Vault-backed test in
+# one group.
 [[profile.default.overrides]]
-filter = 'package(e2e_test) & test(/^kms::kms_vault_test::/)'
+filter = 'package(e2e_test) & (test(/^kms::kms_vault_test::/) | test(/^kms::kms_rekey_sweep_test::/) | test(/^kms::configured_roundtrip_test::test_configured_vault_kms_admin_and_versioned_cleanup$/))'
 test-group = 'e2e-vault'

 # ---------------------------------------------------------------------------
@@ -133,6 +156,14 @@ fail-fast = false
 # marker is the observable signal the flake policy is built around.
 path = "junit.xml"

+[[profile.ci.scripts]]
+filter = 'package(rustfs-ecstore) & test(/^(bucket::lifecycle::bucket_lifecycle_ops::tests::manual_transition_worker_result_recovery_marks_unknown_for_corrupt_marker|services::rebalance::entry::tests::real_rebalance_run_fence_loss_blocks_multipart_publication|store::init::tests::(decommission_entry_(allows_free_version_consumed_before_source_lock|rejects_subquorum_free_version_conflict_and_retains_source|skips_cleanup_only_marker_when_free_version_is_present)|prepared_tier_delete_recovery_(checks_later_pool_then_commits_after_source_removal|finds_directory_source_on_encoded_set|retains_journal_on_source_metadata_error)|tier_mutation_peer_handler_applies_prepare_commit_and_abort_idempotently|transition_response_loss_persists_unknown_outcome_for_provider_recovery|transition_transaction_recovery_(drops_record_after_confirmed_local_commit|keeps_cleanup_pending_local_commit)))$/)'
+setup = 'ecstore-large-stack'
+
+[[profile.ci.scripts]]
+filter = 'binary(lifecycle_integration_test) | (package(rustfs) & test(/^app::lifecycle_transition_api_test::/))'
+setup = 'lifecycle-large-stack'
+
 # ===========================================================================
 # QUARANTINE — flaky tests granted retries = 2 under the ci profile ONLY.
 #
@@ -165,6 +196,15 @@ test-group = 'ecstore-serial-flaky'
 filter = 'package(rustfs-ecstore) & test(walk_dir_does_not_charge_consumer_backpressure_to_the_stall_budget)'
 retries = 2

+# Serialize the relocated-pool GET resume regression under the ci profile too
+# (see the matching default-profile override near the top). No longer a
+# quarantine: the fixture race (rustfs#6701/rustfs#6703) was fixed by #6707,
+# which made the staging tolerate quorum-tolerated disk gaps; only the 8-disk
+# cross-disk-IO serialization remains.
+[[profile.ci.overrides]]
+filter = 'package(rustfs) & test(execute_get_object_resumes_from_relocated_pool_without_splicing_body)'
+test-group = 'ecstore-serial-flaky'
+
 # Serialize the 4-disk reliability / degraded-read e2e tests under the ci
 # profile too (see the e2e-reliability test-group note near the top). Not a
 # quarantine: no retries, just single-threaded so several 4-disk servers never
@@ -180,10 +220,6 @@ test-group = 'e2e-reliability'
 filter = 'package(rustfs-ecstore) & test(/^set_disk::ops::multipart::tests::crash_consistency::/)'
 test-group = 'ecstore-serial-flaky'

-[[profile.ci.overrides]]
-filter = 'package(rustfs) & test(execute_get_object_resumes_from_relocated_pool_without_splicing_body)'
-test-group = 'ecstore-serial-flaky'
-
 # Match the default-profile embedded test isolation without quarantining or
 # retrying failures in CI.
 [[profile.ci.overrides]]
@@ -409,7 +445,7 @@ path = "junit.xml"
 #     `e2e-smoke` (20 fast) and `e2e-repl-nightly` (56 slow) lanes and reserves
 #     it for those, so e2e-full does not double-run it.
 #   * #[ignore]d tests — nextest skips them by default (no --run-ignored); the
-#     manual-localhost:9000 reliant/policy tests are ci-13's migration.
+#     manual-localhost:9000 reliant tests are ci-13's migration.
 #
 # Each e2e test spawns its own single-node rustfs server on a random port with
 # an isolated temp dir (crates/e2e_test/src/common.rs), so the set is
@@ -453,5 +489,5 @@ filter = 'package(e2e_test) & test(/^inline_fast_path_cluster_test::/)'
 test-group = 'e2e-inline-boundaries'

 [[profile.e2e-full.overrides]]
-filter = 'package(e2e_test) & test(/^kms::kms_vault_test::/)'
+filter = 'package(e2e_test) & (test(/^kms::kms_vault_test::/) | test(/^kms::kms_rekey_sweep_test::/) | test(/^kms::configured_roundtrip_test::test_configured_vault_kms_admin_and_versioned_cleanup$/))'
 test-group = 'e2e-vault'
@@ -9,4 +9,4 @@
 # if the selected count drops below this number, so a rename or removal that
 # thins the security smoke gate must update this file in the same PR.
 # Adding tests does not require a bump, but bumping keeps the guard tight.
-16
+18
@@ -0,0 +1,84 @@
+# Copyright 2024 RustFS Team
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+global:
+  scrape_interval: 15s # Evaluate rules every 15 seconds. The default is every 1 minute.
+  evaluation_interval: 15s
+  external_labels:
+    cluster: 'rustfs-dev'  # Label to identify the cluster
+    replica: '1'  # Replica identifier
+
+rule_files:
+  - /etc/prometheus/rules/*.yml
+
+scrape_configs:
+  - job_name: 'otel-collector'
+    static_configs:
+      - targets: [ 'otel-collector:8888' ]  # Scrape metrics from Collector
+    scrape_interval: 10s
+
+  - job_name: 'rustfs-app-metrics'
+    static_configs:
+      - targets: [ 'otel-collector:8889' ]  # Application indicators
+    scrape_interval: 15s
+    metric_relabel_configs:
+      - source_labels: [ __name__ ]
+        regex: 'go_.*'
+        action: drop  # Drop Go runtime metrics if not needed
+
+  - job_name: 'tempo'
+    static_configs:
+      - targets: [ 'tempo:3200' ]  # Scrape metrics from Tempo
+
+  - job_name: 'jaeger'
+    static_configs:
+      - targets: [ 'jaeger:14269' ]  # Jaeger admin port (14269 is standard for admin/metrics)
+
+  - job_name: 'loki'
+    static_configs:
+      - targets: [ 'loki:3100' ]
+
+  - job_name: 'prometheus'
+    static_configs:
+      - targets: [ 'localhost:9090' ]
+
+  - job_name: 'vulture'
+    static_configs:
+      - targets:
+          - 'vulture:8080'
+
+otlp:
+  promote_resource_attributes:
+    - service.instance.id
+    - service.name
+    - service.namespace
+    - cloud.availability_zone
+    - cloud.region
+    - container.name
+    - deployment.environment.name
+    - k8s.cluster.name
+    - k8s.container.name
+    - k8s.cronjob.name
+    - k8s.daemonset.name
+    - k8s.deployment.name
+    - k8s.job.name
+    - k8s.namespace.name
+    - k8s.pod.name
+    - k8s.replicaset.name
+    - k8s.statefulset.name
+  translation_strategy: NoUTF8EscapingWithSuffixes
+
+storage:
+  tsdb:
+    out_of_order_time_window: 30m
@@ -5,3 +5,4 @@ self-hosted-runner:
    - sm-standard-2
    - sm-standard-4
    - dind-sm-standard-2
+    - smoke-testing
@@ -7,8 +7,16 @@
  { "workflow": ".github/workflows/e2e-s3tests.yml", "max_age_hours": 192 },
  { "workflow": ".github/workflows/fuzz.yml", "max_age_hours": 36 },
  { "workflow": ".github/workflows/mint.yml", "max_age_hours": 192 },
-  { "workflow": ".github/workflows/minio-interop.yml", "max_age_hours": 36 },
+  {
+    "workflow": ".github/workflows/minio-interop.yml",
+    "max_age_hours": 36,
+    "never_ran_grace_until": "2026-09-08T00:00:00Z"
+  },
  { "workflow": ".github/workflows/nightly-gnu.yml", "max_age_hours": 36 },
  { "workflow": ".github/workflows/performance-ab.yml", "max_age_hours": 36 },
-  { "workflow": ".github/workflows/runner-hygiene.yml", "max_age_hours": 792 }
+  {
+    "workflow": ".github/workflows/runner-hygiene.yml",
+    "max_age_hours": 792,
+    "never_ran_grace_until": "2026-09-02T06:37:00Z"
+  }
 ]
@@ -1033,6 +1033,55 @@ jobs:
          echo "🎉 Released $TAG successfully!"
          echo "📄 Release URL: ${{ needs.create-release.outputs.release_url }}"

+  # Remove the internal preview releases once the deliverable release is live.
+  # Only the Releases are deleted; the -preview.N tags stay so the validated
+  # commit remains traceable.
+  cleanup-preview-releases:
+    name: Cleanup Preview Releases
+    needs: [ build-check, publish-release ]
+    if: startsWith(github.ref, 'refs/tags/') && (needs.build-check.outputs.build_type == 'release' || needs.build-check.outputs.build_type == 'prerelease')
+    runs-on: ubuntu-latest
+    timeout-minutes: 30
+    permissions:
+      contents: write
+    steps:
+      - name: Delete preview releases for this target
+        env:
+          GH_TOKEN: ${{ github.token }}
+        shell: bash
+        run: |
+          set -euo pipefail
+
+          TAG="${{ needs.build-check.outputs.version }}"
+          RELEASES_JSON="${RUNNER_TEMP}/releases.json"
+
+          # Fetch before filtering: a failed listing must abort here instead of
+          # looking like "nothing to clean up".
+          gh api --paginate "repos/${GITHUB_REPOSITORY}/releases?per_page=100" > "$RELEASES_JSON"
+
+          # Match only <target>-preview.<digits>. String operations, not a
+          # regex over the tag, so dots in the version cannot widen the match.
+          DELETED=0
+          while IFS= read -r preview_tag; do
+            [[ -n "$preview_tag" ]] || continue
+            echo "🧹 Deleting preview release $preview_tag (tag kept)"
+            gh release delete "$preview_tag" --repo "${GITHUB_REPOSITORY}" --yes
+            DELETED=$((DELETED + 1))
+          done < <(
+            jq -r --arg tag "$TAG" '
+              .[]
+              | select(.tag_name | startswith($tag + "-preview."))
+              | select(.tag_name | ltrimstr($tag + "-preview.") | test("^[0-9]+$"))
+              | .tag_name
+            ' "$RELEASES_JSON"
+          )
+
+          if [[ "$DELETED" -eq 0 ]]; then
+            echo "ℹ️  No preview releases to clean up for $TAG"
+          else
+            echo "✅ Removed $DELETED preview release(s) for $TAG"
+          fi
+
  alert-on-failure:
    name: Alert on scheduled failure
    needs: [build-check, prepare-platform-matrix, build-rustfs, build-summary]
@@ -212,6 +212,9 @@ jobs:
          install-build-packaging-tools: 'false'

      - name: Build ci-feat-rio superset
+        env:
+          # --all-targets links the same test binaries as the reader lane.
+          CARGO_BUILD_JOBS: "2"
        run: |
          cargo build -p rustfs -p rustfs-ecstore --all-targets --features rio-v2
          cargo build -p rustfs --bins --features rio-v2,e2e-test-hooks
@@ -240,6 +243,9 @@ jobs:
          install-build-packaging-tools: 'false'

      - name: Build ci-feat-proto superset
+        env:
+          # Avoid an unbounded burst of concurrent test-binary links (#5394).
+          CARGO_BUILD_JOBS: "2"
        run: |
          cargo build -p rustfs -p rustfs-protocols --all-targets --features swift
          cargo build -p rustfs -p rustfs-protocols --all-targets --features sftp
@@ -142,6 +142,9 @@ jobs:
      - name: Check logging guardrails
        run: ./scripts/check_logging_guardrails.sh

+      - name: Check error other(format!) ratchet
+        run: ./scripts/check_error_other_format_ratchet.sh
+
      - name: Check tokio io-uring feature guard
        run: ./scripts/check_no_tokio_io_uring.sh

@@ -181,22 +184,14 @@ jobs:
    needs: [ quick-checks ]
    runs-on: sm-standard-4
    timeout-minutes: 90
-    # Both lines are required. Job-level `permissions` replaces the workflow
-    # block rather than merging with it, so declaring only `actions: write`
-    # would drop `contents: read` and break this job's checkout and the
-    # repo-token the setup action hands to setup-protoc.
-    permissions:
-      contents: read
-      actions: write
    env:
      FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
    steps:
      - name: Checkout repository
        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
        with:
-          # This job's token can cancel runs and delete Actions caches. Checkout
-          # otherwise writes it into .git/config, where a PR's own build.rs or
-          # proc-macro could read it back out.
+          # Checkout otherwise writes the token into .git/config, where a PR's
+          # own build.rs or proc-macro could read it back out.
          persist-credentials: false

      - name: Setup Rust environment
@@ -212,6 +207,9 @@ jobs:
          cache-save-if: 'false'
          install-build-packaging-tools: 'false'

+      - name: Protect Connect test home
+        run: chmod go-w "$(realpath "$HOME")"
+
      - name: Prepare test evidence
        run: |
          mkdir -p artifacts/test-and-lint
@@ -310,6 +308,9 @@ jobs:
          } > artifacts/test-and-lint/doctest-diagnostics.txt
          exit "${status}"

+      - name: Check offline enrollment E2E root boundary
+        run: ./scripts/check_offline_enrollment_e2e.sh
+
      - name: Upload test reports and diagnostics
        if: always()
        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
@@ -344,41 +345,36 @@ jobs:
      - name: Run rebalance/decommission migration proofs
        run: ./scripts/check_migration_gate_count.sh

-      # Early stop. Once this job has failed the PR cannot merge, so the sibling
-      # lanes are burning runners on a result nobody can act on: on run
-      # 30674613104 three lanes had already failed while Test and Lint and the
-      # rio-v2 variant kept going past 70 minutes.
-      #
-      # Only this job may cancel. The lanes that are NOT required checks
-      # (protocols, ILM, e2e, s3-tests) must never hold that power: a flake in
-      # one of them would turn the required "Test and Lint" into `cancelled`,
-      # which blocks the merge. Today a maintainer can merge with sftp red, and
-      # that has to stay true.
-      #
-      # These steps run last so the `if: always()` artifact upload above still
-      # captures logs and diagnostics before the run goes away.
+      # Record the reason before this job completes as FAILURE. A separate
+      # dependent job cancels sibling lanes only after GitHub has preserved this
+      # required check's failure verdict.
      - name: Annotate early-stop reason
-        if: failure() && github.event_name == 'pull_request'
-        run: |
-          {
-            echo "## CI early-stop"
-            echo "Job \`${GITHUB_JOB}\` (Test and Lint) failed; cancelling run ${GITHUB_RUN_ID} to free runners."
-            echo "Sibling jobs showing **cancelled** were stopped by this job, not by their own failure."
-          } >> "$GITHUB_STEP_SUMMARY"
-
-      # curl rather than `gh`: every existing `gh` call in this repo runs on
-      # ubuntu-latest, and the sm-standard-* images are custom and trimmed (they
-      # ship no C toolchain, see the e2e job below), so `gh` is not known to
-      # exist here.
-      #
-      # Fork PRs are excluded explicitly instead of relying on the error path:
-      # their GITHUB_TOKEN is forced read-only and job-level permissions cannot
-      # raise it, so the call would always 403. Skipping keeps their logs clean.
-      - name: Cancel run on failure (same-repo PR only)
        if: >-
          failure() && github.event_name == 'pull_request'
          && github.event.pull_request.head.repo.full_name == github.repository
-        continue-on-error: true
+        run: |
+          {
+            echo "## CI early-stop"
+            echo "Job \`${GITHUB_JOB}\` (Test and Lint) failed; a follow-up job will cancel sibling lanes to free runners."
+            echo "Sibling jobs showing **cancelled** were stopped by the early-stop follow-up, not by their own failure."
+          } >> "$GITHUB_STEP_SUMMARY"
+
+  # Preserve the required Test and Lint FAILURE verdict before stopping sibling
+  # lanes. Cancelling from inside test-and-lint changed its own conclusion to
+  # CANCELLED and hid the actionable failure in the PR checks UI.
+  cancel-after-test-and-lint-failure:
+    name: Cancel siblings after Test and Lint failure
+    if: >-
+      failure() && needs.test-and-lint.result == 'failure'
+      && github.event_name == 'pull_request'
+      && github.event.pull_request.head.repo.full_name == github.repository
+    needs: [ test-and-lint ]
+    runs-on: ubuntu-latest
+    timeout-minutes: 5
+    permissions:
+      actions: write
+    steps:
+      - name: Cancel remaining jobs
        env:
          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
        run: |
@@ -386,7 +382,7 @@ jobs:
            -H "Authorization: Bearer ${GH_TOKEN}" \
            -H "Accept: application/vnd.github+json" \
            -H "X-GitHub-Api-Version: 2022-11-28" \
-            "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/actions/runs/${GITHUB_RUN_ID}/cancel" || true
+            "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/actions/runs/${GITHUB_RUN_ID}/cancel"

  # Dedicated serial lane for the ILM / lifecycle integration tests. These tests
  # drive the object layer through process-global singletons (the GLOBAL_ENV
@@ -433,10 +429,38 @@ jobs:
      #   - test_noncurrent_{expiry,transition}_still_works_after_immediate_compensation_transition:
      #     noncurrent transition/expiry after an immediate compensation transition.
      - name: Run ignored ILM integration tests serially
+        env:
+          # Match the measured Test and Lint link budget. The default exposed
+          # all 14 pod CPUs and a cold cache spent the full 80m compiling
+          # without starting one ILM test (main run 32982910990).
+          CARGO_BUILD_JOBS: ${{ (github.event_name == 'push' || github.event_name == 'workflow_dispatch') && '3' || '2' }}
        run: |
-          cargo nextest run -j1 --run-ignored ignored-only \
+          mkdir -p artifacts/ilm-integration
+          set +e
+          NEXTEST_HIDE_PROGRESS_BAR=1 timeout --verbose --signal=TERM --kill-after=30s 80m \
+            cargo nextest run -j1 --run-ignored ignored-only \
            -p rustfs-scanner -p rustfs \
-            -E '(binary(lifecycle_integration_test) or (package(rustfs) and test(lifecycle_transition_api_test))) and not (test(test_noncurrent_expiry_still_works_after_immediate_compensation_transition) or test(test_noncurrent_transition_still_works_after_immediate_compensation_transition))'
+            -E '(binary(lifecycle_integration_test) or (package(rustfs) and test(lifecycle_transition_api_test))) and not (test(test_noncurrent_expiry_still_works_after_immediate_compensation_transition) or test(test_noncurrent_transition_still_works_after_immediate_compensation_transition))' \
+            --status-level all --final-status-level all \
+            2>&1 | tee artifacts/ilm-integration/nextest.log
+          status=${PIPESTATUS[0]}
+          {
+            echo "exit_status=${status}"
+            echo "finished_at=$(date --utc --iso-8601=seconds)"
+            echo
+            echo "Remaining test-related processes:"
+            pgrep -af 'cargo|nextest|target/.*/deps/' || true
+          } > artifacts/ilm-integration/diagnostics.txt
+          exit "${status}"
+
+      - name: Upload ILM test diagnostics
+        if: always()
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
+        with:
+          name: ilm-integration-${{ github.run_number }}-${{ github.run_attempt }}
+          path: |
+            artifacts/ilm-integration
+            target/nextest/ci/junit.xml

  test-and-lint-rio-v2:
    name: Test and Lint (rio-v2)
@@ -460,14 +484,83 @@ jobs:
          cache-save-if: 'false'
          install-build-packaging-tools: 'false'

+      - name: Protect Connect test home
+        run: chmod go-w "$(realpath "$HOME")"
+
      - name: Run rio-v2 clippy lints
        run: cargo clippy -p rustfs -p rustfs-ecstore --all-targets --features rio-v2 -- -D warnings

      - name: Run rio-v2 feature tests
+        env:
+          # Match the main nextest lane's #5394 link-I/O guard. A cold feature
+          # cache otherwise fans out enough rust-lld processes to exhaust this
+          # job's 90-minute budget before any test starts.
+          CARGO_BUILD_JOBS: "2"
        run: |
-          cargo nextest run -p rustfs -p rustfs-ecstore --features rio-v2
+          # --profile ci so the quarantine list (and its junit flaky markers)
+          # covers this leg too; the default profile is the local no-retry
+          # profile and silently ignored quarantined flakes here (rustfs#6703).
+          cargo nextest run --profile ci -p rustfs -p rustfs-ecstore --features rio-v2
          cargo test -p rustfs --doc --features rio-v2

+  connect-short-credential-boundary:
+    name: Connect Short Credential Boundary
+    if: github.event_name != 'pull_request' || github.event.action != 'closed'
+    needs: [ quick-checks ]
+    runs-on: sm-standard-4
+    timeout-minutes: 60
+    env:
+      FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          persist-credentials: false
+
+      - name: Setup Rust environment
+        uses: ./.github/actions/setup
+        with:
+          rust-version: stable
+          cache-shared-key: ci-dev
+          cache-save-if: 'false'
+          install-build-packaging-tools: 'false'
+          install-test-tools: 'false'
+
+      - name: Run short credential behavior tests
+        env:
+          CARGO_BUILD_JOBS: "2"
+        run: |
+          cargo test -p rustfs --test connect_registration \
+            --features connect-e2e-short-credentials \
+            registration_enforces_build_profile_credential_lifetime -- --exact
+          cargo test -p rustfs --test connect_registration \
+            --features connect-e2e-short-credentials \
+            rotation_waits_for_threshold_and_stops_on_revocation -- --exact
+
+      - name: Reject short credentials in release builds
+        env:
+          CARGO_BUILD_JOBS: "2"
+        run: |
+          log="$(mktemp)"
+          set +e
+          CARGO_TERM_COLOR=never cargo check -p rustfs --release \
+            --features connect-e2e-short-credentials >"$log" 2>&1
+          status=$?
+          set -e
+          cat "$log"
+          expected='error: connect-e2e-short-credentials is restricted to debug builds'
+          summary="error: could not compile \`rustfs\` (lib) due to 1 previous error"
+          expected_count="$(grep -Fxc "$expected" "$log" || true)"
+          summary_count="$(grep -Fc "$summary" "$log" || true)"
+          error_count="$(grep -Ec '^error(:|\[)' "$log" || true)"
+          if [ "$status" -ne 101 ] || [ "$expected_count" -ne 1 ] \
+            || [ "$summary_count" -ne 1 ] || [ "$error_count" -ne 2 ]; then
+            echo "release feature gate did not fail solely at the expected compile_error" >&2
+            rm -f "$log"
+            exit 1
+          fi
+          rm -f "$log"
+
  test-and-lint-protocols:
    name: "Test and Lint (${{ matrix.features.name }})"
    if: github.event_name != 'pull_request' || github.event.action != 'closed'
@@ -504,13 +597,23 @@ jobs:
          cache-save-if: 'false'
          install-build-packaging-tools: 'false'

+      - name: Protect Connect test home
+        run: chmod go-w "$(realpath "$HOME")"
+
      - name: Run clippy with ${{ matrix.features.name }}
        run: |
          cargo clippy -p rustfs -p rustfs-protocols --all-targets ${{ matrix.features.flags }} -- -D warnings

      - name: Run tests with ${{ matrix.features.name }}
+        env:
+          # Keep feature-test linking under the same bounded concurrency as the
+          # main nextest lane; Clippy is metadata-only and needs no such limit.
+          CARGO_BUILD_JOBS: "2"
        run: |
-          cargo nextest run -p rustfs -p rustfs-protocols ${{ matrix.features.flags }}
+          # --profile ci so the quarantine list (and its junit flaky markers)
+          # covers this leg too; the default profile is the local no-retry
+          # profile and silently ignored quarantined flakes here (rustfs#6703).
+          cargo nextest run --profile ci -p rustfs -p rustfs-protocols ${{ matrix.features.flags }}

  build-rustfs-debug-binary:
    name: Build RustFS Debug Binary
@@ -311,7 +311,7 @@ jobs:
      - name: Wait for RustFS ready
        run: |
          for _ in {1..120}; do
-            if curl -sf "http://${S3_HOST}:${S3_PORT}/health" >/dev/null 2>&1; then
+            if curl -sf "http://${S3_HOST}:${S3_PORT}/health/ready" >/dev/null 2>&1; then
              echo "RustFS is ready"
              exit 0
            fi
@@ -0,0 +1,105 @@
+# Copyright 2024 RustFS Team
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+name: Upgrade Compatibility
+
+on:
+  pull_request:
+    paths:
+      - ".github/workflows/e2e-upgrade.yml"
+      - "crates/e2e_test/src/common.rs"
+      - "crates/e2e_test/src/lib.rs"
+      - "crates/e2e_test/src/upgrade_compatibility_test.rs"
+      - "crates/ecstore/**"
+      - "crates/filemeta/**"
+      - "crates/kms/**"
+      - "crates/storage-api/**"
+      - "rustfs/**"
+      - "Cargo.lock"
+  push:
+    tags:
+      - "[0-9]*.[0-9]*.[0-9]*"
+  schedule:
+    - cron: "17 3 * * 1"
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: ${{ github.event_name == 'pull_request' }}
+
+env:
+  CARGO_TERM_COLOR: always
+  RUST_BACKTRACE: 1
+  UPGRADE_SOURCE_VERSION: 1.0.0-rc.2
+  UPGRADE_SOURCE_ASSET: rustfs-linux-x86_64-gnu-v1.0.0-rc.2.zip
+  UPGRADE_SOURCE_SHA256: 7c789386bf85278f865b8e0d359bf4edb84d5aa408cc3fa54a18c25ca74cd6e7
+
+jobs:
+  direct-upgrade:
+    name: Direct upgrade from rc.2
+    runs-on: ubuntu-latest
+    timeout-minutes: 60
+    env:
+      FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          persist-credentials: false
+
+      - name: Setup Rust environment
+        uses: ./.github/actions/setup
+        with:
+          cache-shared-key: e2e-direct-upgrade
+          cache-save-if: ${{ github.ref == 'refs/heads/main' }}
+          install-build-packaging-tools: "false"
+
+      - name: Download pinned previous release
+        env:
+          SOURCE_DIR: ${{ runner.temp }}/rustfs-upgrade-source
+        run: |
+          set -euo pipefail
+          mkdir -p "$SOURCE_DIR"
+          archive="$SOURCE_DIR/$UPGRADE_SOURCE_ASSET"
+          curl --fail --location --retry 3 --output "$archive" \
+            "https://github.com/${GITHUB_REPOSITORY}/releases/download/${UPGRADE_SOURCE_VERSION}/${UPGRADE_SOURCE_ASSET}"
+          echo "$UPGRADE_SOURCE_SHA256  $archive" | sha256sum --check --strict
+          unzip -q "$archive" -d "$SOURCE_DIR"
+          chmod +x "$SOURCE_DIR/rustfs"
+          test -x "$SOURCE_DIR/rustfs"
+          echo "RUSTFS_UPGRADE_SOURCE_BINARY=$SOURCE_DIR/rustfs" >> "$GITHUB_ENV"
+          echo "RUSTFS_E2E_LOG_DIR=$RUNNER_TEMP/rustfs-upgrade-logs" >> "$GITHUB_ENV"
+
+      - name: Build current RustFS binary
+        run: |
+          cargo build --locked -p rustfs --bin rustfs
+          : > target/debug/rustfs.features
+
+      - name: Run direct-upgrade compatibility test
+        run: |
+          cargo test --locked -p e2e_test \
+            upgrade_compatibility_test::direct_upgrade_from_rc2_preserves_object_contracts \
+            -- --ignored --exact --nocapture
+
+      - name: Upload server logs
+        if: always()
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
+        with:
+          name: direct-upgrade-server-logs-${{ github.run_number }}
+          path: ${{ runner.temp }}/rustfs-upgrade-logs
+          if-no-files-found: warn
+          retention-days: 14
@@ -173,7 +173,7 @@ jobs:
          path: |
            fuzz/artifacts/**
            fuzz/corpus/${{ matrix.target }}/**
-          if-no-files-found: ignore
+          if-no-files-found: error
          retention-days: 7

  # ──────────────────────────────────────────────────────────────
@@ -227,7 +227,7 @@ jobs:
          path: |
            fuzz/artifacts/**
            fuzz/corpus/${{ matrix.target }}/**
-          if-no-files-found: ignore
+          if-no-files-found: error
          retention-days: 30

  # ──────────────────────────────────────────────────────────────
@@ -20,22 +20,27 @@
 # each run with Docker and then runs the `#[ignore]` reader tests in
 # rustfs/src/storage/minio_generated_read_test.rs.
 #
-# Scope: end-to-end MinIO-to-RustFS SSE interop is NOT implemented yet. Both
-# envelope parsers reject MinIO's own wrapped-DEK shape — see
-# `is_data_key_envelope` in crates/kms/src/encryption/dek.rs and the
-# `deny_unknown_fields` `LocalSseDekEnvelope` in rustfs/src/storage/sse.rs — and
-# closing that gap is tracked in rustfs/backlog#1638. Treat this job as the
-# harness for #1638, not as standing evidence that a MinIO migration reads back.
+# Scope: MinIO-to-RustFS SSE read interop is implemented behind the `rio-v2`
+# feature for MinIO's builtin static-KMS deployments — SSE-S3 and SSE-KMS
+# (single- and multipart) since rustfs/rustfs#6191, SSE-C detection since the
+# rustfs/backlog#1638 D2 close-out. This job is the standing evidence: it
+# regenerates real MinIO backend trees and proves byte-identical plaintext
+# reconstruction. KES/MinKMS-backed MinIO objects remain unreadable by design
+# (their envelopes are sealed by the KES service, not by a key RustFS can
+# hold), and default RustFS builds do not include the read path — it is a
+# special-purpose migration capability, not a default-build feature.
 #
 # Runner: GitHub-hosted `ubuntu-latest`. It reliably ships Docker + Python,
 # unlike the self-hosted fleet, whose pods drift in Docker/pip availability
 # (see the infra note in e2e-s3tests.yml). Nightly + manual only.
-# DISABLED. This workflow is switched off in the repository's Actions settings
-# (state: disabled_manually) and does not run on any trigger, including its cron
-# and workflow_dispatch. That state lives in GitHub's UI and is invisible when
-# reading this file, which has already misled at least one audit — hence this
-# banner. Re-enabling is a UI action; anyone doing so should first check that the
-# workflow still matches the current CI layout. See rustfs/backlog#1603.
+#
+# Enablement: this workflow was long disabled in the repository's Actions
+# settings (state: disabled_manually — a state that lives in GitHub's UI and is
+# invisible in this file). The change that updated this banner also re-added
+# the .github/scheduled-validations.json entry; both only make sense together
+# with re-enabling the workflow in the Actions settings. If it is ever disabled
+# again, remove the scheduled-validations entry in the same change — a disabled
+# workflow can never satisfy the freshness check. See rustfs/backlog#1603.
 #
 name: minio-interop

@@ -155,7 +155,7 @@ jobs:
      - name: Wait for RustFS ready
        run: |
          for _ in {1..60}; do
-            if curl -sf http://127.0.0.1:9000/health >/dev/null 2>&1; then
+            if curl -sf http://127.0.0.1:9000/health/ready >/dev/null 2>&1; then
              echo "RustFS is ready"
              exit 0
            fi
@@ -34,7 +34,7 @@ env:
 jobs:
  build:
    name: Build x86_64 GNU
-    runs-on: sm-standard-2
+    runs-on: sm-standard-4
    timeout-minutes: 150
    env:
      FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
@@ -55,6 +55,155 @@ jobs:
      - name: Build RustFS
        run: cargo build --release --locked --target x86_64-unknown-linux-gnu -p rustfs --bins

+      - name: Build DEB package
+        id: deb
+        shell: bash
+        run: |
+          set -euo pipefail
+
+          # Nightly snapshot name: rustfs-nightly-<YYYY-MM-DD> (Asia/Shanghai,
+          # matching the schedule timezone so the file date always matches the
+          # cron's intended day).
+          DEB_DATE="$(TZ=Asia/Shanghai date +%Y-%m-%d)"
+          DEB_FILE="rustfs-nightly-${DEB_DATE}.deb"
+          PKG_DIR="rustfs-nightly-${DEB_DATE}"
+
+          command -v fakeroot >/dev/null 2>&1 || sudo apt-get install -y -qq fakeroot
+
+          BIN="target/x86_64-unknown-linux-gnu/release/rustfs"
+          test -x "${BIN}" || { echo "rustfs binary not found: ${BIN}"; exit 1; }
+
+          mkdir -p "${PKG_DIR}/DEBIAN"
+          mkdir -p "${PKG_DIR}/usr/bin"
+          mkdir -p "${PKG_DIR}/etc/default"
+          mkdir -p "${PKG_DIR}/lib/systemd/system"
+          mkdir -p "${PKG_DIR}/usr/share/doc/rustfs"
+
+          cp "${BIN}" "${PKG_DIR}/usr/bin/rustfs"
+          chmod 755 "${PKG_DIR}/usr/bin/rustfs"
+          cp deploy/build/rustfs.service "${PKG_DIR}/lib/systemd/system/"
+
+          cat > "${PKG_DIR}/etc/default/rustfs" << 'ENVEOF'
+          # RustFS Environment Configuration
+          # See https://rustfs.com/docs/ for more information
+          # RUSTFS_VOLUMES=""
+          # RUSTFS_ROOT_USER=""
+          # RUSTFS_ROOT_PASSWORD=""
+          ENVEOF
+
+          # dpkg versions must start with a digit and cannot contain hyphens;
+          # a date-based snapshot version keeps the nightly installable
+          # alongside release packages.
+          DEB_VERSION="${DEB_DATE//-/.}~nightly"
+
+          cat > "${PKG_DIR}/DEBIAN/control" << EOF
+          Package: rustfs
+          Version: ${DEB_VERSION}
+          Section: utils
+          Priority: optional
+          Architecture: amd64
+          Depends: libc6 (>= 2.31)
+          Maintainer: RustFS Team <[email protected]>
+          Description: High-performance distributed object storage
+           RustFS is a high-performance distributed object storage software
+           built using Rust. It is compatible with MinIO and S3 API.
+          Homepage: https://rustfs.com
+          EOF
+
+          cat > "${PKG_DIR}/DEBIAN/conffiles" << 'CONFFILES'
+          /etc/default/rustfs
+          CONFFILES
+
+          cat > "${PKG_DIR}/DEBIAN/postinst" << 'POSTINST'
+          #!/bin/bash
+          set -e
+          if ! getent passwd rustfs > /dev/null 2>&1; then
+              useradd -r -s /bin/false -d /opt/rustfs rustfs
+          fi
+          mkdir -p /opt/rustfs /data/rustfs /var/log/rustfs
+          chown rustfs:rustfs /opt/rustfs /data/rustfs /var/log/rustfs
+          if [ -d /run/systemd/system ]; then
+              systemctl daemon-reload
+          fi
+          echo "RustFS installed. Configure /etc/default/rustfs then: systemctl start rustfs"
+          POSTINST
+          chmod 755 "${PKG_DIR}/DEBIAN/postinst"
+
+          cat > "${PKG_DIR}/DEBIAN/prerm" << 'PRERM'
+          #!/bin/bash
+          set -e
+          if [ -d /run/systemd/system ] && systemctl is-active --quiet rustfs; then
+              systemctl stop rustfs
+          fi
+          PRERM
+          chmod 755 "${PKG_DIR}/DEBIAN/prerm"
+
+          cat > "${PKG_DIR}/DEBIAN/postrm" << 'POSTRM'
+          #!/bin/bash
+          set -e
+          if [ -d /run/systemd/system ]; then
+              systemctl daemon-reload
+          fi
+          POSTRM
+          chmod 755 "${PKG_DIR}/DEBIAN/postrm"
+
+          cp LICENSE "${PKG_DIR}/usr/share/doc/rustfs/"
+          cp README.md "${PKG_DIR}/usr/share/doc/rustfs/"
+
+          fakeroot dpkg-deb --build "${PKG_DIR}"
+          ls -lh "${DEB_FILE}"
+          echo "deb_file=${DEB_FILE}" >> "${GITHUB_OUTPUT}"
+
+      - name: Upload DEB artifact
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
+        with:
+          name: ${{ steps.deb.outputs.deb_file }}
+          path: ${{ steps.deb.outputs.deb_file }}
+          if-no-files-found: error
+
+      # Persist the nightly deb on Cloudflare R2 (same channel as package.yml)
+      # so it can be downloaded later with a stable, unauthenticated URL —
+      # e.g. https://dl.rustfs.com/artifacts/rustfs/packages/nightly/... .
+      # Skipped when the R2 secrets are not configured (artifact-only mode).
+      - name: Upload DEB to Cloudflare R2
+        if: env.R2_ACCESS_KEY_ID != ''
+        env:
+          R2_ACCESS_KEY_ID: ${{ secrets.R2_ACCESS_KEY_ID }}
+          R2_SECRET_ACCESS_KEY: ${{ secrets.R2_SECRET_ACCESS_KEY }}
+          R2_ENDPOINT: ${{ secrets.R2_ENDPOINT }}
+          R2_BUCKET: ${{ secrets.R2_BUCKET }}
+          AWS_EC2_METADATA_DISABLED: true
+        shell: bash
+        run: |
+          set -euo pipefail
+
+          if [[ -z "$R2_ACCESS_KEY_ID" || -z "$R2_SECRET_ACCESS_KEY" || -z "$R2_ENDPOINT" || -z "$R2_BUCKET" ]]; then
+            echo "⚠️ R2 credentials missing, skipping upload"
+            exit 0
+          fi
+
+          if ! command -v aws >/dev/null 2>&1; then
+            sudo apt-get update && sudo apt-get install -y -qq awscli
+          fi
+
+          export AWS_ACCESS_KEY_ID="$R2_ACCESS_KEY_ID"
+          export AWS_SECRET_ACCESS_KEY="$R2_SECRET_ACCESS_KEY"
+          export AWS_DEFAULT_REGION="auto"
+
+          DEB_FILE="${{ steps.deb.outputs.deb_file }}"
+          R2_PREFIX="s3://${R2_BUCKET}/artifacts/rustfs/packages/nightly/"
+
+          echo "📤 Uploading ${DEB_FILE} to ${R2_PREFIX}"
+          aws s3 cp "${DEB_FILE}" "${R2_PREFIX}" --endpoint-url "$R2_ENDPOINT" --only-show-errors
+
+          # Stable "latest" alias so tests can fetch the newest nightly
+          # without knowing today's date.
+          echo "📤 Uploading latest alias"
+          aws s3 cp "${DEB_FILE}" "${R2_PREFIX}rustfs-nightly-latest.deb" \
+            --endpoint-url "$R2_ENDPOINT" --only-show-errors
+
+          echo "✅ R2 upload complete"
+
  # Live-Vault lane for the rustfs-kms suite (rustfs/backlog#1774).
  #
  # RUSTFS_KMS_VAULT_TOKEN is the single switch that adds the Vault KV2 and
@@ -0,0 +1,110 @@
+# Copyright 2024 RustFS Team
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+name: OIDC Keycloak Live
+
+on:
+  pull_request:
+    paths:
+      - ".github/workflows/oidc-keycloak.yml"
+      - "crates/config/src/constants/oidc.rs"
+      - "crates/iam/src/federation/**"
+      - "crates/iam/src/oidc.rs"
+      - "rustfs/src/admin/handlers/oidc.rs"
+      - "rustfs/src/admin/handlers/sts.rs"
+      - "scripts/test/oidc_keycloak_live.sh"
+      - "scripts/test/fixtures/keycloak-rustfs-ci-realm.json"
+  push:
+    branches: [main]
+    paths:
+      - ".github/workflows/oidc-keycloak.yml"
+      - "crates/config/src/constants/oidc.rs"
+      - "crates/iam/src/federation/**"
+      - "crates/iam/src/oidc.rs"
+      - "rustfs/src/admin/handlers/oidc.rs"
+      - "rustfs/src/admin/handlers/sts.rs"
+      - "scripts/test/oidc_keycloak_live.sh"
+      - "scripts/test/fixtures/keycloak-rustfs-ci-realm.json"
+  schedule:
+    - cron: "23 2 * * 1"
+      timezone: "Asia/Shanghai"
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: oidc-keycloak-live-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  oidc-keycloak-live:
+    name: OIDC Keycloak live gate
+    runs-on: ubuntu-latest
+    timeout-minutes: 60
+    env:
+      FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          persist-credentials: false
+
+      - name: Setup Rust environment
+        uses: ./.github/actions/setup
+        with:
+          cache-shared-key: oidc-keycloak-live
+          cache-save-if: "true"
+          install-build-packaging-tools: "false"
+          install-test-tools: "false"
+
+      - name: Build RustFS
+        run: cargo build --locked -p rustfs --bin rustfs
+
+      - name: Install pinned request signer
+        run: |
+          python3 -m pip install --user --upgrade pip "awscurl==0.44"
+          echo "${HOME}/.local/bin" >> "${GITHUB_PATH}"
+
+      - name: Run live Keycloak discovery, JWT and STS checks
+        run: bash scripts/test/oidc_keycloak_live.sh ./target/debug/rustfs
+
+      - name: Upload service logs
+        if: failure()
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
+        with:
+          name: oidc-keycloak-live-${{ github.run_number }}
+          path: ${{ runner.temp }}/rustfs-keycloak-live-*/**/*.log
+          if-no-files-found: ignore
+          retention-days: 3
+
+  alert-on-failure:
+    name: Alert on scheduled failure
+    needs: oidc-keycloak-live
+    if: >-
+      always() && github.event_name == 'schedule' &&
+      (needs.oidc-keycloak-live.result == 'failure' || needs.oidc-keycloak-live.result == 'cancelled')
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    permissions:
+      contents: read
+      issues: write
+    steps:
+      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          persist-credentials: false
+      - name: Open or update failure-tracking issue
+        uses: ./.github/actions/schedule-failure-issue
+        with:
+          github-token: ${{ secrets.GITHUB_TOKEN }}
@@ -0,0 +1,134 @@
+name: RustFS Heal Test
+
+on:
+  workflow_dispatch:
+    inputs:
+      package_url:
+        description: 'Direct .deb URL (nightly/R2). Defaults to the latest nightly deb.'
+        required: false
+        type: string
+      stop_node_gb:
+        description: 'Stop the outage node when surviving nodes reach N GiB'
+        required: false
+        default: '15'
+      warp_stop_gb:
+        description: 'Stop warp when surviving nodes reach N GiB'
+        required: false
+        default: '40'
+      cleanup_before:
+        description: 'Reset the nodes before the test (DESTROYS existing data/config)'
+        type: boolean
+        default: true
+      cleanup_after:
+        description: 'Reset the nodes after the test (DESTROYS test data/config)'
+        type: boolean
+        default: true
+  workflow_run:
+    # Run after the nightly build completes: the nightly deb is what the test
+    # installs. Heal also runs inside the functional suite; this standalone
+    # workflow enables manual single-suite runs as well.
+    workflows: ["Nightly GNU Build"]
+    types: [completed]
+
+permissions:
+  contents: read
+
+# Only one test at a time: both this and the pool-expansion workflow mutate
+# the same test environment, so they share one concurrency group.
+concurrency:
+  group: rustfs-pool-expansion-test
+  cancel-in-progress: false
+
+defaults:
+  run:
+    shell: bash
+
+env:
+  RUSTFS_ACCESS_KEY: ${{ secrets.RUSTFS_ACCESS_KEY }}
+  RUSTFS_SECRET_KEY: ${{ secrets.RUSTFS_SECRET_KEY }}
+  RUSTFS_API_ENDPOINT: ${{ secrets.RUSTFS_API_ENDPOINT || vars.RUSTFS_API_ENDPOINT || vars.RUSTFS_RC_ENDPOINT }}
+  RUSTFS_NODES: ${{ secrets.RUSTFS_NODES || vars.RUSTFS_NODES }}
+  RUSTFS_SSH_USER: ${{ secrets.RUSTFS_SSH_USER || vars.RUSTFS_SSH_USER }}
+  RUSTFS_NIGHTLY_PACKAGE_URL: ${{ vars.RUSTFS_NIGHTLY_PACKAGE_URL || 'https://dl.rustfs.com/artifacts/rustfs/packages/nightly/rustfs-nightly-latest.deb' }}
+
+jobs:
+  heal-test:
+    runs-on: smoke-testing
+    timeout-minutes: 480
+    # Run on manual dispatch, or when the nightly build completed successfully.
+    # Skipped when nightly failed.
+    if: ${{ github.event_name == 'workflow_dispatch' || github.event.workflow_run.conclusion == 'success' }}
+    steps:
+      - name: Checkout auto-testing scripts
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          repository: rustfs/auto-testing
+          ref: main
+          path: auto-testing
+          persist-credentials: false
+          token: ${{ secrets.PF_TESTING_GH_TOKEN }}
+
+      - name: Show environment
+        run: |
+          uname -a
+          jq --version
+          openssl version
+          warp --version || true
+          df -h /data | tail -1
+
+      - name: Reset test environment (before)
+        if: ${{ inputs.cleanup_before != 'false' }}
+        run: |
+          chmod +x auto-testing/rustfs_heal_test.sh
+          ./auto-testing/rustfs_heal_test.sh --reset -y
+
+      - name: Install RustFS package & start cluster
+        run: |
+          ARGS=(--steps "1,2" -y --endpoint "${{ env.RUSTFS_API_ENDPOINT }}")
+          if [ -n "${{ inputs.package_url }}" ]; then
+            ARGS+=(--package-url "${{ inputs.package_url }}")
+          else
+            ARGS+=(--package-url "${{ env.RUSTFS_NIGHTLY_PACKAGE_URL }}")
+          fi
+          ./auto-testing/rustfs_heal_test.sh "${ARGS[@]}"
+
+      - name: Preflight checks
+        run: |
+          ARGS=(--preflight --endpoint "${{ env.RUSTFS_API_ENDPOINT }}")
+          if [ -n "${{ inputs.package_url }}" ]; then
+            ARGS+=(--package-url "${{ inputs.package_url }}")
+          else
+            ARGS+=(--package-url "${{ env.RUSTFS_NIGHTLY_PACKAGE_URL }}")
+          fi
+          ./auto-testing/rustfs_heal_test.sh "${ARGS[@]}"
+
+      - name: Run heal test (write -> outage -> heal -> verify)
+        run: |
+          ./auto-testing/rustfs_heal_test.sh \
+            --steps "3,4,5,6,7" -y \
+            --endpoint "${{ env.RUSTFS_API_ENDPOINT }}" \
+            --stop-node-gb "${{ inputs.stop_node_gb }}" \
+            --warp-stop-gb "${{ inputs.warp_stop_gb }}" \
+            --log-file /tmp/rustfs-heal-test.log
+
+      - name: Upload test logs
+        if: always()
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
+        with:
+          name: rustfs-heal-test-${{ github.run_id }}
+          path: |
+            /tmp/rustfs-heal-test*.log
+            /tmp/rustfs-warp.*.log
+          if-no-files-found: warn
+
+      - name: Reset test environment (after)
+        if: ${{ always() && inputs.cleanup_after != 'false' }}
+        run: |
+          ./auto-testing/rustfs_heal_test.sh --reset -y
+
+      - name: Notify on failure
+        if: failure()
+        run: |
+          echo "RustFS heal test failed"
+          echo "Package source: ${{ inputs.package_url || 'nightly (R2 latest)' }}"
+          echo "See the uploaded log artifact for details."
@@ -0,0 +1,206 @@
+name: RustFS Performance Test
+
+on:
+  workflow_dispatch:
+    inputs:
+      package_url:
+        description: 'Direct .deb URL (nightly/R2). Defaults to the latest nightly deb.'
+        required: false
+        type: string
+      test_method:
+        description: 'Benchmark method(s) to run (manual runs only; "all" = GET+PUT+MIXED)'
+        type: choice
+        options:
+          - all
+          - get
+          - put
+          - mixed
+        default: 'all'
+      object_size:
+        description: 'Object size(s) to test (manual runs only; "all" = all 10 sizes)'
+        type: choice
+        options:
+          - all
+          - 1KiB
+          - 4KiB
+          - 16KiB
+          - 128KiB
+          - 1MiB
+          - 4MiB
+          - 8MiB
+          - 16MiB
+          - 32MiB
+          - 64MiB
+        default: 'all'
+      warp_duration:
+        description: 'warp duration per round (e.g. 5m, 30s)'
+        required: false
+        default: '5m'
+      warp_concurrency:
+        description: 'warp concurrency'
+        required: false
+        default: '64'
+      cleanup_before:
+        description: 'Reset the nodes before the test (DESTROYS existing data/config)'
+        type: boolean
+        default: true
+      cleanup_after:
+        description: 'Reset the nodes after the test (DESTROYS test data/config)'
+        type: boolean
+        default: true
+  workflow_run:
+    # Run after the nightly build completes; the nightly deb is what the test installs.
+    workflows: ["Nightly GNU Build"]
+    types: [completed]
+
+permissions:
+  contents: read
+
+# Dedicated pf-testing runner/environment: own concurrency group so perf runs
+# never block (or are blocked by) the pool-expansion / heal tests.
+concurrency:
+  group: rustfs-performance-test
+  cancel-in-progress: false
+
+defaults:
+  run:
+    shell: bash
+
+env:
+  RUSTFS_ACCESS_KEY: ${{ secrets.RUSTFS_ACCESS_KEY }}
+  RUSTFS_SECRET_KEY: ${{ secrets.RUSTFS_SECRET_KEY }}
+  # Performance test uses its own node list (4 nodes); the shared
+  # RUSTFS_NODES secret is used by the 3-node pool-expansion / heal tests.
+  RUSTFS_NODES: ${{ secrets.RUSTFS_PERF_NODES || vars.RUSTFS_PERF_NODES || 'vm000 vm001 vm002 vm003' }}
+  RUSTFS_SSH_USER: ${{ secrets.RUSTFS_SSH_USER || vars.RUSTFS_SSH_USER }}
+  # Package used by the nightly run (workflow_dispatch inputs are empty for
+  # workflow_run events), i.e. the latest nightly deb published by nightly-gnu.yml.
+  RUSTFS_NIGHTLY_PACKAGE_URL: ${{ vars.RUSTFS_NIGHTLY_PACKAGE_URL || 'https://dl.rustfs.com/artifacts/rustfs/packages/nightly/rustfs-nightly-latest.deb' }}
+  # Fixed benchmark result directory so later steps can read summary.md
+  RUSTFS_RESULT_DIR: /tmp/rustfs-perf-results
+  # Cross-repo token for writing to rustfs/backlog (set in repo settings)
+  PF_TESTING_GH_TOKEN: ${{ secrets.PF_TESTING_GH_TOKEN }}
+
+jobs:
+  performance-test:
+    runs-on: pf-testing
+    timeout-minutes: 900
+    # Run on manual dispatch, or when the nightly build completed successfully.
+    # Skipped when nightly failed.
+    if: ${{ github.event_name == 'workflow_dispatch' || github.event.workflow_run.conclusion == 'success' }}
+    steps:
+      - name: Checkout auto-testing scripts
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          repository: rustfs/auto-testing
+          ref: main
+          path: auto-testing
+          persist-credentials: false
+          token: ${{ secrets.PF_TESTING_GH_TOKEN }}
+
+      - name: Show environment
+        run: |
+          uname -a
+          jq --version
+          warp --version || true
+          df -h /data | tail -1
+
+      - name: Reset test environment (before)
+        if: ${{ inputs.cleanup_before != 'false' }}
+        run: |
+          chmod +x auto-testing/rustfs_performance_test.sh
+          ./auto-testing/rustfs_performance_test.sh --step 1 -y
+
+      - name: Install RustFS package & start cluster (4x4)
+        run: |
+          ARGS=(--steps "2,3,4" -y)
+          if [ -n "${{ inputs.package_url }}" ]; then
+            ARGS+=(--package-url "${{ inputs.package_url }}")
+          else
+            ARGS+=(--package-url "${{ env.RUSTFS_NIGHTLY_PACKAGE_URL }}")
+          fi
+          ./auto-testing/rustfs_performance_test.sh "${ARGS[@]}"
+
+      - name: Preflight checks
+        run: |
+          ARGS=(--preflight)
+          if [ -n "${{ inputs.package_url }}" ]; then
+            ARGS+=(--package-url "${{ inputs.package_url }}")
+          else
+            ARGS+=(--package-url "${{ env.RUSTFS_NIGHTLY_PACKAGE_URL }}")
+          fi
+          ./auto-testing/rustfs_performance_test.sh "${ARGS[@]}"
+
+      - name: Run benchmark (GET/PUT/MIXED)
+        id: benchmark
+        run: |
+          # Empty on automatic (workflow_run) runs -> full 30 rounds.
+          # Manual dispatch can restrict method(s)/size(s).
+          export WARP_METHODS="${{ inputs.test_method }}"
+          export WARP_SIZES="${{ inputs.object_size }}"
+          ./auto-testing/rustfs_performance_test.sh \
+            --step 5 -y \
+            --warp-duration "${{ inputs.warp_duration || '5m' }}" \
+            --warp-concurrency "${{ inputs.warp_concurrency || '64' }}" \
+            --log-file /tmp/rustfs-perf-test.log
+
+      - name: Analyze results
+        if: ${{ steps.benchmark.conclusion == 'success' }}
+        run: |
+          ./auto-testing/rustfs_performance_test.sh --step 6 -y
+
+      - name: Post results to backlog issue
+        if: ${{ steps.benchmark.conclusion == 'success' }}
+        env:
+          GH_TOKEN: ${{ env.PF_TESTING_GH_TOKEN }}
+          RESULT_DIR: ${{ env.RUSTFS_RESULT_DIR }}
+        run: |
+          set -euo pipefail
+          if [ -z "${GH_TOKEN:-}" ]; then
+            echo "PF_TESTING_GH_TOKEN is not configured; skipping issue post"
+            exit 0
+          fi
+          SUMMARY="${RESULT_DIR}/summary.md"
+          [ -f "${SUMMARY}" ] || { echo "summary.md not found at ${SUMMARY}"; exit 1; }
+          DATE="$(date -u +%Y-%m-%d)"
+          {
+            echo "## RustFS nightly build performance testing report"
+            echo ""
+            echo "- **日期**: ${DATE}"
+            echo "- **Run**: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
+            echo "- **触发方式**: ${{ github.event_name }}"
+            echo ""
+            cat "${SUMMARY}"
+          } > /tmp/rustfs-perf-issue-body.md
+          TITLE="RustFS nightly build performance testing report"
+          EXISTING="$(gh issue list --repo rustfs/backlog \
+            --search "in:title \"${TITLE}\"" --state all --limit 5 \
+            --json number --jq '.[0].number // empty')"
+          if [ -n "${EXISTING}" ]; then
+            gh issue comment "${EXISTING}" --repo rustfs/backlog --body-file /tmp/rustfs-perf-issue-body.md
+            echo "commented on existing issue #${EXISTING}"
+          else
+            gh issue create --repo rustfs/backlog --title "${TITLE}" --body-file /tmp/rustfs-perf-issue-body.md
+          fi
+
+      - name: Upload test logs & results
+        if: always()
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
+        with:
+          name: rustfs-perf-test-${{ github.run_id }}
+          path: |
+            /tmp/rustfs-perf-test*.log
+            /tmp/rustfs-perf-results/**
+          if-no-files-found: warn
+
+      - name: Reset test environment (after)
+        if: ${{ always() && inputs.cleanup_after != 'false' }}
+        run: |
+          ./auto-testing/rustfs_performance_test.sh --step 7 -y
+
+      - name: Notify on failure
+        if: failure()
+        run: |
+          echo "RustFS performance test failed"
+          echo "Package source: ${{ inputs.package_url || 'nightly (R2 latest)' }}"
+          echo "See the uploaded log artifact for details."
@@ -0,0 +1,400 @@
+name: RustFS Functional Test Suite (S3/KMS/tier/pool/heal)
+
+on:
+  workflow_dispatch:
+    inputs:
+      rustfs_version:
+        description: 'RustFS release tag to test (e.g. 1.0.0-rc.3)'
+        required: false
+        default: '1.0.0-rc.3'
+      package_url:
+        description: 'Direct .deb URL (nightly/R2/dev). Overrides rustfs_version.'
+        required: false
+        type: string
+      pools:
+        description: 'Number of pools to expand to (2 = first rebalance only)'
+        type: choice
+        options:
+          - '2'
+          - '3'
+        default: '3'
+      storage_threshold:
+        description: 'Stop writing when storage usage reaches N%'
+        required: false
+        default: '50'
+      warp_duration:
+        description: 'warp write duration (e.g. 5m, 10m)'
+        required: false
+        default: '10m'
+      run_decommission:
+        description: 'Run the pool decommission step (3-pool topology only)'
+        type: boolean
+        default: true
+      stop_node_gb:
+        description: 'Heal: stop the outage node when surviving nodes reach N GiB'
+        required: false
+        default: '15'
+      warp_stop_gb:
+        description: 'Heal: stop warp when surviving nodes reach N GiB'
+        required: false
+        default: '40'
+      cleanup_before:
+        description: 'Reset the nodes before the test (DESTROYS existing data/config)'
+        type: boolean
+        default: true
+      cleanup_after:
+        description: 'Reset the nodes after the test (DESTROYS test data/config)'
+        type: boolean
+        default: true
+  workflow_run:
+    # Run after the nightly build completes: S3 -> KMS -> tier -> pool -> heal.
+    workflows: ["Nightly GNU Build"]
+    types: [completed]
+
+permissions:
+  contents: read
+
+# Only one test run at a time: every job mutates the same shared test
+# environment (vm000/vm001/vm002), so concurrent runs must not clobber each
+# other. Jobs inside a run are chained with needs to serialize them.
+concurrency:
+  group: rustfs-pool-expansion-test
+  cancel-in-progress: false
+
+defaults:
+  run:
+    shell: bash
+
+env:
+  RUSTFS_ACCESS_KEY: ${{ secrets.RUSTFS_ACCESS_KEY }}
+  RUSTFS_SECRET_KEY: ${{ secrets.RUSTFS_SECRET_KEY }}
+  RUSTFS_API_ENDPOINT: ${{ secrets.RUSTFS_API_ENDPOINT || vars.RUSTFS_API_ENDPOINT || vars.RUSTFS_RC_ENDPOINT }}
+  RUSTFS_NODES: ${{ secrets.RUSTFS_NODES || vars.RUSTFS_NODES }}
+  RUSTFS_SSH_USER: ${{ secrets.RUSTFS_SSH_USER || vars.RUSTFS_SSH_USER }}
+  # Package used by the nightly run (workflow_dispatch inputs are empty for
+  # workflow_run events), i.e. the latest nightly deb published by nightly-gnu.yml.
+  RUSTFS_NIGHTLY_PACKAGE_URL: ${{ vars.RUSTFS_NIGHTLY_PACKAGE_URL || 'https://dl.rustfs.com/artifacts/rustfs/packages/nightly/rustfs-nightly-latest.deb' }}
+
+jobs:
+  # All test scripts live in rustfs/auto-testing; rustfs stores only this
+  # workflow. Each job checks out auto-testing before running.
+
+  s3-compat-test:
+    name: S3 compatibility test
+    runs-on: smoke-testing
+    timeout-minutes: 240
+    if: ${{ github.event_name == 'workflow_dispatch' || github.event.workflow_run.conclusion == 'success' }}
+    steps:
+      - name: Checkout auto-testing scripts
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          repository: rustfs/auto-testing
+          ref: main
+          path: auto-testing
+          persist-credentials: false
+          token: ${{ secrets.PF_TESTING_GH_TOKEN }}
+
+      - name: Run S3 compatibility suite
+        run: |
+          chmod +x auto-testing/rustfs-s3-compat-test.sh
+          ARGS=(--all-topologies -y --log-file /tmp/rustfs-s3-compat.log)
+          if [ -n "${{ inputs.package_url }}" ]; then
+            ARGS+=(--package-url "${{ inputs.package_url }}")
+          elif [ -n "${{ inputs.rustfs_version }}" ]; then
+            ARGS+=(--version "${{ inputs.rustfs_version }}")
+          else
+            ARGS+=(--package-url "${{ env.RUSTFS_NIGHTLY_PACKAGE_URL }}")
+          fi
+          ./auto-testing/rustfs-s3-compat-test.sh "${ARGS[@]}"
+
+      - name: Upload test logs
+        if: always()
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
+        with:
+          name: rustfs-s3-compat-${{ github.run_id }}
+          path: |
+            /tmp/rustfs-s3-compat*.log
+          if-no-files-found: warn
+
+      - name: Notify on failure
+        if: failure()
+        run: |
+          echo "RustFS S3 compatibility suite failed"
+          echo "See the uploaded log artifact for details."
+
+  kms-test:
+    name: KMS test (after S3)
+    runs-on: smoke-testing
+    timeout-minutes: 360
+    needs: s3-compat-test
+    if: ${{ always() && (github.event_name == 'workflow_dispatch' || github.event.workflow_run.conclusion == 'success') }}
+    steps:
+      - name: Checkout auto-testing scripts
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          repository: rustfs/auto-testing
+          ref: main
+          path: auto-testing
+          persist-credentials: false
+          token: ${{ secrets.PF_TESTING_GH_TOKEN }}
+
+      - name: Ensure docker (Vault container)
+        run: |
+          if ! command -v docker >/dev/null 2>&1; then
+            sudo apt-get update
+            sudo apt-get install -y docker.io
+          fi
+          sudo systemctl enable --now docker
+          docker info >/dev/null 2>&1 || sudo docker info >/dev/null 2>&1
+
+      - name: Run KMS suite
+        run: |
+          chmod +x auto-testing/rustfs-kms-test.sh
+          ARGS=(--all-topologies --backends local,vault-kv2 -y --log-file /tmp/rustfs-kms.log)
+          if [ -n "${{ inputs.package_url }}" ]; then
+            ARGS+=(--package-url "${{ inputs.package_url }}")
+          elif [ -n "${{ inputs.rustfs_version }}" ]; then
+            ARGS+=(--version "${{ inputs.rustfs_version }}")
+          else
+            ARGS+=(--package-url "${{ env.RUSTFS_NIGHTLY_PACKAGE_URL }}")
+          fi
+          ./auto-testing/rustfs-kms-test.sh "${ARGS[@]}"
+
+      - name: Upload test logs
+        if: always()
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
+        with:
+          name: rustfs-kms-test-${{ github.run_id }}
+          path: |
+            /tmp/rustfs-kms*.log
+          if-no-files-found: warn
+
+      - name: Notify on failure
+        if: failure()
+        run: |
+          echo "RustFS KMS suite failed"
+          echo "See the uploaded log artifact for details."
+
+  tier-test:
+    name: Tier / event / audit test (after KMS)
+    runs-on: smoke-testing
+    timeout-minutes: 360
+    needs: kms-test
+    if: ${{ always() && (github.event_name == 'workflow_dispatch' || github.event.workflow_run.conclusion == 'success') }}
+    steps:
+      - name: Checkout auto-testing scripts
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          repository: rustfs/auto-testing
+          ref: main
+          path: auto-testing
+          persist-credentials: false
+          token: ${{ secrets.PF_TESTING_GH_TOKEN }}
+
+      - name: Ensure MQTT broker + clients (event notification tests)
+        run: |
+          if ! command -v mosquitto_sub >/dev/null 2>&1; then
+            sudo apt-get update
+            sudo apt-get install -y mosquitto mosquitto-clients
+          fi
+          sudo mkdir -p /etc/mosquitto/conf.d
+          printf 'listener 1883 0.0.0.0\nallow_anonymous true\n' | sudo tee /etc/mosquitto/conf.d/rustfs-test.conf >/dev/null
+          sudo systemctl restart mosquitto
+          sleep 2
+          ss -tln 2>/dev/null | grep -q ':1883' || { echo "mosquitto not listening on 1883"; exit 1; }
+
+      - name: Run tier / event / audit suite
+        run: |
+          chmod +x auto-testing/rustfs-tier-test.sh
+          ARGS=(--all-topologies -y --log-file /tmp/rustfs-tier.log)
+          if [ -n "${{ inputs.package_url }}" ]; then
+            ARGS+=(--package-url "${{ inputs.package_url }}")
+          elif [ -n "${{ inputs.rustfs_version }}" ]; then
+            ARGS+=(--version "${{ inputs.rustfs_version }}")
+          else
+            ARGS+=(--package-url "${{ env.RUSTFS_NIGHTLY_PACKAGE_URL }}")
+          fi
+          ./auto-testing/rustfs-tier-test.sh "${ARGS[@]}"
+
+      - name: Upload test logs
+        if: always()
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
+        with:
+          name: rustfs-tier-test-${{ github.run_id }}
+          path: |
+            /tmp/rustfs-tier*.log
+          if-no-files-found: warn
+
+      - name: Notify on failure
+        if: failure()
+        run: |
+          echo "RustFS tier/event/audit suite failed"
+          echo "See the uploaded log artifact for details."
+
+  pool-expansion-test:
+    name: Pool expansion / decommission test (after tier)
+    runs-on: smoke-testing
+    timeout-minutes: 360
+    needs: tier-test
+    if: ${{ always() && (github.event_name == 'workflow_dispatch' || github.event.workflow_run.conclusion == 'success') }}
+    steps:
+      - name: Checkout auto-testing scripts
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          repository: rustfs/auto-testing
+          ref: main
+          path: auto-testing
+          persist-credentials: false
+          token: ${{ secrets.PF_TESTING_GH_TOKEN }}
+
+      - name: Show environment
+        run: |
+          uname -a
+          jq --version
+          openssl version
+          warp --version || true
+          df -h /data | tail -1
+
+      - name: Reset test environment (before)
+        if: ${{ inputs.cleanup_before != 'false' }}
+        run: |
+          chmod +x auto-testing/rustfs_pool_expand.sh
+          ./auto-testing/rustfs_pool_expand.sh --reset -y
+
+      - name: Install RustFS package & start first pool
+        run: |
+          ARGS=(--steps "1,2,3" -y --endpoint "${{ env.RUSTFS_API_ENDPOINT }}")
+          if [ -n "${{ inputs.package_url }}" ]; then
+            ARGS+=(--package-url "${{ inputs.package_url }}")
+          elif [ -n "${{ inputs.rustfs_version }}" ]; then
+            ARGS+=(--version "${{ inputs.rustfs_version }}")
+          else
+            ARGS+=(--package-url "${{ env.RUSTFS_NIGHTLY_PACKAGE_URL }}")
+          fi
+          ./auto-testing/rustfs_pool_expand.sh "${ARGS[@]}"
+
+      - name: Preflight checks
+        run: |
+          ARGS=(--preflight --endpoint "${{ env.RUSTFS_API_ENDPOINT }}")
+          if [ -n "${{ inputs.package_url }}" ]; then
+            ARGS+=(--package-url "${{ inputs.package_url }}")
+          elif [ -n "${{ inputs.rustfs_version }}" ]; then
+            ARGS+=(--version "${{ inputs.rustfs_version }}")
+          else
+            ARGS+=(--package-url "${{ env.RUSTFS_NIGHTLY_PACKAGE_URL }}")
+          fi
+          ./auto-testing/rustfs_pool_expand.sh "${ARGS[@]}"
+
+      - name: Run pool expansion & decommission test
+        id: pool_test
+        run: |
+          set -o pipefail
+          STEPS="4,5,6"
+          if [ "${{ inputs.pools || '3' }}" = "3" ]; then
+            STEPS="$STEPS,7,8"
+            if [ "${{ inputs.run_decommission != 'false' }}" = "true" ]; then
+              STEPS="$STEPS,9"
+            fi
+          fi
+          ./auto-testing/rustfs_pool_expand.sh \
+            --steps "$STEPS" --with-warp -y \
+            --endpoint "${{ env.RUSTFS_API_ENDPOINT }}" \
+            --storage-threshold "${{ inputs.storage_threshold || '50' }}" \
+            --warp-duration "${{ inputs.warp_duration || '10m' }}" \
+            --log-file /tmp/rustfs-pool-test.log
+
+      - name: Upload test logs
+        if: always()
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
+        with:
+          name: rustfs-pool-test-${{ github.run_id }}
+          path: |
+            /tmp/rustfs-pool-test*.log
+            /tmp/rustfs-warp.*.log
+          if-no-files-found: warn
+
+      - name: Reset test environment (after)
+        if: ${{ always() && inputs.cleanup_after != 'false' }}
+        run: |
+          ./auto-testing/rustfs_pool_expand.sh --reset -y
+
+      - name: Notify on failure
+        if: failure()
+        run: |
+          echo "RustFS pool expansion test failed"
+          echo "Package source: ${{ inputs.package_url || inputs.rustfs_version || 'nightly (R2 latest)' }}"
+          echo "See the uploaded log artifact for details."
+
+  # Heal regression runs after the pool test regardless of its outcome: a pool
+  # failure must be reported (it makes the run red) but must not block heal.
+  heal-test:
+    name: Heal test (after pool test)
+    runs-on: smoke-testing
+    timeout-minutes: 480
+    needs: pool-expansion-test
+    if: ${{ always() && (github.event_name == 'workflow_dispatch' || github.event.workflow_run.conclusion == 'success') }}
+    steps:
+      - name: Checkout auto-testing scripts
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          repository: rustfs/auto-testing
+          ref: main
+          path: auto-testing
+          persist-credentials: false
+          token: ${{ secrets.PF_TESTING_GH_TOKEN }}
+
+      - name: Reset test environment (before)
+        if: ${{ inputs.cleanup_before != 'false' }}
+        run: |
+          chmod +x auto-testing/rustfs_heal_test.sh
+          ./auto-testing/rustfs_heal_test.sh --reset -y
+
+      - name: Install RustFS package & start cluster
+        run: |
+          ARGS=(--steps "1,2" -y --endpoint "${{ env.RUSTFS_API_ENDPOINT }}")
+          if [ -n "${{ inputs.package_url }}" ]; then
+            ARGS+=(--package-url "${{ inputs.package_url }}")
+          else
+            ARGS+=(--package-url "${{ env.RUSTFS_NIGHTLY_PACKAGE_URL }}")
+          fi
+          ./auto-testing/rustfs_heal_test.sh "${ARGS[@]}"
+
+      - name: Preflight checks
+        run: |
+          ARGS=(--preflight --endpoint "${{ env.RUSTFS_API_ENDPOINT }}")
+          if [ -n "${{ inputs.package_url }}" ]; then
+            ARGS+=(--package-url "${{ inputs.package_url }}")
+          else
+            ARGS+=(--package-url "${{ env.RUSTFS_NIGHTLY_PACKAGE_URL }}")
+          fi
+          ./auto-testing/rustfs_heal_test.sh "${ARGS[@]}"
+
+      - name: Run heal test (write -> outage -> heal -> verify)
+        run: |
+          ./auto-testing/rustfs_heal_test.sh \
+            --steps 3,4,5,6,7 -y \
+            --endpoint "${{ env.RUSTFS_API_ENDPOINT }}" \
+            --stop-node-gb "${{ inputs.stop_node_gb || '15' }}" \
+            --warp-stop-gb "${{ inputs.warp_stop_gb || '40' }}" \
+            --log-file /tmp/rustfs-heal-test.log
+
+      - name: Upload test logs
+        if: always()
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
+        with:
+          name: rustfs-heal-test-${{ github.run_id }}
+          path: |
+            /tmp/rustfs-heal-test.log
+            /tmp/rustfs-warp.*.log
+          if-no-files-found: warn
+
+      - name: Reset test environment (after)
+        if: ${{ always() && inputs.cleanup_after != 'false' }}
+        run: |
+          ./auto-testing/rustfs_heal_test.sh --reset -y
+
+      - name: Notify on failure
+        if: failure()
+        run: |
+          echo "RustFS heal test failed"
+          echo "See the uploaded log artifact for details."
@@ -0,0 +1,192 @@
+# Copyright 2024 RustFS Team
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+name: Targets Integration
+
+on:
+  pull_request:
+    branches: [main]
+    paths:
+      - ".github/actions/setup/**"
+      - ".github/workflows/targets-integration.yml"
+      - "crates/targets/**"
+      - "Cargo.lock"
+  schedule:
+    - cron: "17 2 * * *"
+      timezone: "Asia/Shanghai"
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: targets-integration-${{ github.ref }}-${{ github.event_name }}
+  cancel-in-progress: ${{ github.event_name != 'schedule' }}
+
+env:
+  CARGO_TERM_COLOR: always
+  RUST_BACKTRACE: 1
+
+jobs:
+  targets-live:
+    name: PostgreSQL, MySQL, AMQP, and NATS
+    runs-on: ubuntu-latest
+    timeout-minutes: 90
+    env:
+      FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
+      NO_PROXY: 127.0.0.1,localhost
+      RUSTFS_TEST_PG_DSN: postgres://postgres:[email protected]:5432/rustfs_events
+      RUSTFS_TEST_MYSQL_DSN: root:testpass@tcp(127.0.0.1:3306)/testdb
+      RUSTFS_TEST_AMQP_URL: amqp://rustfs:[email protected]:5672/%2f
+      RUSTFS_TEST_NATS_URL: nats://127.0.0.1:4222
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          persist-credentials: false
+
+      - name: Setup Rust environment
+        uses: ./.github/actions/setup
+        with:
+          cache-shared-key: targets-live-lane
+          cache-save-if: ${{ github.ref == 'refs/heads/main' || github.event_name == 'schedule' }}
+          install-build-packaging-tools: 'false'
+          install-test-tools: 'false'
+
+      - name: Start target services
+        run: |
+          set -euo pipefail
+          mkdir -p artifacts/targets-live/services
+          docker run -d --name rustfs-targets-postgres \
+            -e POSTGRES_PASSWORD=rustfs \
+            -e POSTGRES_DB=rustfs_events \
+            -p 5432:5432 postgres:16
+          docker run -d --name rustfs-targets-mysql \
+            -e MYSQL_ROOT_PASSWORD=testpass \
+            -e MYSQL_DATABASE=testdb \
+            -p 3306:3306 mysql:8.0.36
+          docker run -d --name rustfs-targets-rabbitmq \
+            -e RABBITMQ_DEFAULT_USER=rustfs \
+            -e RABBITMQ_DEFAULT_PASS=rustfs \
+            -p 5672:5672 rabbitmq:3
+          docker run -d --name rustfs-targets-nats \
+            -p 4222:4222 -p 8222:8222 nats:2 -js -m 8222
+
+          for _ in $(seq 1 120); do
+            docker exec rustfs-targets-postgres pg_isready -U postgres -d rustfs_events >/dev/null 2>&1 && break
+            sleep 1
+          done
+          docker exec rustfs-targets-postgres pg_isready -U postgres -d rustfs_events
+
+          for _ in $(seq 1 120); do
+            docker exec rustfs-targets-mysql mysqladmin ping -h 127.0.0.1 -uroot -ptestpass --silent >/dev/null 2>&1 && break
+            sleep 1
+          done
+          docker exec rustfs-targets-mysql mysqladmin ping -h 127.0.0.1 -uroot -ptestpass --silent
+
+          for _ in $(seq 1 120); do
+            docker exec rustfs-targets-rabbitmq rabbitmq-diagnostics -q ping >/dev/null 2>&1 && break
+            sleep 1
+          done
+          docker exec rustfs-targets-rabbitmq rabbitmq-diagnostics -q ping
+
+          for _ in $(seq 1 120); do
+            curl -fsS http://127.0.0.1:8222/healthz >/dev/null 2>&1 && break
+            sleep 1
+          done
+          curl -fsS http://127.0.0.1:8222/healthz
+
+      - name: Run live target tests
+        env:
+          CARGO_BUILD_JOBS: "2"
+        run: |
+          set +e
+          timeout --verbose --signal=TERM --kill-after=30s 75m bash <<'TESTS' \
+            2>&1 | tee artifacts/targets-live/tests.log
+          result=0
+
+          echo "::group::PostgreSQL"
+          cargo test --locked -p rustfs-targets --test postgres_integration -- --ignored --test-threads=1 || result=1
+          echo "::endgroup::"
+
+          echo "::group::MySQL"
+          cargo test --locked -p rustfs-targets --test mysql_integration -- --ignored --test-threads=1 || result=1
+          echo "::endgroup::"
+
+          echo "::group::AMQP"
+          cargo test --locked -p rustfs-targets --test amqp_integration -- --ignored --test-threads=1 || result=1
+          echo "::endgroup::"
+
+          echo "::group::NATS integration"
+          cargo test --locked -p rustfs-targets --test nats_jetstream_validation_integration -- --ignored --test-threads=1 || result=1
+          cargo test --locked -p rustfs-targets --test nats_jetstream_regression_guards -- --ignored --test-threads=1 || result=1
+          cargo test --locked -p rustfs-targets --lib target::nats::jetstream -- --ignored --test-threads=1 || result=1
+          echo "::endgroup::"
+
+          exit "${result}"
+          TESTS
+          status=${PIPESTATUS[0]}
+          {
+            echo "exit_status=${status}"
+            echo "finished_at=$(date --utc --iso-8601=seconds)"
+            echo
+            echo "Remaining test-related processes:"
+            pgrep -af 'cargo|target/.*/deps/' || true
+          } > artifacts/targets-live/diagnostics.txt
+          exit "${status}"
+
+      - name: Collect service logs
+        if: always()
+        run: |
+          mkdir -p artifacts/targets-live/services
+          for container in postgres mysql rabbitmq nats; do
+            docker logs --tail 500 "rustfs-targets-${container}" \
+              > "artifacts/targets-live/services/${container}.log" 2>&1 || true
+          done
+
+      - name: Stop target services
+        if: always()
+        run: |
+          docker rm -f \
+            rustfs-targets-postgres \
+            rustfs-targets-mysql \
+            rustfs-targets-rabbitmq \
+            rustfs-targets-nats >/dev/null 2>&1 || true
+
+      - name: Upload target integration diagnostics
+        if: always()
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
+        with:
+          name: targets-integration-${{ github.run_number }}-${{ github.run_attempt }}
+          path: artifacts/targets-live
+
+  alert-on-failure:
+    name: Alert on scheduled failure
+    needs: [targets-live]
+    if: >-
+      always() && github.event_name == 'schedule' &&
+      (contains(needs.*.result, 'failure') || contains(needs.*.result, 'cancelled'))
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    permissions:
+      contents: read
+      issues: write
+    steps:
+      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          persist-credentials: false
+      - name: Open or update failure-tracking issue
+        uses: ./.github/actions/schedule-failure-issue
+        with:
+          github-token: ${{ secrets.GITHUB_TOKEN }}
@@ -31,8 +31,13 @@ This file contains repository-wide rules. Use the nearest subdirectory
 - An existing clean, isolated task worktree is sufficient. Create another
  worktree only when the current checkout is shared, dirty with unrelated work,
  or belongs to another task.
- Never commit from a shared checkout. Use an `overtrue/` feature branch unless
-  the user requests another name.
+- Never commit from a shared checkout.
+- Use a task-specific branch named `<type>/<topic>`, such as `fix/...`,
+  `feat/...`, `test/...`, or `docs/...`, unless the user specifies a name.
+- Do not include agent, tool, contributor, account, or organization names in
+  branch names.
+- Push to the user-requested remote or the repository's configured push remote.
+  Do not hard-code or infer a remote from an account name.
 - Check free space before artifact-heavy builds, tests, coverage, or downloads.
  Re-check before a broad gate when space is tight.
 - Remove only task-owned temporary/build artifacts. Never delete another task's
@@ -73,7 +73,7 @@ The main crate is organized in layers, top to bottom:
 |-------|-----------|----------------|
 | **Server** | `server/` | HTTP listener, TLS, CORS, compression, middleware, graceful shutdown |
 | **Admin** | `admin/` | Admin API routing, 30+ handler modules, web console |
-| **App** | `app/` | Use-case orchestration: object_usecase, bucket_usecase, multipart_usecase |
+| **App** | `app/` | Use-case orchestration: object (per-operation modules under `app/object/`, re-exported as `object_usecase`), bucket_usecase, multipart_usecase |
 | **Storage** | `storage/` | S3 API translation, erasure-coded FS, SSE encryption, RPC, concurrency |
 | **Auth** | `auth.rs` | S3 signature verification, credential validation |
 | **Config** | `config/` | CLI parsing, config struct, workload profiles |
@@ -92,8 +92,8 @@ refactors.

 | Domain | Current workspace crates | Responsibility |
 |--------|--------------------------|----------------|
-| Foundation | `checksums`, `common`, `config`, `data-usage`, `utils` | Shared configuration, data-usage models, utilities, and checksums. |
-| I/O and storage | `concurrency`, `ecstore`, `filemeta`, `heal`, `io-core`, `io-metrics`, `lifecycle`, `lock`, `object-capacity`, `object-data-cache`, `replication`, `rio`, `rio-v2`, `scanner`, `storage-api` | Erasure-coded object storage, metadata, recovery, lifecycle, replication, locking, cache, and I/O pipelines. |
+| Foundation | `checksums`, `common`, `config`, `data-usage`, `heal-contracts`, `scanner-contracts`, `utils` | Shared configuration, data-usage models, heal/scanner domain contracts, utilities, and checksums. |
+| I/O and storage | `concurrency`, `ecstore`, `filemeta`, `heal`, `io-core`, `io-metrics`, `lifecycle`, `lock`, `object-capacity`, `object-data-cache`, `replication`, `rio`, `rio-v2`, `s3-client`, `scanner`, `storage-api` | Erasure-coded object storage, metadata, recovery, lifecycle, replication, locking, cache, I/O pipelines, and the engine-side S3 client for remote tier/transition targets. |
 | Security and identity | `credentials`, `crypto`, `iam`, `keystone`, `kms`, `policy`, `security-governance`, `signer`, `tls-runtime`, `trusted-proxies` | Credentials, authentication, authorization, encryption, key management, TLS, and security contracts. |
 | Protocols and contracts | `extension-schema`, `madmin`, `protos`, `protocols`, `s3-ops`, `s3-types`, `s3select-api`, `s3select-query` | Admin, inter-node, S3, S3 Select, and optional protocol contracts. |
 | Operations and integration | `audit`, `notify`, `obs`, `targets`, `zip` | Auditing, observability, event delivery, notification targets, and archive support. |
@@ -115,8 +115,15 @@ default build (lifecycle:
 1. **Layers flow downward.** Server → Admin/App → Storage → ecstore → rio/io-core.
   No upward imports.

-2. **Leaf crates have zero internal dependencies.** `config`, `credentials`, `crypto`,
-   `io-metrics`, and `madmin` should depend only on external crates.
+2. **Leaf crates depend only on external crates, with adjudicated exceptions
+   pinned by a guard.** `config`, `credentials`, and `crypto` take no internal
+   dependency. `io-metrics` takes exactly `rustfs-s3-ops` (transitively
+   `rustfs-s3-types`), a pure contract crate with no I/O and no global state —
+   adjudicated in rustfs/backlog#1834. `madmin` left the leaf set when #6166 made
+   it the SigV4-signed admin SDK client; its internal dependency surface is pinned
+   to exactly `rustfs-signer`. Both pins live in the leaf allowlist in
+   `scripts/check_architecture_migration_rules.sh`; any other internal dependency
+   fails the guard ([crate boundaries](docs/architecture/crate-boundaries.md)).
   - ✅ RESOLVED: the historical `utils → config` and `common → filemeta`/`madmin`
     edges were removed; do not reintroduce them (see Known Structural Issues).

@@ -128,7 +135,7 @@ default build (lifecycle:
     `crates/ecstore/src/bucket/replication/replication_state.rs`) — a naming
     collision, not copies; renaming is tracked in rustfs/backlog#1847.
   - `LastMinuteLatency` has two deliberately different implementations: the
-     per-second bucketed accumulator in `crates/common/src/last_minute.rs` and
+     per-second bucketed accumulator in `crates/scanner-contracts/src/last_minute.rs` and
     the in-memory endpoint-health sample tracker in
     `crates/ecstore/src/bucket/bucket_target_sys.rs` (its doc comment explains
     why it stays local).
@@ -138,15 +145,19 @@ default build (lifecycle:
     `BackpressureSettings` copy that lingered in io-metrics was removed
     (rustfs/backlog#1833).

-4. **ecstore does not know about HTTP or S3 protocol details.** It operates on
-   storage-level abstractions (objects, buckets, disks, pools).
-   - ⚠️ VIOLATED: 58 files under `crates/ecstore/src` reference `s3s`
-     (`rg -l 's3s' crates/ecstore/src | wc -l`), `crates/ecstore/src/client/`
-     is a ~9.4K-line embedded S3 HTTP client, and `crates/ecstore/Cargo.toml`
-     depends on `s3s`, `http`, `hyper`/`hyper-util`/`hyper-rustls`, and
-     `reqwest`. Target state: the engine's need to act as an S3 client
-     (tiering, replication targets) is served by an extracted client crate,
-     and ecstore holds no wire or DTO types.
+4. **ecstore does not *serve* HTTP or the S3 wire protocol.** It operates on
+   storage-level abstractions (objects, buckets, disks, pools) and holds no
+   wire or DTO types of the serving surface. *Consuming* remote S3-compatible
+   endpoints (ILM tier warm backends, transition targets) is a legitimate
+   engine capability, but it lives in the dedicated `rustfs-s3-client` crate
+   (`crates/s3-client`, extracted from the formerly embedded
+   `crates/ecstore/src/client/` by rustfs/backlog#1842), not inside ecstore.
+   - ⚠️ PARTIALLY VIOLATED: serving-side `s3s` references remain in ecstore
+     (bucket metadata/replication/lifecycle DTOs and error mapping). The
+     count is ratcheted shrink-only by `scripts/check_s3s_footprint.sh`
+     (`S3S_ECSTORE_FILES_BASELINE`; the `object_lock` module was converted to
+     storage-level types as the first ratchet step). Target state: the
+     baseline reaches zero and ecstore's `Cargo.toml` drops `s3s`.

 5. **The `rustfs` binary crate is the only place that wires everything together.**
   Individual crates should be testable in isolation.
@@ -321,6 +332,8 @@ The binary (`main.rs`) boots in this order:

 - **"Where is replication configured?"**
  `admin/handlers/replication.rs` and `admin/handlers/site_replication.rs` for API,
+  `rustfs/src/site_replication/` for the site-replication service subsystem
+  (state, peer transport, retry queue, repair, hooks),
  `ecstore/src/bucket/replication/` for engine

 - **"Where do I add a new admin endpoint?"**
@@ -26,6 +26,7 @@ members = [
    "crates/e2e_test", # End-to-end test suite
    "crates/filemeta", # File metadata management
    "crates/heal", # Erasure set and object healing
+    "crates/heal-contracts", # Heal request/response channel contracts
    "crates/iam", # Identity and Access Management
    "crates/keystone", # OpenStack Keystone integration
    "crates/lifecycle", # Lifecycle rule evaluation contracts
@@ -44,11 +45,13 @@ members = [
    "crates/rio-v2", # MinIO on-disk format compatibility I/O layer (feature-gated, ships in no default build)
    "crates/replication", # Replication contracts and wire formats
    "crates/concurrency", # Concurrency management for RustFS - timeout, locking, backpressure, and I/O scheduling
+    "crates/s3-client", # S3 client for engine-side consumption of remote S3 endpoints (tiering, transition targets)
    "crates/s3-types", # S3 event type definitions
    "crates/s3-ops", # S3 operation definitions and mapping
    "crates/s3select-api", # S3 Select API interface
    "crates/s3select-query", # S3 Select query engine
    "crates/scanner", # Scanner for data integrity checks and health monitoring
+    "crates/scanner-contracts", # Scanner metrics and cycle contracts
    "crates/security-governance", # Security governance contracts
    "crates/extension-schema", # Extension schema contracts
    "crates/signer", # client signer
@@ -69,7 +72,7 @@ edition = "2024"
 license = "Apache-2.0"
 repository = "https://github.com/rustfs/rustfs"
 rust-version = "1.97.1"
-version = "1.0.0-rc.3"
+version = "1.0.0-rc.4"
 homepage = "https://rustfs.com"
 description = "RustFS is a high-performance distributed object storage software built using Rust, one of the most popular languages worldwide. "
 keywords = ["RustFS", "Minio", "object-storage", "filesystem", "s3"]
@@ -86,52 +89,55 @@ redundant_clone = "warn"

 [workspace.dependencies]
 # RustFS Internal Crates
-rustfs = { path = "./rustfs", version = "1.0.0-rc.3" }
-rustfs-heal = { path = "crates/heal", version = "1.0.0-rc.3" }
-rustfs-audit = { path = "crates/audit", version = "1.0.0-rc.3" }
-rustfs-checksums = { path = "crates/checksums", version = "1.0.0-rc.3" }
-rustfs-common = { path = "crates/common", version = "1.0.0-rc.3" }
-rustfs-data-usage = { path = "crates/data-usage", version = "1.0.0-rc.3" }
-rustfs-config = { path = "./crates/config", version = "1.0.0-rc.3" }
-rustfs-concurrency = { path = "./crates/concurrency", version = "1.0.0-rc.3" }
-rustfs-credentials = { path = "crates/credentials", version = "1.0.0-rc.3" }
-rustfs-crypto = { path = "crates/crypto", version = "1.0.0-rc.3" }
-rustfs-ecstore = { path = "crates/ecstore", version = "1.0.0-rc.3" }
-rustfs-filemeta = { path = "crates/filemeta", version = "1.0.0-rc.3" }
-rustfs-iam = { path = "crates/iam", version = "1.0.0-rc.3" }
-rustfs-keystone = { path = "crates/keystone", version = "1.0.0-rc.3" }
-rustfs-lifecycle = { path = "crates/lifecycle", version = "1.0.0-rc.3" }
-rustfs-kms = { path = "crates/kms", version = "1.0.0-rc.3" }
-rustfs-lock = { path = "crates/lock", version = "1.0.0-rc.3" }
-rustfs-madmin = { path = "crates/madmin", version = "1.0.0-rc.3" }
-rustfs-notify = { path = "crates/notify", version = "1.0.0-rc.3" }
-rustfs-io-metrics = { path = "crates/io-metrics", version = "1.0.0-rc.3" }
-rustfs-io-core = { path = "crates/io-core", version = "1.0.0-rc.3" }
-rustfs-object-capacity = { path = "crates/object-capacity", version = "1.0.0-rc.3" }
-rustfs-object-data-cache = { path = "crates/object-data-cache", version = "1.0.0-rc.3", default-features = false }
-rustfs-log-analyzer = { path = "crates/log-analyzer", version = "1.0.0-rc.3" }
-rustfs-obs = { path = "crates/obs", version = "1.0.0-rc.3" }
-rustfs-policy = { path = "crates/policy", version = "1.0.0-rc.3" }
-rustfs-protos = { path = "crates/protos", version = "1.0.0-rc.3" }
-rustfs-protocols = { path = "crates/protocols", version = "1.0.0-rc.3" }
-rustfs-replication = { path = "crates/replication", version = "1.0.0-rc.3" }
-rustfs-rio = { path = "crates/rio", version = "1.0.0-rc.3" }
-rustfs-rio-v2 = { path = "crates/rio-v2", version = "1.0.0-rc.3" }
-rustfs-s3-types = { path = "crates/s3-types", version = "1.0.0-rc.3" }
-rustfs-s3-ops = { path = "crates/s3-ops", version = "1.0.0-rc.3" }
-rustfs-s3select-api = { path = "crates/s3select-api", version = "1.0.0-rc.3" }
-rustfs-s3select-query = { path = "crates/s3select-query", version = "1.0.0-rc.3" }
-rustfs-scanner = { path = "crates/scanner", version = "1.0.0-rc.3" }
-rustfs-security-governance = { path = "crates/security-governance", version = "1.0.0-rc.3" }
-rustfs-extension-schema = { path = "crates/extension-schema", version = "1.0.0-rc.3" }
-rustfs-signer = { path = "crates/signer", version = "1.0.0-rc.3" }
-rustfs-storage-api = { path = "crates/storage-api", version = "1.0.0-rc.3" }
-rustfs-trusted-proxies = { path = "crates/trusted-proxies", version = "1.0.0-rc.3" }
-rustfs-targets = { path = "crates/targets", version = "1.0.0-rc.3" }
-rustfs-test-utils = { path = "crates/test-utils", version = "1.0.0-rc.3" }
-rustfs-tls-runtime = { path = "crates/tls-runtime", version = "1.0.0-rc.3" }
-rustfs-utils = { path = "crates/utils", version = "1.0.0-rc.3" }
-rustfs-zip = { path = "./crates/zip", version = "1.0.0-rc.3" }
+rustfs = { path = "./rustfs", version = "1.0.0-rc.4" }
+rustfs-heal = { path = "crates/heal", version = "1.0.0-rc.4" }
+rustfs-heal-contracts = { path = "crates/heal-contracts", version = "1.0.0-rc.4" }
+rustfs-scanner-contracts = { path = "crates/scanner-contracts", version = "1.0.0-rc.4" }
+rustfs-audit = { path = "crates/audit", version = "1.0.0-rc.4" }
+rustfs-checksums = { path = "crates/checksums", version = "1.0.0-rc.4" }
+rustfs-common = { path = "crates/common", version = "1.0.0-rc.4" }
+rustfs-data-usage = { path = "crates/data-usage", version = "1.0.0-rc.4" }
+rustfs-config = { path = "./crates/config", version = "1.0.0-rc.4" }
+rustfs-concurrency = { path = "./crates/concurrency", version = "1.0.0-rc.4" }
+rustfs-credentials = { path = "crates/credentials", version = "1.0.0-rc.4" }
+rustfs-crypto = { path = "crates/crypto", version = "1.0.0-rc.4" }
+rustfs-ecstore = { path = "crates/ecstore", version = "1.0.0-rc.4" }
+rustfs-filemeta = { path = "crates/filemeta", version = "1.0.0-rc.4" }
+rustfs-iam = { path = "crates/iam", version = "1.0.0-rc.4" }
+rustfs-keystone = { path = "crates/keystone", version = "1.0.0-rc.4" }
+rustfs-lifecycle = { path = "crates/lifecycle", version = "1.0.0-rc.4" }
+rustfs-kms = { path = "crates/kms", version = "1.0.0-rc.4" }
+rustfs-lock = { path = "crates/lock", version = "1.0.0-rc.4" }
+rustfs-madmin = { path = "crates/madmin", version = "1.0.0-rc.4" }
+rustfs-notify = { path = "crates/notify", version = "1.0.0-rc.4" }
+rustfs-io-metrics = { path = "crates/io-metrics", version = "1.0.0-rc.4" }
+rustfs-io-core = { path = "crates/io-core", version = "1.0.0-rc.4" }
+rustfs-object-capacity = { path = "crates/object-capacity", version = "1.0.0-rc.4" }
+rustfs-object-data-cache = { path = "crates/object-data-cache", version = "1.0.0-rc.4", default-features = false }
+rustfs-log-analyzer = { path = "crates/log-analyzer", version = "1.0.0-rc.4" }
+rustfs-obs = { path = "crates/obs", version = "1.0.0-rc.4" }
+rustfs-policy = { path = "crates/policy", version = "1.0.0-rc.4" }
+rustfs-protos = { path = "crates/protos", version = "1.0.0-rc.4" }
+rustfs-protocols = { path = "crates/protocols", version = "1.0.0-rc.4" }
+rustfs-replication = { path = "crates/replication", version = "1.0.0-rc.4" }
+rustfs-rio = { path = "crates/rio", version = "1.0.0-rc.4" }
+rustfs-rio-v2 = { path = "crates/rio-v2", version = "1.0.0-rc.4" }
+rustfs-s3-client = { path = "crates/s3-client", version = "1.0.0-rc.4" }
+rustfs-s3-types = { path = "crates/s3-types", version = "1.0.0-rc.4" }
+rustfs-s3-ops = { path = "crates/s3-ops", version = "1.0.0-rc.4" }
+rustfs-s3select-api = { path = "crates/s3select-api", version = "1.0.0-rc.4" }
+rustfs-s3select-query = { path = "crates/s3select-query", version = "1.0.0-rc.4" }
+rustfs-scanner = { path = "crates/scanner", version = "1.0.0-rc.4" }
+rustfs-security-governance = { path = "crates/security-governance", version = "1.0.0-rc.4" }
+rustfs-extension-schema = { path = "crates/extension-schema", version = "1.0.0-rc.4" }
+rustfs-signer = { path = "crates/signer", version = "1.0.0-rc.4" }
+rustfs-storage-api = { path = "crates/storage-api", version = "1.0.0-rc.4" }
+rustfs-trusted-proxies = { path = "crates/trusted-proxies", version = "1.0.0-rc.4" }
+rustfs-targets = { path = "crates/targets", version = "1.0.0-rc.4" }
+rustfs-test-utils = { path = "crates/test-utils", version = "1.0.0-rc.4" }
+rustfs-tls-runtime = { path = "crates/tls-runtime", version = "1.0.0-rc.4" }
+rustfs-utils = { path = "crates/utils", version = "1.0.0-rc.4" }
+rustfs-zip = { path = "./crates/zip", version = "1.0.0-rc.4" }

 # Async Runtime and Networking
 async-channel = "2.5.0"
@@ -147,7 +153,7 @@ futures-core = "0.3.34"
 futures-lite = "2.6.1"
 futures-util = "0.3.34"
 pollster = "1.0.1"
-pulsar = { default-features = false, version = "6.8.0" }
+pulsar = { default-features = false, version = "6.9.0" }
 lapin = { default-features = false, version = "4.10.0" }
 hyper = { version = "1.11.0" }
 hyper-rustls = { default-features = false, version = "0.27.9" }
@@ -192,14 +198,15 @@ serde_urlencoded = "0.7.1"
 # have incompatible APIs. Keep them exact-pinned and monitor upstream for stable
 # releases.
 aes-gcm = { version = "=0.11.1" }
-argon2 = { version = "=0.6.0-rc.8" }
-blake2 = "=0.11.0-rc.6"
+argon2 = { version = "=0.6.0" }
+blake2 = "=0.11.0"
 chacha20poly1305 = { version = "=0.11.0" }
 crc-fast = "1.10.0"
 hmac = { version = "0.13.0" }
 jsonwebtoken = { version = "11.0.0" }
 openidconnect = { default-features = false, version = "4.0" }
 pbkdf2 = "0.13.0"
+p256 = { version = "0.14.0", features = ["ecdsa", "pkcs8"] }
 rsa = { version = "=0.10.0-rc.18" }
 rustls = { default-features = false, version = "0.23.43" }
 rustls-native-certs = "0.8"
@@ -209,6 +216,7 @@ sha1 = "0.11.0"
 sha2 = "0.11.0"
 subtle = "2.6"
 zeroize = { version = "1.9.0" }
+proptest = "1"

 # Time and Date
 chrono = { version = "0.4.45" }
@@ -229,18 +237,17 @@ atoi = "3.1.0"
 atomic_enum = "0.3.0"
 aws-config = { version = "1.11.0" }
 aws-credential-types = { version = "1.3.0" }
-aws-sdk-kms = { default-features = false, version = "1.116.0" }
-aws-sdk-s3 = { default-features = false, version = "1.143.0" }
-aws-sdk-sts = { default-features = false, version = "1.112.0" }
+aws-sdk-kms = { default-features = false, version = "1.117.0" }
+aws-sdk-s3 = { default-features = false, version = "1.144.0" }
+aws-sdk-sts = { default-features = false, version = "1.113.0" }
 aws-smithy-http-client = { default-features = false, version = "1.4.0" }
 aws-smithy-runtime-api = { version = "1.15.0" }
 aws-smithy-types = { version = "1.6.2" }
-base64 = "0.23.1"
 base64-simd = "0.8.0"
 brotli = "8.0.4"
 clap = { version = "4.6.6" }
 const-str = { version = "1.1.0" }
-convert_case = "0.11.0"
+convert_case = "0.12.0"
 criterion = { version = "0.8" }
 crossbeam-queue = "0.3.13"
 crossbeam-channel = "0.5.16"
@@ -252,10 +259,12 @@ enumset = "1.1.14"
 faster-hex = "0.10.0"
 flate2 = "1.1.9"
 glob = "0.3.4"
-google-cloud-storage = "1.17.0"
-google-cloud-auth = "1.15.0"
+google-cloud-storage = "1.18.0"
+google-cloud-auth = "1.16.0"
 hashbrown = { version = "0.17.1" }
-hex = "0.4.3"
+# Base32 for RFC 6238 TOTP shared secrets (RFC 4648 unpadded, the alphabet
+# every authenticator app expects). Already in the graph transitively.
+data-encoding = "2.11.1"
 hex-simd = "0.8.0"
 highway = { version = "1.3.0" }
 hostname = "0.4.2"
@@ -277,6 +286,10 @@ nvml-wrapper = "0.12.1"
 parking_lot = "0.12.5"
 path-absolutize = "4.0.1"
 percent-encoding = "2.3.2"
+# Server-side QR rendering for TOTP enrollment, so neither the console nor the
+# CLI needs its own QR encoder. No default features: the image/render backends
+# pull in an image stack this only needs SVG and text output from.
+qrcode-rs = { version = "2.0.0", default-features = false, features = ["std", "svg"] }
 pin-project-lite = "0.2.17"
 pretty_assertions = "1.4.1"
 rand = { version = "0.10.2" }
@@ -291,14 +304,14 @@ rustify = { version = "0.7", default-features = false }
 rustix = { version = "1.1.4" }
 rust-embed = { version = "8.12.0" }
 rustc-hash = { version = "2.1.3" }
-s3s = { git = "https://github.com/rustfs/s3s.git", rev = "e080e38c56a3b43acbacce55710d765a5ce9003d" }
+s3s = { git = "https://github.com/rustfs/s3s.git", rev = "6e7b41252c7ba218a90886f58d297716ddf68acf", version = "0.15.0", features = ["minio"] }
 serial_test = "4.0.1"
 shadow-rs = { default-features = false, version = "2.0.0" }
 siphasher = "1.0.3"
 smallvec = { version = "1.15.2" }
 compact_str = "0.10.0"
 snap = "1.1.2"
-starshard = { version = "2.2.2" }
+starshard = { version = "2.3.0" }
 strum = { version = "0.28.0" }
 sysinfo = "0.39.6"
 temp-env = "0.3.6"
@@ -314,7 +327,7 @@ tracing-subscriber = { version = "0.3.23" }
 transform-stream = "0.3.1"
 url = "2.5.8"
 urlencoding = "2.1.3"
-uuid = { version = "1.25.0" }
+uuid = { version = "1.26.0" }
 vaultrs = { version = "0.8.0" }
 tar = "0.4.46"
 walkdir = "2.5.0"
@@ -328,7 +341,7 @@ zstd = "0.13.3"
 # Observability and Metrics
 metrics = "0.24.6"
 metrics-util = "0.20"
-dial9-tokio-telemetry = "0.3"
+dial9-tokio-telemetry = "0.5.0"
 opentelemetry = { version = "0.32.0" }
 opentelemetry-appender-tracing = { version = "0.32.0" }
 opentelemetry-otlp = { version = "0.32.0" }
@@ -343,15 +356,14 @@ libunftp = { version = "0.23.0" }
 unftp-core = "0.1.0"
 suppaftp = { version = "10.0.2" }
 rcgen = { version = "0.14.9", default-features = false, features = ["aws_lc_rs", "crypto", "pem"] }
-russh = { version = "0.63.0" }
+russh = { version = "0.63.1" }
 russh-sftp = "2.4.0"

 # WebDAV
 dav-server = "0.11.0"

 # Performance Analysis and Memory Profiling
-rustfs-mimalloc = { version = "0.5.0" }
-rustfs-mimalloc-sys = { version = "0.5.0" }
+rustfs-mimalloc = { version = "0.5.1" }
 hotpath = { version = "0.24.0", default-features = false }
 # Snapshot testing for output format regression detection
 insta = { version = "1.48" }
@@ -371,8 +383,8 @@ opt-level = 3
 lto = "thin"
 codegen-units = 1
 debug = 0
-split-debuginfo = "off"
 strip = "symbols"
+split-debuginfo = "off"

 [profile.production]
 inherits = "release"
@@ -12,12 +12,11 @@
 </p>

 <p align="center">
-<a href="https://trendshift.io/repositories/14181" target="_blank"><img src="https://trendshift.io/api/badge/repositories/14181" alt="rustfs%2Frustfs | Trendshift" style="width: 250px; height: 55px;" width="250" height="55"/></a> 
-<a href="https://runacap.com/ross-index/q4-2025/" target="_blank" rel="noopener"><img style="width: 260px; height: 55px" src="https://runacap.com/wp-content/uploads/2026/01/ROSS_badge_white_Q4_2025.svg" alt="ROSS Index - Fastest Growing Open-Source Startups in Q4 2025 | Runa Capital" height="55" /></a>
+<a href="https://trendshift.io/repositories/14181" target="_blank"><img src="https://trendshift.io/api/badge/repositories/14181" alt="rustfs%2Frustfs | Trendshift" style="width: 250px; height: 55px;" width="250" height="55"/></a>
 </p>

 <p align="center">
-  <a href="https://docs.rustfs.com/installation/">Getting Started</a>
+  <a href="https://docs.rustfs.com/en/installation">Getting Started</a>
  · <a href="https://docs.rustfs.com/">Docs</a>
  · <a href="https://github.com/rustfs/rustfs/issues">Bug reports</a>
  · <a href="https://github.com/rustfs/rustfs/discussions">Discussions</a>
@@ -116,7 +115,7 @@ chown -R 10001:10001 data logs
 docker run -d -p 9000:9000 -p 9001:9001 -v $(pwd)/data:/data -v $(pwd)/logs:/logs rustfs/rustfs:latest

 # Using specific version
-docker run -d -p 9000:9000 -p 9001:9001 -v $(pwd)/data:/data -v $(pwd)/logs:/logs rustfs/rustfs:1.0.0-rc.3
+docker run -d -p 9000:9000 -p 9001:9001 -v $(pwd)/data:/data -v $(pwd)/logs:/logs rustfs/rustfs:1.0.0-rc.4
 ```

 If you use [podman](https://github.com/containers/podman) instead of docker, you can install the RustFS with the below command
@@ -12,12 +12,11 @@


 <p align="center">
-<a href="https://trendshift.io/repositories/14181" target="_blank"><img src="https://trendshift.io/api/badge/repositories/14181" alt="rustfs%2Frustfs | Trendshift" style="width: 250px; height: 55px;" width="250" height="55"/></a> 
-<a href="https://runacap.com/ross-index/q4-2025/" target="_blank" rel="noopener"><img style="width: 260px; height: 55px" src="https://runacap.com/wp-content/uploads/2026/01/ROSS_badge_white_Q4_2025.svg" alt="ROSS Index - Fastest Growing Open-Source Startups in Q4 2025 | Runa Capital" height="55" /></a>
+<a href="https://trendshift.io/repositories/14181" target="_blank"><img src="https://trendshift.io/api/badge/repositories/14181" alt="rustfs%2Frustfs | Trendshift" style="width: 250px; height: 55px;" width="250" height="55"/></a>
 </p>

 <p align="center">
-  <a href="https://docs.rustfs.com/installation/">快速开始</a>
+  <a href="https://docs.rustfs.com/zh/installation">快速开始</a>
  · <a href="https://docs.rustfs.com/">文档</a>
  · <a href="https://github.com/rustfs/rustfs/issues">报告 Bug</a>
  · <a href="https://github.com/rustfs/rustfs/discussions">社区讨论</a>
@@ -113,7 +112,7 @@ chown -R 10001:10001 data logs
 docker run -d -p 9000:9000 -p 9001:9001 -v $(pwd)/data:/data -v $(pwd)/logs:/logs rustfs/rustfs:latest

 # 使用指定版本运行
-docker run -d -p 9000:9000 -p 9001:9001 -v $(pwd)/data:/data -v $(pwd)/logs:/logs rustfs/rustfs:1.0.0-rc.3
+docker run -d -p 9000:9000 -p 9001:9001 -v $(pwd)/data:/data -v $(pwd)/logs:/logs rustfs/rustfs:1.0.0-rc.4
 ```

 如果您通过绑定挂载启用 TLS 证书目录，也请用同样方式准备该目录：
@@ -67,7 +67,7 @@ tokio = { workspace = true, features = ["sync", "fs", "rt-multi-thread", "time",
 tracing = { workspace = true, features = ["std", "attributes"] }

 [dev-dependencies]
-async-trait = { workspace = true }
+rustfs-targets = { workspace = true, features = ["test-support"] }
 temp-env = { workspace = true }
 url = { workspace = true }

@@ -564,88 +564,21 @@ impl AuditRuntimeFacade {
 mod tests {
    use super::AuditPipeline;
    use crate::{AuditEntry, AuditError, AuditRegistry};
-    use async_trait::async_trait;
-    use rustfs_targets::arn::TargetID;
-    use rustfs_targets::store::{Key, Store};
-    use rustfs_targets::target::{EntityTarget, QueuedPayload, QueuedPayloadMeta};
-    use rustfs_targets::{StoreError, Target, TargetError};
+    use rustfs_targets::testkit::MockTarget;
    use std::sync::Arc;
    use tokio::sync::{Mutex, Notify};

-    /// Mock target whose `save()` outcome is fixed at construction so tests can
-    /// force full-success / full-failure / partial-failure fan-outs.
-    #[derive(Clone)]
-    struct MockTarget {
-        id: TargetID,
-        fail: bool,
-        health_gate: Option<(Arc<Notify>, Arc<Notify>)>,
-    }
-
-    impl MockTarget {
-        fn new(id: &str, fail: bool) -> Self {
-            Self {
-                id: TargetID::new(id.to_string(), "webhook".to_string()),
-                fail,
-                health_gate: None,
-            }
-        }
-
-        fn with_health_gate(mut self, started: Arc<Notify>, release: Arc<Notify>) -> Self {
-            self.health_gate = Some((started, release));
-            self
-        }
-    }
-
-    #[async_trait]
-    impl<E> Target<E> for MockTarget
-    where
-        E: rustfs_targets::PluginEvent,
-    {
-        fn id(&self) -> TargetID {
-            self.id.clone()
-        }
-
-        async fn is_active(&self) -> Result<bool, TargetError> {
-            if let Some((started, release)) = &self.health_gate {
-                started.notify_one();
-                release.notified().await;
-            }
-            Ok(true)
-        }
-
-        async fn save(&self, _event: Arc<EntityTarget<E>>) -> Result<(), TargetError> {
-            if self.fail {
-                Err(TargetError::Configuration("forced save failure".to_string()))
-            } else {
-                Ok(())
-            }
-        }
-
-        async fn send_raw_from_store(&self, _key: Key, _body: Vec<u8>, _meta: QueuedPayloadMeta) -> Result<(), TargetError> {
-            Ok(())
-        }
-
-        async fn close(&self) -> Result<(), TargetError> {
-            Ok(())
-        }
-
-        fn store(&self) -> Option<&(dyn Store<QueuedPayload, Error = StoreError, Key = Key> + Send + Sync)> {
-            None
-        }
-
-        fn clone_dyn(&self) -> Box<dyn Target<E> + Send + Sync> {
-            Box::new(self.clone())
-        }
-
-        fn is_enabled(&self) -> bool {
-            true
-        }
+    /// Builds a mock target whose `save()` outcome is fixed at construction so tests can force
+    /// full-success / full-failure / partial-failure fan-outs.
+    fn mock_target(id: &str, fail: bool) -> MockTarget {
+        let target = MockTarget::new(id, "webhook");
+        if fail { target.with_save_failures(usize::MAX) } else { target }
    }

    fn pipeline_with(targets: Vec<MockTarget>) -> AuditPipeline {
        let mut registry = AuditRegistry::new();
        for target in targets {
-            registry.add_target(target.id.to_string(), Box::new(target));
+            registry.add_target(target.target_id().to_string(), Box::new(target));
        }
        AuditPipeline::new(Arc::new(Mutex::new(registry)))
    }
@@ -658,7 +591,7 @@ mod tests {
    // dispatch must return Err rather than swallowing the failures as Ok.
    #[tokio::test]
    async fn dispatch_returns_err_when_all_targets_fail() {
-        let pipeline = pipeline_with(vec![MockTarget::new("a:webhook", true), MockTarget::new("b:webhook", true)]);
+        let pipeline = pipeline_with(vec![mock_target("a:webhook", true), mock_target("b:webhook", true)]);
        let result = pipeline.dispatch(entry()).await;
        assert!(matches!(result, Err(AuditError::Target(_))), "expected Err, got {result:?}");
    }
@@ -667,13 +600,13 @@ mod tests {
    // so dispatch reports success (degradation is logged, not propagated).
    #[tokio::test]
    async fn dispatch_returns_ok_on_partial_failure() {
-        let pipeline = pipeline_with(vec![MockTarget::new("ok:webhook", false), MockTarget::new("bad:webhook", true)]);
+        let pipeline = pipeline_with(vec![mock_target("ok:webhook", false), mock_target("bad:webhook", true)]);
        pipeline.dispatch(entry()).await.expect("partial success should return Ok");
    }

    #[tokio::test]
    async fn dispatch_returns_ok_when_all_targets_succeed() {
-        let pipeline = pipeline_with(vec![MockTarget::new("a:webhook", false), MockTarget::new("b:webhook", false)]);
+        let pipeline = pipeline_with(vec![mock_target("a:webhook", false), mock_target("b:webhook", false)]);
        pipeline.dispatch(entry()).await.expect("all-success should return Ok");
    }

@@ -686,9 +619,10 @@ mod tests {

    #[tokio::test]
    async fn health_probe_does_not_hold_the_registry_lock() {
-        let started = Arc::new(Notify::new());
        let release = Arc::new(Notify::new());
-        let pipeline = pipeline_with(vec![MockTarget::new("blocked", false).with_health_gate(started.clone(), release.clone())]);
+        let target = mock_target("blocked", false).with_health_gate(release.clone());
+        let started = target.health_started();
+        let pipeline = pipeline_with(vec![target]);
        let registry = Arc::clone(&pipeline.registry);
        let snapshot_task = tokio::spawn(async move { pipeline.snapshot_target_health().await });
        started.notified().await;
@@ -706,14 +640,14 @@ mod tests {
    // whole-batch loss instead of returning Ok.
    #[tokio::test]
    async fn dispatch_batch_returns_err_when_all_targets_fail() {
-        let pipeline = pipeline_with(vec![MockTarget::new("a:webhook", true)]);
+        let pipeline = pipeline_with(vec![mock_target("a:webhook", true)]);
        let result = pipeline.dispatch_batch(vec![entry(), entry()]).await;
        assert!(matches!(result, Err(AuditError::Target(_))), "expected Err, got {result:?}");
    }

    #[tokio::test]
    async fn dispatch_batch_returns_ok_when_all_targets_succeed() {
-        let pipeline = pipeline_with(vec![MockTarget::new("a:webhook", false), MockTarget::new("b:webhook", false)]);
+        let pipeline = pipeline_with(vec![mock_target("a:webhook", false), mock_target("b:webhook", false)]);
        pipeline
            .dispatch_batch(vec![entry(), entry()])
            .await
@@ -286,70 +286,10 @@ impl AuditRegistry {
 #[cfg(test)]
 mod tests {
    use super::AuditRegistry;
-    use crate::{AuditEntry, AuditError};
-    use rustfs_targets::arn::TargetID;
-    use rustfs_targets::store::{Key, Store};
-    use rustfs_targets::target::{ChannelTargetType, EntityTarget, QueuedPayload, QueuedPayloadMeta};
-    use rustfs_targets::{StoreError, Target, TargetError};
-    use std::sync::Arc;
-    use std::sync::atomic::{AtomicUsize, Ordering};
-
-    #[derive(Clone)]
-    struct CloseTestTarget {
-        id: TargetID,
-        close_calls: Arc<AtomicUsize>,
-        fail_on_close: bool,
-    }
-
-    impl CloseTestTarget {
-        fn new(id: TargetID, close_calls: Arc<AtomicUsize>, fail_on_close: bool) -> Self {
-            Self {
-                id,
-                close_calls,
-                fail_on_close,
-            }
-        }
-    }
-
-    #[async_trait::async_trait]
-    impl Target<AuditEntry> for CloseTestTarget {
-        fn id(&self) -> TargetID {
-            self.id.clone()
-        }
-
-        async fn is_active(&self) -> Result<bool, TargetError> {
-            Ok(true)
-        }
-
-        async fn save(&self, _event: Arc<EntityTarget<AuditEntry>>) -> Result<(), TargetError> {
-            Ok(())
-        }
-
-        async fn send_raw_from_store(&self, _key: Key, _body: Vec<u8>, _meta: QueuedPayloadMeta) -> Result<(), TargetError> {
-            Ok(())
-        }
-
-        async fn close(&self) -> Result<(), TargetError> {
-            self.close_calls.fetch_add(1, Ordering::SeqCst);
-            if self.fail_on_close {
-                Err(TargetError::Unknown("close failed".to_string()))
-            } else {
-                Ok(())
-            }
-        }
-
-        fn store(&self) -> Option<&(dyn Store<QueuedPayload, Error = StoreError, Key = Key> + Send + Sync)> {
-            None
-        }
-
-        fn clone_dyn(&self) -> Box<dyn Target<AuditEntry> + Send + Sync> {
-            Box::new(self.clone())
-        }
-
-        fn is_enabled(&self) -> bool {
-            true
-        }
-    }
+    use crate::AuditError;
+    use rustfs_targets::TargetError;
+    use rustfs_targets::target::ChannelTargetType;
+    use rustfs_targets::testkit::MockTarget;

    #[test]
    fn registry_registers_amqp_factory() {
@@ -361,23 +301,21 @@ mod tests {
    #[tokio::test]
    async fn close_all_returns_first_error_and_clears_targets() {
        let mut registry = AuditRegistry::new();
-        let ok_calls = Arc::new(AtomicUsize::new(0));
-        let fail_calls = Arc::new(AtomicUsize::new(0));
+        let ok = MockTarget::new("ok", "webhook");
+        let ok_observer = ok.clone();
+        let fail = MockTarget::new("fail", "webhook")
+            .with_close_failures(usize::MAX)
+            .with_close_failure_error(|| TargetError::Unknown("close failed".to_string()));
+        let fail_observer = fail.clone();

-        let ok_id = TargetID::new("ok".to_string(), "webhook".to_string());
-        let fail_id = TargetID::new("fail".to_string(), "webhook".to_string());
-
-        registry.add_target(ok_id.to_string(), Box::new(CloseTestTarget::new(ok_id, Arc::clone(&ok_calls), false)));
-        registry.add_target(
-            fail_id.to_string(),
-            Box::new(CloseTestTarget::new(fail_id, Arc::clone(&fail_calls), true)),
-        );
+        registry.add_target(ok.target_id().to_string(), Box::new(ok));
+        registry.add_target(fail.target_id().to_string(), Box::new(fail));

        let result = registry.close_all().await;

        assert!(matches!(result, Err(AuditError::Target(TargetError::Unknown(_)))));
-        assert_eq!(ok_calls.load(Ordering::SeqCst), 1);
-        assert_eq!(fail_calls.load(Ordering::SeqCst), 1);
+        assert_eq!(ok_observer.close_call_count(), 1);
+        assert_eq!(fail_observer.close_call_count(), 1);
        assert!(registry.list_targets().is_empty());
    }
 }
@@ -577,76 +577,17 @@ fn warn_audit_state(state: &str, reason: Option<&str>) {
 mod tests {
    use super::{AuditSystem, AuditSystemState};
    use crate::{AuditEntry, AuditError};
-    use async_trait::async_trait;
    use rustfs_targets::ReplayWorkerManager;
-    use rustfs_targets::arn::TargetID;
-    use rustfs_targets::store::{Key, Store};
-    use rustfs_targets::target::{EntityTarget, QueuedPayload, QueuedPayloadMeta};
-    use rustfs_targets::{StoreError, Target, TargetError};
+    use rustfs_targets::testkit::MockTarget;
    use std::collections::HashMap;
    use std::sync::Arc;
-    use std::sync::atomic::{AtomicUsize, Ordering};
    use tokio::sync::mpsc;

-    #[derive(Clone)]
-    struct TestTarget {
-        close_calls: Arc<AtomicUsize>,
-        id: TargetID,
-    }
-
-    impl TestTarget {
-        fn new(id: &str, name: &str) -> Self {
-            Self {
-                close_calls: Arc::new(AtomicUsize::new(0)),
-                id: TargetID::new(id.to_string(), name.to_string()),
-            }
-        }
-    }
-
-    #[async_trait]
-    impl<E> Target<E> for TestTarget
-    where
-        E: rustfs_targets::PluginEvent,
-    {
-        fn id(&self) -> TargetID {
-            self.id.clone()
-        }
-
-        async fn is_active(&self) -> Result<bool, TargetError> {
-            Ok(true)
-        }
-
-        async fn save(&self, _event: Arc<EntityTarget<E>>) -> Result<(), TargetError> {
-            Ok(())
-        }
-
-        async fn send_raw_from_store(&self, _key: Key, _body: Vec<u8>, _meta: QueuedPayloadMeta) -> Result<(), TargetError> {
-            Ok(())
-        }
-
-        async fn close(&self) -> Result<(), TargetError> {
-            self.close_calls.fetch_add(1, Ordering::SeqCst);
-            Ok(())
-        }
-
-        fn store(&self) -> Option<&(dyn Store<QueuedPayload, Error = StoreError, Key = Key> + Send + Sync)> {
-            None
-        }
-
-        fn clone_dyn(&self) -> Box<dyn Target<E> + Send + Sync> {
-            Box::new(self.clone())
-        }
-
-        fn is_enabled(&self) -> bool {
-            true
-        }
-    }
-
    #[tokio::test]
    async fn reload_with_empty_config_stops_existing_runtime() {
        let system = AuditSystem::new();
-        let target = TestTarget::new("primary", "webhook");
-        let close_calls = Arc::clone(&target.close_calls);
+        let target = MockTarget::new("primary", "webhook");
+        let observer = target.clone();

        {
            let mut registry = system.registry.lock().await;
@@ -671,7 +612,7 @@ mod tests {
        assert_eq!(system.get_state().await, AuditSystemState::Stopped);
        assert!(system.list_targets().await.is_empty());
        assert_eq!(system.runtime_status_snapshot().await, ReplayWorkerManager::new().snapshot(0));
-        assert_eq!(close_calls.load(Ordering::SeqCst), 1);
+        assert_eq!(observer.close_call_count(), 1);
        assert_eq!(*system.config.read().await, Some(rustfs_config::server_config::Config(HashMap::new())));
    }

@@ -693,7 +634,7 @@ mod tests {
        // Seed a target + replay worker so both critical sections touch real state.
        {
            let mut registry = system.registry.lock().await;
-            registry.add_target("primary:webhook".to_string(), Box::new(TestTarget::new("primary", "webhook")));
+            registry.add_target("primary:webhook".to_string(), Box::new(MockTarget::new("primary", "webhook")));
        }
        {
            let mut replay_workers = system.stream_cancellers.write().await;
@@ -793,8 +734,8 @@ mod tests {
    async fn commit_closes_old_targets_before_installing_new() {
        let system = AuditSystem::new();

-        let old = TestTarget::new("old", "webhook");
-        let old_close = Arc::clone(&old.close_calls);
+        let old = MockTarget::new("old", "webhook");
+        let old_observer = old.clone();
        {
            let mut registry = system.registry.lock().await;
            registry.add_target("old:webhook".to_string(), Box::new(old));
@@ -809,17 +750,17 @@ mod tests {
            *state = AuditSystemState::Running;
        }

-        let new = TestTarget::new("new", "webhook");
-        let new_close = Arc::clone(&new.close_calls);
+        let new = MockTarget::new("new", "webhook");
+        let new_observer = new.clone();
        system
            .commit_runtime_targets(vec![Box::new(new)], AuditSystemState::Running)
            .await
            .expect("commit should succeed");

        // Old target closed exactly once during the pre-install shutdown.
-        assert_eq!(old_close.load(Ordering::SeqCst), 1);
+        assert_eq!(old_observer.close_call_count(), 1);
        // New target installed and left open.
-        assert_eq!(new_close.load(Ordering::SeqCst), 0);
+        assert_eq!(new_observer.close_call_count(), 0);
        assert_eq!(system.list_targets().await, vec!["new:webhook".to_string()]);
        // Old replay worker stopped; the store-less new target adds none.
        assert_eq!(system.runtime_status_snapshot().await.replay_worker_count, 0);
@@ -12,136 +12,16 @@
 //  See the License for the specific language governing permissions and
 //  limitations under the License.

-use async_trait::async_trait;
 use rustfs_audit::{AuditEntry, AuditError, AuditPipeline, AuditRegistry, AuditRuntimeFacade, AuditRuntimeView};
-use rustfs_targets::arn::TargetID;
-use rustfs_targets::store::{Key, Store};
-use rustfs_targets::target::{EntityTarget, QueuedPayload, QueuedPayloadMeta};
-use rustfs_targets::{SharedTarget, StoreError, Target, TargetError};
-use serde::{Serialize, de::DeserializeOwned};
+use rustfs_targets::SharedTarget;
+use rustfs_targets::testkit::MockTarget;
 use std::sync::Arc;
-use std::sync::atomic::{AtomicUsize, Ordering};
 use tokio::sync::{Mutex, RwLock};

-#[derive(Clone)]
-struct TestTarget {
-    close_calls: Arc<AtomicUsize>,
-    id: TargetID,
-    init_calls: Arc<AtomicUsize>,
-}
-
-impl TestTarget {
-    fn new(id: &str, name: &str) -> Self {
-        Self {
-            close_calls: Arc::new(AtomicUsize::new(0)),
-            id: TargetID::new(id.to_string(), name.to_string()),
-            init_calls: Arc::new(AtomicUsize::new(0)),
-        }
-    }
-}
-
-#[async_trait]
-impl<E> Target<E> for TestTarget
-where
-    E: Send + Sync + 'static + Clone + Serialize + DeserializeOwned,
-{
-    fn id(&self) -> TargetID {
-        self.id.clone()
-    }
-
-    async fn is_active(&self) -> Result<bool, TargetError> {
-        Ok(true)
-    }
-
-    async fn save(&self, _event: Arc<EntityTarget<E>>) -> Result<(), TargetError> {
-        Ok(())
-    }
-
-    async fn send_raw_from_store(&self, _key: Key, _body: Vec<u8>, _meta: QueuedPayloadMeta) -> Result<(), TargetError> {
-        Ok(())
-    }
-
-    async fn close(&self) -> Result<(), TargetError> {
-        self.close_calls.fetch_add(1, Ordering::SeqCst);
-        Ok(())
-    }
-
-    fn store(&self) -> Option<&(dyn Store<QueuedPayload, Error = StoreError, Key = Key> + Send + Sync)> {
-        None
-    }
-
-    fn clone_dyn(&self) -> Box<dyn Target<E> + Send + Sync> {
-        Box::new(self.clone())
-    }
-
-    async fn init(&self) -> Result<(), TargetError> {
-        self.init_calls.fetch_add(1, Ordering::SeqCst);
-        Ok(())
-    }
-
-    fn is_enabled(&self) -> bool {
-        true
-    }
-}
-
-/// A target whose `save()` always fails, used to exercise the dispatch
+/// Builds a target whose `save()` always fails, used to exercise the dispatch
 /// failure-propagation paths.
-#[derive(Clone)]
-struct FailingTarget {
-    id: TargetID,
-    save_calls: Arc<AtomicUsize>,
-}
-
-impl FailingTarget {
-    fn new(id: &str, name: &str) -> Self {
-        Self {
-            id: TargetID::new(id.to_string(), name.to_string()),
-            save_calls: Arc::new(AtomicUsize::new(0)),
-        }
-    }
-}
-
-#[async_trait]
-impl<E> Target<E> for FailingTarget
-where
-    E: Send + Sync + 'static + Clone + Serialize + DeserializeOwned,
-{
-    fn id(&self) -> TargetID {
-        self.id.clone()
-    }
-
-    async fn is_active(&self) -> Result<bool, TargetError> {
-        Ok(true)
-    }
-
-    async fn save(&self, _event: Arc<EntityTarget<E>>) -> Result<(), TargetError> {
-        self.save_calls.fetch_add(1, Ordering::SeqCst);
-        Err(TargetError::Storage("disk full".to_string()))
-    }
-
-    async fn send_raw_from_store(&self, _key: Key, _body: Vec<u8>, _meta: QueuedPayloadMeta) -> Result<(), TargetError> {
-        Ok(())
-    }
-
-    async fn close(&self) -> Result<(), TargetError> {
-        Ok(())
-    }
-
-    fn store(&self) -> Option<&(dyn Store<QueuedPayload, Error = StoreError, Key = Key> + Send + Sync)> {
-        None
-    }
-
-    fn clone_dyn(&self) -> Box<dyn Target<E> + Send + Sync> {
-        Box::new(self.clone())
-    }
-
-    async fn init(&self) -> Result<(), TargetError> {
-        Ok(())
-    }
-
-    fn is_enabled(&self) -> bool {
-        true
-    }
+fn failing_target(id: &str, name: &str) -> MockTarget {
+    MockTarget::new(id, name).with_save_failures(usize::MAX)
 }

 fn pipeline_with_targets(targets: Vec<(&str, SharedTarget<AuditEntry>)>) -> AuditPipeline {
@@ -154,8 +34,8 @@ fn pipeline_with_targets(targets: Vec<(&str, SharedTarget<AuditEntry>)>) -> Audi

 #[tokio::test]
 async fn audit_pipeline_dispatch_propagates_total_failure() {
-    let failing = FailingTarget::new("primary", "webhook");
-    let save_calls = Arc::clone(&failing.save_calls);
+    let failing = failing_target("primary", "webhook");
+    let observer = failing.clone();
    let pipeline = pipeline_with_targets(vec![("primary:webhook", Arc::new(failing))]);

    let result = pipeline.dispatch(Arc::new(AuditEntry::default())).await;
@@ -164,13 +44,13 @@ async fn audit_pipeline_dispatch_propagates_total_failure() {
        matches!(result, Err(AuditError::Target(_))),
        "dispatch must surface an error when every target fails, got {result:?}"
    );
-    assert_eq!(save_calls.load(Ordering::SeqCst), 1, "the failing target should have been invoked");
+    assert_eq!(observer.save_call_count(), 1, "the failing target should have been invoked");
 }

 #[tokio::test]
 async fn audit_pipeline_dispatch_tolerates_partial_failure() {
-    let failing = FailingTarget::new("primary", "webhook");
-    let healthy = TestTarget::new("secondary", "webhook");
+    let failing = failing_target("primary", "webhook");
+    let healthy = MockTarget::new("secondary", "webhook");
    let pipeline = pipeline_with_targets(vec![
        ("primary:webhook", Arc::new(failing)),
        ("secondary:webhook", Arc::new(healthy)),
@@ -186,7 +66,7 @@ async fn audit_pipeline_dispatch_tolerates_partial_failure() {

 #[tokio::test]
 async fn audit_pipeline_dispatch_batch_propagates_total_failure() {
-    let failing = FailingTarget::new("primary", "webhook");
+    let failing = failing_target("primary", "webhook");
    let pipeline = pipeline_with_targets(vec![("primary:webhook", Arc::new(failing))]);

    let entries = vec![Arc::new(AuditEntry::default()), Arc::new(AuditEntry::default())];
@@ -200,8 +80,8 @@ async fn audit_pipeline_dispatch_batch_propagates_total_failure() {

 #[tokio::test]
 async fn audit_pipeline_dispatch_batch_tolerates_partial_failure() {
-    let failing = FailingTarget::new("primary", "webhook");
-    let healthy = TestTarget::new("secondary", "webhook");
+    let failing = failing_target("primary", "webhook");
+    let healthy = MockTarget::new("secondary", "webhook");
    let pipeline = pipeline_with_targets(vec![
        ("primary:webhook", Arc::new(failing)),
        ("secondary:webhook", Arc::new(healthy)),
@@ -266,9 +146,8 @@ async fn audit_runtime_facade_activates_empty_target_list() {
 async fn audit_runtime_view_upsert_and_remove_target() {
    let registry = Arc::new(Mutex::new(AuditRegistry::new()));
    let runtime_view = AuditRuntimeView::new(registry.clone());
-    let target = TestTarget::new("primary", "webhook");
-    let init_calls = Arc::clone(&target.init_calls);
-    let close_calls = Arc::clone(&target.close_calls);
+    let target = MockTarget::new("primary", "webhook");
+    let observer = target.clone();

    runtime_view
        .upsert_target("primary:webhook".to_string(), Box::new(target))
@@ -276,7 +155,7 @@ async fn audit_runtime_view_upsert_and_remove_target() {
        .expect("upsert should succeed");

    assert_eq!(runtime_view.list_targets().await, vec!["primary:webhook".to_string()]);
-    assert_eq!(init_calls.load(Ordering::SeqCst), 1);
+    assert_eq!(observer.init_call_count(), 1);

    runtime_view
        .remove_target("primary:webhook")
@@ -284,7 +163,7 @@ async fn audit_runtime_view_upsert_and_remove_target() {
        .expect("remove should succeed");

    assert!(runtime_view.list_targets().await.is_empty());
-    assert_eq!(close_calls.load(Ordering::SeqCst), 1);
+    assert_eq!(observer.close_call_count(), 1);
 }

 #[tokio::test]
@@ -292,7 +171,7 @@ async fn audit_runtime_facade_replace_targets_commits_runtime_state() {
    let registry = Arc::new(Mutex::new(AuditRegistry::new()));
    let replay_workers = Arc::new(RwLock::new(rustfs_targets::ReplayWorkerManager::new()));
    let facade = AuditRuntimeFacade::new(registry.clone(), replay_workers.clone());
-    let target = TestTarget::new("primary", "webhook");
+    let target = MockTarget::new("primary", "webhook");
    let activation = rustfs_targets::RuntimeActivation {
        replay_workers: rustfs_targets::ReplayWorkerManager::new(),
        targets: vec![Arc::new(target) as rustfs_targets::SharedTarget<rustfs_audit::AuditEntry>],
@@ -41,14 +41,22 @@ pub const XXHASH_64_NAME: &str = "xxhash64";
 pub const XXHASH_128_NAME: &str = "xxhash128";
 pub const MD5_NAME: &str = "md5";

-/// One of three deliberately separate checksum registries (backlog#1833):
-/// this enum owns the **streaming-hash algorithm registry**, including the
-/// RustFS extensions (sha512, xxhash3/64/128). The on-disk xl.meta bitset
-/// lives in `rustfs_rio::ChecksumType` (crates/rio/src/checksum.rs, varint
-/// bits are append-only), and the MinIO-port client keeps its own
-/// `ChecksumMode` (crates/ecstore/src/client/checksum.rs). When adding an
-/// algorithm, extend all three (or record why not) — they do not derive from
-/// each other.
+/// The canonical checksum-algorithm registry (backlog#1833, backlog#1844):
+/// this enum owns the streaming-hash implementations and, via the exhaustive
+/// per-algorithm metadata methods below, the wire names, header names, digest
+/// lengths, and checksum-type capabilities — including the RustFS extensions
+/// (sha512, xxhash3/64/128). The MinIO-port client's `ChecksumMode`
+/// (crates/s3-client/src/checksum.rs) delegates all per-algorithm dispatch
+/// here through its `algorithm()` bridge. The on-disk xl.meta bitset remains
+/// deliberately separate in `rustfs_rio::ChecksumType`
+/// (crates/rio/src/checksum.rs, varint bits are append-only), and rio also
+/// keeps its own hot-path hasher shells — equivalence with this crate's
+/// hashers is enforced by both test suites pinning the same official
+/// known-answer vectors (backlog#1844 PR3 verdict, recorded on
+/// `rustfs_rio::ChecksumType`). When adding an algorithm: add the variant
+/// here (the exhaustive matches force every metadata decision), bridge it in
+/// the client, and allocate an xl.meta bit + hasher + shared vector in rio
+/// (or record why not).
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
 #[non_exhaustive]
 pub enum ChecksumAlgorithm {
@@ -120,6 +128,84 @@ impl ChecksumAlgorithm {
            Self::Xxhash128 => XXHASH_128_NAME,
        }
    }
+
+    // Per-algorithm wire metadata. These matches are deliberately exhaustive
+    // (no `_` arm): adding a ChecksumAlgorithm variant without deciding its
+    // name, header, digest length, and checksum-type support must fail to
+    // compile rather than silently inherit a default (backlog#1844).
+
+    /// The canonical `x-amz-checksum-algorithm` wire value (uppercase), as
+    /// carried in S3 requests/responses and stored checksum maps.
+    pub fn s3_algorithm_name(&self) -> &'static str {
+        match self {
+            Self::Crc32 => "CRC32",
+            Self::Crc32c => "CRC32C",
+            Self::Crc64Nvme => "CRC64NVME",
+            Self::Sha1 => "SHA1",
+            Self::Sha256 => "SHA256",
+            Self::Sha512 => "SHA512",
+            Self::Xxhash3 => "XXHASH3",
+            Self::Xxhash64 => "XXHASH64",
+            Self::Xxhash128 => "XXHASH128",
+        }
+    }
+
+    /// The `x-amz-checksum-*` HTTP header that carries this algorithm's
+    /// base64-encoded digest.
+    pub fn http_header_name(&self) -> &'static str {
+        match self {
+            Self::Crc32 => http::CRC_32_HEADER_NAME,
+            Self::Crc32c => http::CRC_32_C_HEADER_NAME,
+            Self::Crc64Nvme => http::CRC_64_NVME_HEADER_NAME,
+            Self::Sha1 => http::SHA_1_HEADER_NAME,
+            Self::Sha256 => http::SHA_256_HEADER_NAME,
+            Self::Sha512 => http::SHA_512_HEADER_NAME,
+            Self::Xxhash3 => http::XXHASH_3_HEADER_NAME,
+            Self::Xxhash64 => http::XXHASH_64_HEADER_NAME,
+            Self::Xxhash128 => http::XXHASH_128_HEADER_NAME,
+        }
+    }
+
+    /// Raw (unencoded) digest length in bytes.
+    pub fn raw_len(&self) -> usize {
+        match self {
+            Self::Crc32 | Self::Crc32c => 4,
+            Self::Crc64Nvme => 8,
+            Self::Sha1 => 20,
+            Self::Sha256 => 32,
+            Self::Sha512 => 64,
+            Self::Xxhash3 | Self::Xxhash64 => 8,
+            Self::Xxhash128 => 16,
+        }
+    }
+
+    /// Whether the algorithm supports the S3 COMPOSITE multipart checksum
+    /// type. Per the AWS registry, every algorithm does except CRC64NVME,
+    /// which is FULL_OBJECT-only.
+    pub fn supports_composite(&self) -> bool {
+        match self {
+            Self::Crc64Nvme => false,
+            Self::Crc32
+            | Self::Crc32c
+            | Self::Sha1
+            | Self::Sha256
+            | Self::Sha512
+            | Self::Xxhash3
+            | Self::Xxhash64
+            | Self::Xxhash128 => true,
+        }
+    }
+
+    /// Whether the algorithm supports the S3 FULL_OBJECT checksum type, i.e.
+    /// part digests can be linearly combined into the whole-object digest.
+    /// Only the CRC family has this property; the hash algorithms are
+    /// COMPOSITE-only.
+    pub fn supports_full_object(&self) -> bool {
+        match self {
+            Self::Crc32 | Self::Crc32c | Self::Crc64Nvme => true,
+            Self::Sha1 | Self::Sha256 | Self::Sha512 | Self::Xxhash3 | Self::Xxhash64 | Self::Xxhash128 => false,
+        }
+    }
 }

 pub trait Checksum: Send + Sync {
@@ -731,6 +817,77 @@ mod tests {
        assert_eq!(&raw[..], reference.digest128().to_be_bytes().as_slice());
    }

+    #[test]
+    fn test_algorithm_metadata_is_consistent_for_every_variant() {
+        use crate::Checksum;
+
+        // Cross-checks the per-algorithm metadata methods against the hasher
+        // implementations themselves, so the registry cannot drift from the
+        // code that computes digests (backlog#1844). The list must cover every
+        // variant; the metadata methods use exhaustive matches, so a new
+        // variant that is missing here still fails to compile there first.
+        let all = [
+            ChecksumAlgorithm::Crc32,
+            ChecksumAlgorithm::Crc32c,
+            ChecksumAlgorithm::Crc64Nvme,
+            ChecksumAlgorithm::Sha1,
+            ChecksumAlgorithm::Sha256,
+            ChecksumAlgorithm::Sha512,
+            ChecksumAlgorithm::Xxhash3,
+            ChecksumAlgorithm::Xxhash64,
+            ChecksumAlgorithm::Xxhash128,
+        ];
+
+        for algorithm in all {
+            // Digest length must match what the hasher actually produces.
+            let mut hasher = algorithm.into_impl();
+            hasher.update(b"metadata consistency probe");
+            assert_eq!(
+                algorithm.raw_len(),
+                Checksum::size(&*algorithm.into_impl()) as usize,
+                "{algorithm:?} raw_len() != hasher size()"
+            );
+            assert_eq!(hasher.finalize().len(), algorithm.raw_len(), "{algorithm:?} finalize length != raw_len()");
+
+            // Header name must match the hasher's own header binding.
+            assert_eq!(
+                algorithm.http_header_name(),
+                algorithm.into_impl().header_name(),
+                "{algorithm:?} http_header_name() != HttpChecksum::header_name()"
+            );
+            assert_eq!(
+                algorithm.http_header_name(),
+                format!("x-amz-checksum-{}", algorithm.as_str()),
+                "{algorithm:?} header must be x-amz-checksum-<name>"
+            );
+
+            // The uppercase wire name and the lowercase parse name must be the
+            // same word, and the wire name must parse back to the variant.
+            assert!(
+                algorithm.s3_algorithm_name().eq_ignore_ascii_case(algorithm.as_str()),
+                "{algorithm:?} s3_algorithm_name() and as_str() diverge"
+            );
+            assert_eq!(algorithm.s3_algorithm_name().parse::<ChecksumAlgorithm>().unwrap(), algorithm);
+        }
+
+        // AWS checksum-type support table: CRC64NVME is FULL_OBJECT-only, the
+        // CRC family supports FULL_OBJECT, everything else is COMPOSITE-only.
+        for algorithm in all {
+            let composite = algorithm.supports_composite();
+            let full_object = algorithm.supports_full_object();
+            assert!(composite || full_object, "{algorithm:?} supports no checksum type at all");
+            match algorithm {
+                ChecksumAlgorithm::Crc32 | ChecksumAlgorithm::Crc32c => {
+                    assert!(composite && full_object, "{algorithm:?} must support both checksum types")
+                }
+                ChecksumAlgorithm::Crc64Nvme => {
+                    assert!(!composite && full_object, "CRC64NVME must be FULL_OBJECT-only")
+                }
+                _ => assert!(composite && !full_object, "{algorithm:?} must be COMPOSITE-only"),
+            }
+        }
+    }
+
    #[test]
    fn test_xxhash64_matches_direct_computation_big_endian_seed0() {
        use crate::Xxhash64;
@@ -38,16 +38,9 @@ hotpath.workspace = true
 tokio = { workspace = true, features = ["fs", "rt-multi-thread"] }
 tonic = { workspace = true, features = ["gzip", "deflate"] }
 uuid = { workspace = true, features = ["v4", "fast-rng", "macro-diagnostics"] }
-chrono = { workspace = true, features = ["serde"] }
-jiff = { workspace = true, features = ["serde"] }
 metrics = { workspace = true }
-serde = { workspace = true, features = ["derive"] }
 smallvec = { workspace = true }
-rmp-serde = { workspace = true }
 tracing = { workspace = true }

-[dev-dependencies]
-serde_json = { workspace = true }
-
 [lib]
 doctest = false
@@ -12,7 +12,6 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

-use chrono::{DateTime, Utc};
 use std::collections::HashMap;
 use std::sync::LazyLock;
 use std::sync::atomic::{AtomicU64, Ordering};
@@ -27,8 +26,6 @@ pub static GLOBAL_CONN_MAP: LazyLock<RwLock<HashMap<String, Channel>>> = LazyLoc
 pub static GLOBAL_ROOT_CERT: LazyLock<RwLock<Option<Vec<u8>>>> = LazyLock::new(|| RwLock::new(None));
 pub static GLOBAL_MTLS_IDENTITY: LazyLock<RwLock<Option<MtlsIdentityPem>>> = LazyLock::new(|| RwLock::new(None));
 pub static GLOBAL_OUTBOUND_TLS_GENERATION: LazyLock<AtomicU64> = LazyLock::new(|| AtomicU64::new(0));
-/// Global initialization time of the RustFS node.
-pub static GLOBAL_INIT_TIME: LazyLock<RwLock<Option<DateTime<Utc>>>> = LazyLock::new(|| RwLock::new(None));

 /// Log level to use when reporting cached gRPC connection eviction.
 #[derive(Clone, Copy, Debug, Eq, PartialEq)]
@@ -63,20 +60,6 @@ pub fn try_get_global_local_node_name() -> Option<String> {
        .filter(|name| !name.is_empty())
 }

-/// Set the global RustFS initialization time to the current UTC time.
-pub async fn set_global_init_time_now() {
-    let now = Utc::now();
-    *GLOBAL_INIT_TIME.write().await = Some(now);
-}
-
-/// Get the global RustFS initialization time.
-///
-/// # Returns
-/// * `Option<DateTime<Utc>>` - The initialization time if set.
-pub async fn get_global_init_time() -> Option<DateTime<Utc>> {
-    *GLOBAL_INIT_TIME.read().await
-}
-
 /// Set the global RustFS address used for gRPC connections.
 ///
 /// # Arguments
@@ -14,9 +14,6 @@

 // pub mod error;
 pub mod globals;
-pub mod heal_channel;
-pub mod last_minute;
-pub mod metrics;
 pub mod mrf_channel;
 mod readiness;
 pub mod table_catalog;
@@ -178,6 +178,76 @@ pub trait WorkloadAdmissionSnapshotProvider {
    fn workload_admission_snapshot(&self) -> WorkloadAdmissionRegistrySnapshot;
 }

+/// Foreground workload pressure observed against a configured utilization threshold.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub struct ForegroundPressure {
+    /// Foreground workload class whose utilization reached its threshold.
+    pub class: WorkloadClass,
+    /// Observed utilization percentage for the class.
+    pub usage_pct: usize,
+    /// Configured threshold percentage that the observed utilization reached.
+    pub threshold_pct: usize,
+}
+
+impl ForegroundPressure {
+    /// Return a stable reason label for logs and metrics.
+    pub const fn reason(self) -> &'static str {
+        match self.class {
+            WorkloadClass::ForegroundRead => "foreground_read_pressure",
+            WorkloadClass::ForegroundWrite => "foreground_write_pressure",
+            _ => "foreground_pressure",
+        }
+    }
+}
+
+/// Return the strongest foreground pressure in `snapshot`, if any.
+///
+/// A zero threshold disables its class. `Saturated` counts as full utilization
+/// regardless of the reported limit; otherwise a class contributes only when it
+/// reports a non-zero limit, with a missing active count read as zero. When both
+/// classes are above their threshold the higher utilization wins.
+///
+/// Callers own the enable switch: this function evaluates thresholds only.
+pub fn foreground_pressure(
+    snapshot: &WorkloadAdmissionRegistrySnapshot,
+    read_threshold_pct: usize,
+    write_threshold_pct: usize,
+) -> Option<ForegroundPressure> {
+    [
+        (WorkloadClass::ForegroundRead, read_threshold_pct),
+        (WorkloadClass::ForegroundWrite, write_threshold_pct),
+    ]
+    .into_iter()
+    .filter_map(|(class, threshold_pct)| {
+        if threshold_pct == 0 {
+            return None;
+        }
+
+        let entry = snapshot.get(class)?;
+        let usage_pct = if matches!(entry.state, AdmissionState::Saturated) {
+            100
+        } else {
+            let limit = entry.limit?;
+            if limit == 0 {
+                return None;
+            }
+            entry
+                .active
+                .unwrap_or(0)
+                .saturating_mul(100)
+                .checked_div(limit)
+                .unwrap_or(100)
+        };
+
+        (usage_pct >= threshold_pct).then_some(ForegroundPressure {
+            class,
+            usage_pct,
+            threshold_pct,
+        })
+    })
+    .max_by_key(|pressure| pressure.usage_pct)
+}
+
 #[cfg(test)]
 mod tests {
    use super::*;
@@ -314,4 +384,205 @@ mod tests {

        assert!(err.to_string().contains("unexpected"));
    }
+
+    fn counted(
+        class: WorkloadClass,
+        state: AdmissionState,
+        active: Option<usize>,
+        limit: Option<usize>,
+    ) -> WorkloadAdmissionSnapshot {
+        WorkloadAdmissionSnapshot::new(class, state).with_counts(active, None, limit)
+    }
+
+    fn registry(entries: Vec<WorkloadAdmissionSnapshot>) -> WorkloadAdmissionRegistrySnapshot {
+        WorkloadAdmissionRegistrySnapshot::new(entries)
+    }
+
+    #[test]
+    fn foreground_pressure_reason_labels_cover_non_foreground_classes() {
+        let read = ForegroundPressure {
+            class: WorkloadClass::ForegroundRead,
+            usage_pct: 90,
+            threshold_pct: 80,
+        };
+        let write = ForegroundPressure {
+            class: WorkloadClass::ForegroundWrite,
+            usage_pct: 90,
+            threshold_pct: 80,
+        };
+        let repair = ForegroundPressure {
+            class: WorkloadClass::Repair,
+            usage_pct: 90,
+            threshold_pct: 80,
+        };
+
+        assert_eq!(read.reason(), "foreground_read_pressure");
+        assert_eq!(write.reason(), "foreground_write_pressure");
+        assert_eq!(repair.reason(), "foreground_pressure");
+    }
+
+    #[test]
+    fn foreground_pressure_is_disabled_when_both_thresholds_are_zero() {
+        let snapshot = registry(vec![
+            counted(WorkloadClass::ForegroundRead, AdmissionState::Saturated, Some(8), Some(8)),
+            counted(WorkloadClass::ForegroundWrite, AdmissionState::Saturated, Some(8), Some(8)),
+        ]);
+
+        assert_eq!(foreground_pressure(&snapshot, 0, 0), None);
+    }
+
+    #[test]
+    fn foreground_pressure_skips_only_the_class_whose_threshold_is_zero() {
+        let snapshot = registry(vec![
+            counted(WorkloadClass::ForegroundRead, AdmissionState::Open, Some(10), Some(10)),
+            counted(WorkloadClass::ForegroundWrite, AdmissionState::Open, Some(9), Some(10)),
+        ]);
+
+        assert_eq!(
+            foreground_pressure(&snapshot, 0, 80),
+            Some(ForegroundPressure {
+                class: WorkloadClass::ForegroundWrite,
+                usage_pct: 90,
+                threshold_pct: 80,
+            })
+        );
+        assert_eq!(
+            foreground_pressure(&snapshot, 80, 0),
+            Some(ForegroundPressure {
+                class: WorkloadClass::ForegroundRead,
+                usage_pct: 100,
+                threshold_pct: 80,
+            })
+        );
+    }
+
+    #[test]
+    fn foreground_pressure_ignores_missing_entries() {
+        let snapshot = registry(vec![counted(WorkloadClass::Scanner, AdmissionState::Saturated, Some(8), Some(8))]);
+
+        assert_eq!(foreground_pressure(&snapshot, 1, 1), None);
+    }
+
+    #[test]
+    fn foreground_pressure_ignores_missing_and_zero_limits() {
+        let missing_limit = registry(vec![counted(
+            WorkloadClass::ForegroundRead,
+            AdmissionState::Throttled,
+            Some(8),
+            None,
+        )]);
+        let zero_limit = registry(vec![counted(
+            WorkloadClass::ForegroundWrite,
+            AdmissionState::Throttled,
+            Some(8),
+            Some(0),
+        )]);
+
+        assert_eq!(foreground_pressure(&missing_limit, 1, 1), None);
+        assert_eq!(foreground_pressure(&zero_limit, 1, 1), None);
+    }
+
+    #[test]
+    fn foreground_pressure_treats_saturated_as_full_without_reading_limit() {
+        let snapshot = registry(vec![
+            counted(WorkloadClass::ForegroundRead, AdmissionState::Saturated, None, None),
+            counted(WorkloadClass::ForegroundWrite, AdmissionState::Saturated, Some(0), Some(0)),
+        ]);
+
+        assert_eq!(
+            foreground_pressure(&snapshot, 100, 0),
+            Some(ForegroundPressure {
+                class: WorkloadClass::ForegroundRead,
+                usage_pct: 100,
+                threshold_pct: 100,
+            })
+        );
+        assert_eq!(
+            foreground_pressure(&snapshot, 0, 100),
+            Some(ForegroundPressure {
+                class: WorkloadClass::ForegroundWrite,
+                usage_pct: 100,
+                threshold_pct: 100,
+            })
+        );
+    }
+
+    #[test]
+    fn foreground_pressure_reads_missing_active_as_zero() {
+        let snapshot = registry(vec![counted(WorkloadClass::ForegroundRead, AdmissionState::Open, None, Some(8))]);
+
+        assert_eq!(foreground_pressure(&snapshot, 1, 1), None);
+    }
+
+    #[test]
+    fn foreground_pressure_returns_the_higher_utilization_when_both_classes_exceed() {
+        let read_higher = registry(vec![
+            counted(WorkloadClass::ForegroundRead, AdmissionState::Open, Some(19), Some(20)),
+            counted(WorkloadClass::ForegroundWrite, AdmissionState::Open, Some(17), Some(20)),
+        ]);
+        let write_higher = registry(vec![
+            counted(WorkloadClass::ForegroundRead, AdmissionState::Open, Some(17), Some(20)),
+            counted(WorkloadClass::ForegroundWrite, AdmissionState::Open, Some(19), Some(20)),
+        ]);
+
+        assert_eq!(
+            foreground_pressure(&read_higher, 80, 80),
+            Some(ForegroundPressure {
+                class: WorkloadClass::ForegroundRead,
+                usage_pct: 95,
+                threshold_pct: 80,
+            })
+        );
+        assert_eq!(
+            foreground_pressure(&write_higher, 80, 80),
+            Some(ForegroundPressure {
+                class: WorkloadClass::ForegroundWrite,
+                usage_pct: 95,
+                threshold_pct: 80,
+            })
+        );
+    }
+
+    #[test]
+    fn foreground_pressure_breaks_utilization_ties_toward_the_write_class() {
+        let snapshot = registry(vec![
+            counted(WorkloadClass::ForegroundRead, AdmissionState::Open, Some(18), Some(20)),
+            counted(WorkloadClass::ForegroundWrite, AdmissionState::Open, Some(18), Some(20)),
+        ]);
+
+        assert_eq!(
+            foreground_pressure(&snapshot, 80, 80),
+            Some(ForegroundPressure {
+                class: WorkloadClass::ForegroundWrite,
+                usage_pct: 90,
+                threshold_pct: 80,
+            })
+        );
+    }
+
+    #[test]
+    fn foreground_pressure_triggers_exactly_at_the_threshold_and_not_below() {
+        let at_threshold = registry(vec![counted(
+            WorkloadClass::ForegroundRead,
+            AdmissionState::Open,
+            Some(8),
+            Some(10),
+        )]);
+        let below_threshold = registry(vec![counted(
+            WorkloadClass::ForegroundRead,
+            AdmissionState::Open,
+            Some(7),
+            Some(10),
+        )]);
+
+        assert_eq!(
+            foreground_pressure(&at_threshold, 80, 80),
+            Some(ForegroundPressure {
+                class: WorkloadClass::ForegroundRead,
+                usage_pct: 80,
+                threshold_pct: 80,
+            })
+        );
+        assert_eq!(foreground_pressure(&below_threshold, 80, 80), None);
+    }
 }
@@ -150,9 +150,10 @@ Drive timeout health-action policy:
 Drive timeout profile preset:
 - `RUSTFS_DRIVE_TIMEOUT_PROFILE`
  - `default` (default): keep current timeout defaults.
-  - `high_latency`: use 60s default timeout for scanner-sensitive operations when no per-operation timeout override is set (`read_metadata`, `disk_info`, `list_dir`, `walk_dir`, `walk_dir_stall`).
+  - `high_latency`: use 60s default timeout for scanner-sensitive operations when no operation-specific override is set (`read_metadata`, `disk_info`, `list_dir`, `walk_dir`, `walk_dir_stall`, and object-capacity scan base/maximum budgets).
 - Precedence:
  - Explicit per-operation timeout env (`RUSTFS_DRIVE_*_TIMEOUT_SECS`) takes highest precedence.
+  - Explicit object-capacity timeout env (`RUSTFS_CAPACITY_STAT_TIMEOUT`, `RUSTFS_CAPACITY_MAX_TIMEOUT`) takes precedence for capacity scans.
  - Then `RUSTFS_DRIVE_MAX_TIMEOUT_DURATION` legacy fallback.
  - Then the profile-derived default (`default` or `high_latency`).

@@ -40,6 +40,14 @@ pub const HEALTH_OBJECT_PROGRESS_LOCK_MARGIN_MS: u64 = 5_000;
 pub const ENV_HEALTH_CLUSTER_TIMEOUT_MS: &str = "RUSTFS_HEALTH_CLUSTER_TIMEOUT_MS";
 pub const DEFAULT_HEALTH_CLUSTER_TIMEOUT_MS: u64 = 2000;

+/// Timeout for one remote lock-client online check used by readiness (milliseconds).
+///
+/// This is intentionally shorter than the generic lock RPC timeout so
+/// `/health/ready` can report degradation instead of riding a dead peer's
+/// connect or HTTP/2 keepalive budget.
+pub const ENV_HEALTH_LOCK_ONLINE_TIMEOUT_MS: &str = "RUSTFS_HEALTH_LOCK_ONLINE_TIMEOUT_MS";
+pub const DEFAULT_HEALTH_LOCK_ONLINE_TIMEOUT_MS: u64 = 1000;
+
 /// Maximum time to wait for local node runtime readiness (storage / IAM / lock
 /// quorum) during startup before failing fast (seconds).
 ///
@@ -168,6 +168,35 @@ pub const DEFAULT_DATA_MOVEMENT_PART_CHECKSUMS_FLEET_CONFIRMED: bool = false;
 const _: () = assert!(!DEFAULT_DATA_MOVEMENT_PART_CHECKSUMS_WRITE);
 const _: () = assert!(!DEFAULT_DATA_MOVEMENT_PART_CHECKSUMS_FLEET_CONFIRMED);

+/// Request writing pool metadata version 2.
+///
+/// This remains ineffective until [`ENV_POOL_META_V2_FLEET_CONFIRMED`] is also enabled.
+pub const ENV_POOL_META_V2_WRITE: &str = "RUSTFS_POOL_META_V2_WRITE";
+pub const DEFAULT_POOL_META_V2_WRITE: bool = false;
+
+/// Operator-attested confirmation that every pool metadata reader and writer understands version 2.
+pub const ENV_POOL_META_V2_FLEET_CONFIRMED: &str = "RUSTFS_POOL_META_V2_FLEET_CONFIRMED";
+pub const DEFAULT_POOL_META_V2_FLEET_CONFIRMED: bool = false;
+
+const _: () = assert!(!DEFAULT_POOL_META_V2_WRITE);
+const _: () = assert!(!DEFAULT_POOL_META_V2_FLEET_CONFIRMED);
+
+/// Request writing pool metadata version 3 with durable generations.
+///
+/// Existing deployments remain on their observed version until
+/// [`ENV_POOL_META_V3_FLEET_CONFIRMED`] is also enabled. Fresh deployments may
+/// initialize directly at version 3 because they have no legacy readers.
+pub const ENV_POOL_META_V3_WRITE: &str = "RUSTFS_POOL_META_V3_WRITE";
+pub const DEFAULT_POOL_META_V3_WRITE: bool = false;
+
+/// Operator-attested confirmation that every pool metadata reader and writer
+/// understands the version 3 generation and recovery protocol.
+pub const ENV_POOL_META_V3_FLEET_CONFIRMED: &str = "RUSTFS_POOL_META_V3_FLEET_CONFIRMED";
+pub const DEFAULT_POOL_META_V3_FLEET_CONFIRMED: bool = false;
+
+const _: () = assert!(!DEFAULT_POOL_META_V3_WRITE);
+const _: () = assert!(!DEFAULT_POOL_META_V3_FLEET_CONFIRMED);
+
 // =============================================================================
 // Concurrent Request Fix - Timeout and Backpressure Configuration
 // =============================================================================
@@ -259,6 +288,49 @@ pub const DEFAULT_PUT_FOREGROUND_ADMISSION_WAIT_TIMEOUT_MS: u64 = 0;

 const _: () = assert!(!DEFAULT_PUT_FOREGROUND_ADMISSION_ENABLE);

+/// Enable automatic foreground admission for large or unknown-size PutObject requests.
+///
+/// Unlike the strict experimental gate above, this default-on path only applies
+/// to requests that are large enough to create sustained erasure/RPC pressure.
+/// Small PUTs continue on the legacy path unless the strict gate is explicitly
+/// enabled.
+pub const ENV_PUT_LARGE_FOREGROUND_ADMISSION_ENABLE: &str = "RUSTFS_PUT_LARGE_FOREGROUND_ADMISSION_ENABLE";
+pub const DEFAULT_PUT_LARGE_FOREGROUND_ADMISSION_ENABLE: bool = true;
+
+/// Maximum automatic foreground write requests admitted concurrently per process.
+///
+/// `0` derives a conservative default from the local disk-read scheduler cap,
+/// currently clamped to protect the commit path without making ordinary high
+/// throughput uploads single-file.
+pub const ENV_PUT_LARGE_FOREGROUND_ADMISSION_LIMIT: &str = "RUSTFS_PUT_LARGE_FOREGROUND_ADMISSION_LIMIT";
+pub const DEFAULT_PUT_LARGE_FOREGROUND_ADMISSION_LIMIT: usize = 0;
+
+/// Minimum direct PutObject size that enters automatic foreground write admission.
+///
+/// Requests with an unknown size are treated as large because the write pressure
+/// cannot be bounded from headers.
+pub const ENV_PUT_LARGE_FOREGROUND_ADMISSION_MIN_SIZE_BYTES: &str = "RUSTFS_PUT_LARGE_FOREGROUND_ADMISSION_MIN_SIZE_BYTES";
+pub const DEFAULT_PUT_LARGE_FOREGROUND_ADMISSION_MIN_SIZE_BYTES: usize = 32 * 1024 * 1024;
+
+/// Minimum UploadPart size that enters automatic foreground write admission.
+///
+/// Multipart pressure is often many moderate-sized parts rather than one very
+/// large request. The default gates every multipart part through the same permit
+/// pool as large/unknown-size PutObject while keeping small direct PUTs on the
+/// legacy path.
+pub const ENV_PUT_MULTIPART_FOREGROUND_ADMISSION_MIN_SIZE_BYTES: &str =
+    "RUSTFS_PUT_MULTIPART_FOREGROUND_ADMISSION_MIN_SIZE_BYTES";
+pub const DEFAULT_PUT_MULTIPART_FOREGROUND_ADMISSION_MIN_SIZE_BYTES: usize = 0;
+
+/// Time in milliseconds an automatic foreground write waits for a permit.
+///
+/// A short wait smooths transient bursts while still returning S3
+/// `SlowDown`/503 before body ingest when the node is already saturated.
+pub const ENV_PUT_LARGE_FOREGROUND_ADMISSION_WAIT_TIMEOUT_MS: &str = "RUSTFS_PUT_LARGE_FOREGROUND_ADMISSION_WAIT_TIMEOUT_MS";
+pub const DEFAULT_PUT_LARGE_FOREGROUND_ADMISSION_WAIT_TIMEOUT_MS: u64 = 250;
+
+const _: () = assert!(DEFAULT_PUT_LARGE_FOREGROUND_ADMISSION_ENABLE);
+
 /// Environment variable for minimum GetObject timeout in seconds.
 ///
 /// When dynamic timeout calculation is enabled, this is the minimum timeout
@@ -736,4 +808,16 @@ mod remote_version_state_tests {
            "RUSTFS_OBJECT_TRANSACTION_FENCING_FLEET_CONFIRMED"
        );
    }
+
+    #[test]
+    fn pool_meta_v2_gate_uses_stable_environment_names() {
+        assert_eq!(super::ENV_POOL_META_V2_WRITE, "RUSTFS_POOL_META_V2_WRITE");
+        assert_eq!(super::ENV_POOL_META_V2_FLEET_CONFIRMED, "RUSTFS_POOL_META_V2_FLEET_CONFIRMED");
+    }
+
+    #[test]
+    fn pool_meta_v3_gate_uses_stable_environment_names() {
+        assert_eq!(super::ENV_POOL_META_V3_WRITE, "RUSTFS_POOL_META_V3_WRITE");
+        assert_eq!(super::ENV_POOL_META_V3_FLEET_CONFIRMED, "RUSTFS_POOL_META_V3_FLEET_CONFIRMED");
+    }
 }
@@ -103,7 +103,7 @@ pub const ENV_ALLOCATOR_RECLAIM_ENABLED: &str = "RUSTFS_ALLOCATOR_RECLAIM_ENABLE
 pub const ENV_ALLOCATOR_RECLAIM_INTERVAL_SECS: &str = "RUSTFS_ALLOCATOR_RECLAIM_INTERVAL_SECS";
 pub const ENV_ALLOCATOR_RECLAIM_FORCE: &str = "RUSTFS_ALLOCATOR_RECLAIM_FORCE";
 pub const ENV_ALLOCATOR_RECLAIM_IDLE_INTERVALS: &str = "RUSTFS_ALLOCATOR_RECLAIM_IDLE_INTERVALS";
-pub const DEFAULT_ALLOCATOR_RECLAIM_ENABLED: bool = false;
+pub const DEFAULT_ALLOCATOR_RECLAIM_ENABLED: bool = true;
 pub const DEFAULT_ALLOCATOR_RECLAIM_INTERVAL_SECS: u64 = 30;
 pub const DEFAULT_ALLOCATOR_RECLAIM_FORCE: bool = true;
 pub const DEFAULT_ALLOCATOR_RECLAIM_IDLE_INTERVALS: u64 = 3;
@@ -573,6 +573,10 @@ pub struct DataUsageInfo {
    /// cycle (or retained a compatible last-known-good cache).
    #[serde(default)]
    pub usage_snapshot_partial: bool,
+    /// Durable marker for a first-start bootstrap that has not produced an
+    /// authoritative usage snapshot yet.
+    #[serde(default, skip_serializing_if = "std::ops::Not::not")]
+    pub usage_snapshot_bootstrap_pending: bool,
    /// Deprecated kept here for backward compatibility reasons
    pub bucket_sizes: HashMap<String, u64>,
    /// Per-disk snapshot information when available
@@ -1962,7 +1966,8 @@ impl DataUsageInfo {

    /// Whether this snapshot authoritatively covers every reported bucket.
    pub fn is_complete_bucket_usage_snapshot(&self) -> bool {
-        self.usage_snapshot_complete
+        !self.usage_snapshot_bootstrap_pending
+            && self.usage_snapshot_complete
            && self.last_update.is_some()
            && u64::try_from(self.buckets_usage.len()).ok() == Some(self.buckets_count)
    }
@@ -1971,7 +1976,8 @@ impl DataUsageInfo {
    /// admin display. Partial data is accepted only with unique set states,
    /// a plan digest for every state, and at least one usable generation.
    pub fn is_valid_partial_snapshot(&self) -> bool {
-        if !self.usage_snapshot_partial
+        if self.usage_snapshot_bootstrap_pending
+            || !self.usage_snapshot_partial
            || self.usage_snapshot_converged != Some(false)
            || self.last_update.is_none()
            || self.scanner_cycle.is_none()
@@ -109,15 +109,22 @@ hyper = { workspace = true, features = ["http2", "http1", "server"] }
 hyper-util = { workspace = true, features = ["tokio", "server-auto", "server-graceful", "tracing"] }
 reqwest = { workspace = true, features = ["json", "multipart", "stream"] }
 rustfs-signer.workspace = true
+
+# The MFA e2e test computes RFC 6238 codes itself rather than calling the
+# server's implementation: a shared helper could agree with a bug on both sides.
+data-encoding = { workspace = true }
+hmac = { workspace = true }
+sha1 = { workspace = true }
+serde_urlencoded = { workspace = true }
 tracing = { workspace = true }
 tracing-subscriber = { workspace = true, features = ["env-filter", "time"] }
 uuid = { workspace = true, features = ["v4", "fast-rng", "macro-diagnostics"] }
 urlencoding.workspace = true
 walkdir.workspace = true
-base64 = { workspace = true }
+base64-simd = { workspace = true }
 rand = { workspace = true, features = ["serde"] }
 chrono = { workspace = true, features = ["serde"] }
-hex = { workspace = true }
+hex-simd = { workspace = true }
 md-5 = { workspace = true }
 opentelemetry-proto = { workspace = true }
 prost.workspace = true
@@ -27,6 +27,7 @@ Registered in [`src/lib.rs`](src/lib.rs). Grouped by concern:
 | **reliant** | [`src/reliant/`](src/reliant) | Tests that reuse an **externally started** server (SQL/select, conditional writes, lifecycle, deleted-object reads, node-interact). Run via [`scripts/run_e2e_tests.sh`](../../scripts/run_e2e_tests.sh); see [`src/reliant/README.md`](src/reliant/README.md) |
 | **cluster** | `cluster_concurrency_test`, `stale_multipart_cleanup_cluster_test`, `namespace_lock_quorum_test`, `admin_timeout_regression_test`, `object_lambda_test`, `replication_extension_test` | Multi-node scenarios via `RustFSTestClusterEnvironment` |
 | **chaos / reliability** | [`src/chaos.rs`](src/chaos.rs), `reliability_disk_fault_test`, `heal_erasure_disk_rebuild_test`, `server_startup_failfast_test` | Disk offline/replace/corrupt, EC rebuild, heal, fail-fast startup |
+| **upgrade compatibility** | `upgrade_compatibility_test` | Pinned previous-release writes followed by current-build reads on the same data directory |

 ## How to run

@@ -72,7 +73,7 @@ The reason string on each attribute is the classifier. Current classes:

 - **Needs a pre-started server** — `"requires running RustFS server at
  localhost:9000"` / `"Connects to existing rustfs server"`. These are the
-  `reliant/*` and `policy/test_runner` tests; start a server first (e.g.
+  `reliant/*` tests; start a server first (e.g.
  [`scripts/run_e2e_tests.sh`](../../scripts/run_e2e_tests.sh)) or use
  `--run-ignored`.
 - **Heavy / external tool** — `"Starts a rustfs server; enable when running
@@ -168,6 +169,7 @@ the same profile for membership and execution with one nightly worker.
 | `s3s-e2e` black-box | `e2e-tests` + `e2e-tests-rio-v2` jobs | **Active** (external conformance tool) |
 | ILM / lifecycle (ignored) | `test-ilm-integration-serial` lane, `-j1` | **Active** (backlog#1148 ilm-1) |
 | KMS suite | `e2e-full` job, merge queue + main | **Active** |
+| Direct upgrade from pinned previous release | `e2e-upgrade.yml`, storage-sensitive PRs + release tags + weekly | **Active** |
 | Cluster faults (`e2e-nightly` profile) | consolidated nightly workflow | **Active** (backlog#1149 ci-7) |
 | Protocols (FTPS/WebDAV/SFTP) | consolidated nightly workflow, serial | **Active** (backlog#1149 ci-7) |
 | Replication (fast subset) | `e2e-smoke` profile, `e2e-tests` job, every PR | **Active** (backlog#1147 repl-1) |
@@ -31,13 +31,9 @@

 #[cfg(test)]
 mod tests {
-    use crate::common::{RustFSTestEnvironment, init_logging, local_http_client, rustfs_binary_path};
-    use aws_sdk_s3::config::{Credentials, Region};
-    use aws_sdk_s3::{Client, Config};
-    use http::header::HOST;
-    use rustfs_signer::constants::UNSIGNED_PAYLOAD;
-    use rustfs_signer::sign_v4;
-    use s3s::Body;
+    use crate::common::{RustFSTestEnvironment, init_logging, rustfs_binary_path};
+    use aws_sdk_s3::Client;
+    use aws_sdk_s3::error::ProvideErrorMetadata;
    use std::error::Error;
    use std::io::Read;
    use std::process::{Command, Stdio};
@@ -86,10 +82,10 @@ mod tests {
    }

    /// Send a SigV4-signed request to `path` (optionally with a JSON `body`) and
-    /// return `(status, body)`. Uses the `UNSIGNED_PAYLOAD` content hash so a
-    /// request body can be attached without the caller pre-hashing it — the
-    /// server verifies the signature against the same sentinel, exactly as the
-    /// AWS SDKs / MinIO client do for streaming/unsigned payloads.
+    /// return `(status, body)`.
+    ///
+    /// Thin wrapper over [`crate::common::admin_request`], kept local so the
+    /// call sites below keep their `Option<&str>` body shape.
    async fn signed_request(
        base_url: &str,
        method: http::Method,
@@ -98,47 +94,13 @@ mod tests {
        access_key: &str,
        secret_key: &str,
    ) -> Result<(reqwest::StatusCode, String), Box<dyn Error + Send + Sync>> {
-        let url = format!("{base_url}{path}");
-        let uri = url.parse::<http::Uri>()?;
-        let authority = uri.authority().ok_or("missing authority")?.to_string();
-        let body_bytes = body.map(|b| b.as_bytes().to_vec()).unwrap_or_default();
-
-        // The signature is computed over `UNSIGNED_PAYLOAD`, so the body bytes do
-        // not participate in the SigV4 hash — sign over an empty body and attach
-        // the real payload to the wire request below.
-        let request = http::Request::builder()
-            .method(method.clone())
-            .uri(uri)
-            .header(HOST, authority)
-            .header("x-amz-content-sha256", UNSIGNED_PAYLOAD);
-        let signed = sign_v4(request.body(Body::empty())?, 0, access_key, secret_key, "", "us-east-1");
-
-        let client = local_http_client();
-        let mut rb = client.request(method, url.as_str());
-        for (name, value) in signed.headers() {
-            rb = rb.header(name, value);
-        }
-        if !body_bytes.is_empty() {
-            rb = rb.body(body_bytes);
-        }
-        let resp = rb.send().await?;
-        let status = resp.status();
-        let text = resp.text().await?;
-        Ok((status, text))
+        crate::common::admin_request(base_url, method, path, body.map(str::to_string), access_key, secret_key).await
    }

    /// Build an S3 client bound to explicit credentials (used to exercise the S3
    /// data plane with rotated / stale root credentials).
    fn s3_client_with(env: &RustFSTestEnvironment, access_key: &str, secret_key: &str) -> Client {
-        let credentials = Credentials::new(access_key, secret_key, None, None, "sec4-admin-auth");
-        let config = Config::builder()
-            .credentials_provider(credentials)
-            .region(Region::new("us-east-1"))
-            .endpoint_url(&env.url)
-            .force_path_style(true)
-            .behavior_version_latest()
-            .build();
-        Client::from_conf(config)
+        env.create_s3_client_with_credentials(access_key, secret_key)
    }

    /// Create a non-admin IAM user via the admin `add-user` API using the root
@@ -150,12 +112,7 @@ mod tests {
        access_key: &str,
        secret_key: &str,
    ) -> Result<(), Box<dyn Error + Send + Sync>> {
-        let path = format!("/rustfs/admin/v3/add-user?accessKey={access_key}");
-        let body = serde_json::json!({ "secretKey": secret_key, "status": "enabled" }).to_string();
-        let (status, resp) =
-            signed_request(&env.url, http::Method::PUT, &path, Some(&body), &env.access_key, &env.secret_key).await?;
-        assert!(status.is_success(), "add-user should succeed (status={status}, body={resp})");
-        Ok(())
+        crate::common::admin_create_user(env, access_key, secret_key).await
    }

    /// A fully authenticated but non-admin credential must be rejected with
@@ -368,10 +325,15 @@ mod tests {
            reqwest::StatusCode::FORBIDDEN,
            "stale root must be rejected on the admin API after rotation, body: {body}"
        );
-        let s3_old = s3_client_with(&env, &old_ak, &old_sk).list_buckets().send().await;
-        assert!(
-            s3_old.is_err(),
-            "stale root must be rejected on the S3 plane after rotation, got: {s3_old:?}"
+        let s3_old = s3_client_with(&env, &old_ak, &old_sk)
+            .list_buckets()
+            .send()
+            .await
+            .expect_err("stale root must be rejected on the S3 plane after rotation");
+        assert_eq!(
+            s3_old.as_service_error().and_then(ProvideErrorMetadata::code),
+            Some("InvalidAccessKeyId"),
+            "stale root must receive InvalidAccessKeyId after rotation: {s3_old:?}"
        );

        env.stop_server();
@@ -30,6 +30,7 @@ use crate::common::{
    RustFSTestEnvironment, admin_ok, admin_request, admin_request_with_session_token, build_test_sts_client, init_logging,
 };
 use aws_sdk_s3::config::{Credentials, Region};
+use aws_sdk_s3::error::ProvideErrorMetadata;
 use aws_sdk_s3::primitives::ByteStream;
 use aws_sdk_s3::{Client, Config};
 use reqwest::StatusCode;
@@ -411,8 +412,13 @@ async fn test_admin_user_policy_service_account_crud_lifecycle() -> TestResult {
        .key("before-attach")
        .body(ByteStream::from_static(b"x"))
        .send()
-        .await;
-    assert!(denied.is_err(), "user without a policy must not be able to write to {bucket}");
+        .await
+        .expect_err("user without a policy must not be able to write to the bucket");
+    assert_eq!(
+        denied.as_service_error().and_then(ProvideErrorMetadata::code),
+        Some("AccessDenied"),
+        "user without a policy must receive AccessDenied: {denied:?}"
+    );

    // --- attach policy: the credential actually gains S3 access -----------------
    admin_ok(
@@ -499,13 +505,19 @@ async fn test_admin_user_policy_service_account_crud_lifecycle() -> TestResult {
            .body(ByteStream::from_static(b"x"))
            .send()
            .await;
-        if revoked.is_err() {
-            break;
+        match revoked {
+            Ok(_) if tokio::time::Instant::now() >= deadline => {
+                return Err("deleted service account credential still works".into());
+            }
+            Ok(_) => sleep(Duration::from_millis(500)).await,
+            Err(error) => {
+                let code = error.as_service_error().and_then(ProvideErrorMetadata::code);
+                if matches!(code, Some("AccessDenied" | "InvalidAccessKeyId")) {
+                    break;
+                }
+                return Err(format!("deleted service account must fail with an authorization error, got {error:?}").into());
+            }
        }
-        if tokio::time::Instant::now() >= deadline {
-            return Err("deleted service account credential still works".into());
-        }
-        sleep(Duration::from_millis(500)).await;
    }

    // Disable then remove the user; the credential must stop working.
@@ -525,13 +537,19 @@ async fn test_admin_user_policy_service_account_crud_lifecycle() -> TestResult {
            .body(ByteStream::from_static(b"x"))
            .send()
            .await;
-        if disabled.is_err() {
-            break;
+        match disabled {
+            Ok(_) if tokio::time::Instant::now() >= deadline => {
+                return Err("disabled user credential still works".into());
+            }
+            Ok(_) => sleep(Duration::from_millis(500)).await,
+            Err(error) => {
+                let code = error.as_service_error().and_then(ProvideErrorMetadata::code);
+                if matches!(code, Some("AccessDenied" | "InvalidAccessKeyId")) {
+                    break;
+                }
+                return Err(format!("disabled user must fail with an authorization error, got {error:?}").into());
+            }
        }
-        if tokio::time::Instant::now() >= deadline {
-            return Err("disabled user credential still works".into());
-        }
-        sleep(Duration::from_millis(500)).await;
    }

    admin_ok(
@@ -0,0 +1,426 @@
+// Copyright 2024 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+//! End-to-end coverage for the self-service account and two-factor surface.
+//!
+//! The unit tests cover the state machine at its edges; what only an end-to-end
+//! test can prove is that the pieces are wired together and that the *existing*
+//! authentication paths still behave. Specifically:
+//!
+//! 1. Enrollment is refused when `RUSTFS_IAM_MASTER_KEY` is absent, so a TOTP
+//!    secret is never written where an attacker could read it off a disk.
+//! 2. With a master key, the full flow works: enroll, activate with a real
+//!    RFC 6238 code, and receive single-use recovery codes.
+//! 3. Once a factor is enrolled, `AssumeRole` refuses to mint a session without
+//!    one, and accepts a valid code — the actual login gate.
+//! 4. A direct SigV4 admin request keeps working with a factor enrolled. This is
+//!    the regression that matters most: gating it would break every script and
+//!    CLI the moment somebody enabled 2FA.
+//! 5. `AssumeRole` for an identity with no enrollment is byte-for-byte the old
+//!    behaviour, so existing deployments are untouched.
+//! 6. Rotating a password through `/account/password` invalidates the sessions
+//!    minted under the old secret.
+
+#[cfg(test)]
+mod tests {
+    use crate::common::{RustFSTestEnvironment, init_logging, local_http_client};
+    use hmac::{Hmac, KeyInit as _, Mac};
+    use http::header::HOST;
+    use rustfs_signer::constants::UNSIGNED_PAYLOAD;
+    use rustfs_signer::sign_v4;
+    use s3s::Body;
+    use sha1::Sha1;
+    use std::error::Error;
+    use std::time::{SystemTime, UNIX_EPOCH};
+
+    const ACCOUNT_INFO_PATH: &str = "/rustfs/admin/v3/account/info";
+    const ACCOUNT_PASSWORD_PATH: &str = "/rustfs/admin/v3/account/password";
+    const ACCOUNT_MFA_PATH: &str = "/rustfs/admin/v3/account/mfa";
+    const ACCOUNT_MFA_ENROLL_PATH: &str = "/rustfs/admin/v3/account/mfa/enroll";
+    const ACCOUNT_MFA_ACTIVATE_PATH: &str = "/rustfs/admin/v3/account/mfa/activate";
+    const MFA_CHALLENGE_PATH: &str = "/rustfs/admin/v3/mfa/challenge";
+    const ADMIN_INFO_PATH: &str = "/rustfs/admin/v3/info";
+
+    /// A master key so the server will accept an enrollment. Test-only value.
+    const TEST_MASTER_KEY: &str = "e2e-mfa-master-key-do-not-reuse";
+
+    type HmacSha1 = Hmac<Sha1>;
+
+    /// One signed admin request, returning the status and the raw body.
+    ///
+    /// Thin wrapper over [`crate::common::admin_request`], kept local so the
+    /// call sites below keep their `Option<&str>` body shape.
+    async fn signed_request(
+        base_url: &str,
+        method: http::Method,
+        path: &str,
+        body: Option<&str>,
+        access_key: &str,
+        secret_key: &str,
+    ) -> Result<(reqwest::StatusCode, String), Box<dyn Error + Send + Sync>> {
+        crate::common::admin_request(base_url, method, path, body.map(str::to_string), access_key, secret_key).await
+    }
+
+    /// A SigV4-signed `AssumeRole` form POST, optionally carrying a second factor.
+    ///
+    /// Uses STS's own `SerialNumber`/`TokenCode` fields, which is the point: a
+    /// script or SDK can present the factor without a RustFS-specific protocol.
+    async fn assume_role(
+        base_url: &str,
+        access_key: &str,
+        secret_key: &str,
+        second_factor: Option<(&str, &str)>,
+    ) -> Result<(reqwest::StatusCode, String), Box<dyn Error + Send + Sync>> {
+        let mut form = vec![
+            ("Action", "AssumeRole".to_string()),
+            ("Version", "2011-06-15".to_string()),
+            ("RoleArn", "arn:aws:iam::*:role/Admin".to_string()),
+            ("RoleSessionName", "e2e".to_string()),
+            ("DurationSeconds", "3600".to_string()),
+        ];
+        if let Some((challenge, code)) = second_factor {
+            form.push(("SerialNumber", challenge.to_string()));
+            form.push(("TokenCode", code.to_string()));
+        }
+        let body = serde_urlencoded::to_string(&form)?;
+
+        let uri = base_url.parse::<http::Uri>()?;
+        let authority = uri.authority().ok_or("missing authority")?.to_string();
+        let request = http::Request::builder()
+            .method(http::Method::POST)
+            .uri(format!("{base_url}/"))
+            .header(HOST, authority)
+            .header("content-type", "application/x-www-form-urlencoded")
+            .header("x-amz-content-sha256", UNSIGNED_PAYLOAD);
+        let signed = sign_v4(request.body(Body::empty())?, 0, access_key, secret_key, "", "us-east-1");
+
+        let client = local_http_client();
+        let mut builder = client.request(http::Method::POST, format!("{base_url}/"));
+        for (name, value) in signed.headers() {
+            builder = builder.header(name, value);
+        }
+        let response = builder.body(body).send().await?;
+        let status = response.status();
+        let text = response.text().await?;
+        Ok((status, text))
+    }
+
+    /// Generate the current RFC 6238 code for a base32 secret.
+    ///
+    /// Computed independently of the server implementation: a shared helper
+    /// could agree with a bug on both sides.
+    fn totp_now(secret_base32: &str) -> String {
+        let secret = data_encoding::BASE32_NOPAD
+            .decode(secret_base32.as_bytes())
+            .expect("server must return unpadded base32");
+        let step = SystemTime::now()
+            .duration_since(UNIX_EPOCH)
+            .expect("clock after the epoch")
+            .as_secs()
+            / 30;
+
+        let mut mac = HmacSha1::new_from_slice(&secret).expect("HMAC accepts any key length");
+        mac.update(&step.to_be_bytes());
+        let digest = mac.finalize().into_bytes();
+
+        let offset = (digest[digest.len() - 1] & 0x0f) as usize;
+        let binary = u32::from_be_bytes([
+            digest[offset] & 0x7f,
+            digest[offset + 1],
+            digest[offset + 2],
+            digest[offset + 3],
+        ]);
+        format!("{:06}", binary % 1_000_000)
+    }
+
+    fn json(body: &str) -> serde_json::Value {
+        serde_json::from_str(body).unwrap_or_else(|error| panic!("expected JSON, got {body}: {error}"))
+    }
+
+    #[tokio::test]
+    async fn enrollment_is_refused_without_at_rest_protection() -> Result<(), Box<dyn Error + Send + Sync>> {
+        init_logging();
+        let mut env = RustFSTestEnvironment::new().await?;
+        // Deliberately no RUSTFS_IAM_MASTER_KEY.
+        env.start_rustfs_server(vec![]).await?;
+
+        let (access_key, secret_key) = (env.access_key.clone(), env.secret_key.clone());
+
+        // The account surface itself works.
+        let (status, body) =
+            signed_request(&env.url, http::Method::GET, ACCOUNT_INFO_PATH, None, &access_key, &secret_key).await?;
+        assert_eq!(status, reqwest::StatusCode::OK, "account info must be reachable, body: {body}");
+        let info = json(&body);
+        assert_eq!(info["access_key"], access_key.as_str());
+        assert_eq!(info["identity_type"], "root");
+        assert_eq!(info["credentials_source"], "env");
+        // Root credentials come from a process-wide OnceLock that also derives
+        // the internode RPC secret, so they are immutable at runtime.
+        assert_eq!(info["mutable"]["password"], false);
+
+        // Status reports the refusal rather than pretending enrollment is possible.
+        let (status, body) =
+            signed_request(&env.url, http::Method::GET, ACCOUNT_MFA_PATH, None, &access_key, &secret_key).await?;
+        assert_eq!(status, reqwest::StatusCode::OK, "mfa status must be reachable, body: {body}");
+        let mfa = json(&body);
+        assert_eq!(mfa["enabled"], false);
+        assert_eq!(mfa["enrollment_available"], false);
+        assert!(
+            mfa["enrollment_blocked_reason"]
+                .as_str()
+                .is_some_and(|reason| reason.contains("RUSTFS_IAM_MASTER_KEY")),
+            "the refusal must name the variable an operator has to set, body: {body}"
+        );
+
+        // And enrolling actually fails, rather than writing a plaintext secret.
+        let (status, body) = signed_request(
+            &env.url,
+            http::Method::POST,
+            ACCOUNT_MFA_ENROLL_PATH,
+            Some("{}"),
+            &access_key,
+            &secret_key,
+        )
+        .await?;
+        assert!(
+            status.is_client_error() || status.is_server_error(),
+            "enrollment must fail without a master key, status: {status}, body: {body}"
+        );
+        assert!(
+            body.contains("RUSTFS_IAM_MASTER_KEY"),
+            "the failure must explain the remedy, body: {body}"
+        );
+
+        env.stop_server();
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn assume_role_is_unchanged_for_an_identity_with_no_second_factor() -> Result<(), Box<dyn Error + Send + Sync>> {
+        // The regression that protects every existing deployment: an identity
+        // with no enrollment must take no new code path.
+        init_logging();
+        let mut env = RustFSTestEnvironment::new().await?;
+        env.start_rustfs_server_with_env(vec![], &[("RUSTFS_IAM_MASTER_KEY", TEST_MASTER_KEY)])
+            .await?;
+
+        let (access_key, secret_key) = (env.access_key.clone(), env.secret_key.clone());
+
+        let (status, body) =
+            signed_request(&env.url, http::Method::GET, MFA_CHALLENGE_PATH, None, &access_key, &secret_key).await?;
+        assert_eq!(status, reqwest::StatusCode::OK, "challenge must be reachable, body: {body}");
+        let challenge = json(&body);
+        assert_eq!(challenge["required"], false, "no enrollment means no challenge");
+        assert!(challenge["challenge"].is_null());
+
+        let (status, body) = assume_role(&env.url, &access_key, &secret_key, None).await?;
+        assert_eq!(status, reqwest::StatusCode::OK, "AssumeRole must still work, body: {body}");
+        assert!(body.contains("<AccessKeyId>"), "expected STS credentials, body: {body}");
+
+        env.stop_server();
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn the_full_second_factor_lifecycle_gates_only_session_minting() -> Result<(), Box<dyn Error + Send + Sync>> {
+        init_logging();
+        let mut env = RustFSTestEnvironment::new().await?;
+        env.start_rustfs_server_with_env(vec![], &[("RUSTFS_IAM_MASTER_KEY", TEST_MASTER_KEY)])
+            .await?;
+
+        let (access_key, secret_key) = (env.access_key.clone(), env.secret_key.clone());
+
+        // --- Enroll ---
+        let (status, body) = signed_request(
+            &env.url,
+            http::Method::POST,
+            ACCOUNT_MFA_ENROLL_PATH,
+            Some("{}"),
+            &access_key,
+            &secret_key,
+        )
+        .await?;
+        assert_eq!(status, reqwest::StatusCode::OK, "enrollment must succeed, body: {body}");
+        let enrollment = json(&body);
+        let secret_base32 = enrollment["secret_base32"].as_str().expect("secret").to_string();
+        assert!(
+            enrollment["otpauth_uri"]
+                .as_str()
+                .is_some_and(|uri| uri.starts_with("otpauth://totp/RustFS:")),
+            "body: {body}"
+        );
+        assert!(!enrollment["qr_svg"].as_str().unwrap_or_default().is_empty(), "expected an SVG");
+        assert!(!enrollment["qr_utf8"].as_str().unwrap_or_default().is_empty(), "expected block art");
+
+        // A pending enrollment must not gate anything yet: a mis-scanned QR
+        // cannot be allowed to lock the operator out.
+        let (status, body) =
+            signed_request(&env.url, http::Method::GET, MFA_CHALLENGE_PATH, None, &access_key, &secret_key).await?;
+        assert_eq!(status, reqwest::StatusCode::OK);
+        assert_eq!(json(&body)["required"], false, "a pending enrollment must not gate login");
+
+        // --- Activate ---
+        let code = totp_now(&secret_base32);
+        let (status, body) = signed_request(
+            &env.url,
+            http::Method::POST,
+            ACCOUNT_MFA_ACTIVATE_PATH,
+            Some(&format!(r#"{{"code":"{code}"}}"#)),
+            &access_key,
+            &secret_key,
+        )
+        .await?;
+        assert_eq!(status, reqwest::StatusCode::OK, "activation must succeed, body: {body}");
+        let activated = json(&body);
+        let recovery_codes = activated["recovery_codes"].as_array().expect("recovery codes").clone();
+        assert_eq!(recovery_codes.len(), 10, "expected a full recovery set, body: {body}");
+
+        // --- The gate is now on for session minting ---
+        let (status, body) =
+            signed_request(&env.url, http::Method::GET, MFA_CHALLENGE_PATH, None, &access_key, &secret_key).await?;
+        assert_eq!(status, reqwest::StatusCode::OK);
+        let challenge_body = json(&body);
+        assert_eq!(challenge_body["required"], true, "body: {body}");
+        let challenge = challenge_body["challenge"].as_str().expect("challenge").to_string();
+
+        let (status, body) = assume_role(&env.url, &access_key, &secret_key, None).await?;
+        assert!(status.is_client_error(), "AssumeRole must refuse without a factor, body: {body}");
+        assert!(
+            body.contains("MultiFactorAuthRequired"),
+            "clients match on this code to prompt instead of reporting a failed login, body: {body}"
+        );
+
+        // --- ... but direct SigV4 access is untouched ---
+        let (status, body) = signed_request(&env.url, http::Method::GET, ADMIN_INFO_PATH, None, &access_key, &secret_key).await?;
+        assert_eq!(
+            status,
+            reqwest::StatusCode::OK,
+            "a direct admin request must keep working with a factor enrolled, body: {body}"
+        );
+
+        // --- A valid factor mints the session ---
+        // A fresh code: activation consumed the previous time step, so reusing
+        // that code would be refused as a replay.
+        let code = wait_for_a_fresh_code(&secret_base32).await;
+        let (status, body) = assume_role(&env.url, &access_key, &secret_key, Some((&challenge, &code))).await?;
+        assert_eq!(status, reqwest::StatusCode::OK, "a valid factor must mint a session, body: {body}");
+        assert!(body.contains("<AccessKeyId>"), "expected STS credentials, body: {body}");
+
+        // --- A recovery code also works, once ---
+        let recovery_code = recovery_codes[0].as_str().expect("recovery code").to_string();
+        let (status, body) = assume_role(&env.url, &access_key, &secret_key, Some((&challenge, &recovery_code))).await?;
+        assert_eq!(status, reqwest::StatusCode::OK, "a recovery code must mint a session, body: {body}");
+
+        let (status, body) = assume_role(&env.url, &access_key, &secret_key, Some((&challenge, &recovery_code))).await?;
+        assert!(
+            status.is_client_error(),
+            "a spent recovery code must not work twice, status: {status}, body: {body}"
+        );
+
+        env.stop_server();
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn an_iam_user_can_rotate_its_own_password_and_lose_its_sessions() -> Result<(), Box<dyn Error + Send + Sync>> {
+        init_logging();
+        let mut env = RustFSTestEnvironment::new().await?;
+        env.start_rustfs_server_with_env(vec![], &[("RUSTFS_IAM_MASTER_KEY", TEST_MASTER_KEY)])
+            .await?;
+
+        let (root_ak, root_sk) = (env.access_key.clone(), env.secret_key.clone());
+        let user_ak = "mfarotationuser";
+        let old_sk = "mfarotationsecret";
+        let new_sk = "mfarotationsecret2";
+
+        // Root creates the user.
+        let (status, body) = signed_request(
+            &env.url,
+            http::Method::PUT,
+            &format!("/rustfs/admin/v3/add-user?accessKey={user_ak}"),
+            Some(&format!(r#"{{"secretKey":"{old_sk}","status":"enabled"}}"#)),
+            &root_ak,
+            &root_sk,
+        )
+        .await?;
+        assert_eq!(status, reqwest::StatusCode::OK, "user creation must succeed, body: {body}");
+
+        // The user sees itself as mutable, unlike root.
+        let (status, body) = signed_request(&env.url, http::Method::GET, ACCOUNT_INFO_PATH, None, user_ak, old_sk).await?;
+        assert_eq!(status, reqwest::StatusCode::OK, "body: {body}");
+        let info = json(&body);
+        assert_eq!(info["identity_type"], "iam");
+        assert_eq!(info["credentials_source"], "iam");
+        assert_eq!(info["mutable"]["password"], true);
+
+        // The wrong current secret is refused, so a live session alone cannot
+        // rewrite the credential.
+        let (status, body) = signed_request(
+            &env.url,
+            http::Method::POST,
+            ACCOUNT_PASSWORD_PATH,
+            Some(&format!(r#"{{"current_secret_key":"wrong-secret","new_secret_key":"{new_sk}"}}"#)),
+            user_ak,
+            old_sk,
+        )
+        .await?;
+        assert!(status.is_client_error(), "a wrong current secret must be refused, body: {body}");
+
+        // The correct one rotates it.
+        let (status, body) = signed_request(
+            &env.url,
+            http::Method::POST,
+            ACCOUNT_PASSWORD_PATH,
+            Some(&format!(r#"{{"current_secret_key":"{old_sk}","new_secret_key":"{new_sk}"}}"#)),
+            user_ak,
+            old_sk,
+        )
+        .await?;
+        assert_eq!(status, reqwest::StatusCode::OK, "rotation must succeed, body: {body}");
+
+        // The new secret works and the old one does not.
+        let (status, body) = signed_request(&env.url, http::Method::GET, ACCOUNT_INFO_PATH, None, user_ak, new_sk).await?;
+        assert_eq!(status, reqwest::StatusCode::OK, "the new secret must work, body: {body}");
+
+        let (status, _) = signed_request(&env.url, http::Method::GET, ACCOUNT_INFO_PATH, None, user_ak, old_sk).await?;
+        assert!(status.is_client_error(), "the old secret must stop working, status: {status}");
+
+        env.stop_server();
+        Ok(())
+    }
+
+    /// Wait until the current time step differs from the one a code was just
+    /// consumed in, then return a code for it.
+    ///
+    /// Anti-replay burns the step, so a test that reuses a code inside its own
+    /// window would fail for the right reason at the wrong moment.
+    async fn wait_for_a_fresh_code(secret_base32: &str) -> String {
+        let step_at_start = SystemTime::now()
+            .duration_since(UNIX_EPOCH)
+            .expect("clock after the epoch")
+            .as_secs()
+            / 30;
+
+        loop {
+            let now = SystemTime::now()
+                .duration_since(UNIX_EPOCH)
+                .expect("clock after the epoch")
+                .as_secs();
+            if now / 30 > step_at_start {
+                return totp_now(secret_base32);
+            }
+            tokio::time::sleep(std::time::Duration::from_millis(500)).await;
+        }
+    }
+}
@@ -16,8 +16,10 @@

 #[cfg(test)]
 mod tests {
+    use std::borrow::Borrow;
+
    use crate::common::{RustFSTestEnvironment, init_logging, signed_s3_request};
-    use aws_sdk_s3::error::ProvideErrorMetadata;
+    use aws_sdk_s3::error::{ProvideErrorMetadata, SdkError};
    use aws_sdk_s3::types::{
        AccelerateConfiguration, BucketAccelerateStatus, BucketLoggingStatus, IndexDocument, LoggingEnabled, Payer,
        RequestPaymentConfiguration, WebsiteConfiguration,
@@ -26,6 +28,26 @@ mod tests {
    use http::header::CONTENT_TYPE;
    use tracing::info;

+    fn assert_s3_error<T, E, R>(result: Result<T, R>, expected_status: u16, expected_code: &str, context: &str)
+    where
+        T: std::fmt::Debug,
+        E: ProvideErrorMetadata + std::fmt::Debug,
+        R: Borrow<SdkError<E>> + std::fmt::Debug,
+    {
+        let error = result.expect_err(context);
+        let sdk_error = error.borrow();
+        assert_eq!(
+            sdk_error.raw_response().map(|response| response.status().as_u16()),
+            Some(expected_status),
+            "{context}: expected HTTP {expected_status}, got: {error:?}"
+        );
+        assert_eq!(
+            sdk_error.as_service_error().and_then(ProvideErrorMetadata::code),
+            Some(expected_code),
+            "{context}: expected {expected_code}, got: {error:?}"
+        );
+    }
+
    #[tokio::test]
    async fn test_dummy_bucket_compatibility_endpoints() {
        init_logging();
@@ -217,17 +239,11 @@ mod tests {
            .expect("DeleteBucketWebsite should return success");

        let website_after_delete = client.get_bucket_website().bucket(bucket).send().await;
-        assert!(
-            website_after_delete.is_err(),
-            "GetBucketWebsite should return NoSuchWebsiteConfiguration after deletion"
-        );
-        let website_err = website_after_delete.err().unwrap();
-        let website_code = website_err.as_service_error().and_then(|e| e.code());
-        assert!(
-            matches!(website_code, Some("NoSuchWebsiteConfiguration")),
-            "Unexpected GetBucketWebsite error code: {:?}, err: {:?}",
-            website_code,
-            website_err
+        assert_s3_error(
+            website_after_delete,
+            404,
+            "NoSuchWebsiteConfiguration",
+            "GetBucketWebsite after deleting the website configuration",
        );

        env.stop_server();
@@ -245,15 +261,7 @@ mod tests {
        let missing_bucket = "test-dummy-bucket-missing";

        let get_logging = client.get_bucket_logging().bucket(missing_bucket).send().await;
-        assert!(get_logging.is_err(), "GetBucketLogging should fail for missing bucket");
-        let get_logging_err = get_logging.err().unwrap();
-        let get_logging_code = get_logging_err.as_service_error().and_then(|e| e.code());
-        assert!(
-            matches!(get_logging_code, Some("NoSuchBucket")),
-            "Unexpected GetBucketLogging error code: {:?}, err: {:?}",
-            get_logging_code,
-            get_logging_err
-        );
+        assert_s3_error(get_logging, 404, "NoSuchBucket", "GetBucketLogging for a missing bucket");

        let put_logging = client
            .put_bucket_logging()
@@ -261,41 +269,22 @@ mod tests {
            .bucket_logging_status(BucketLoggingStatus::builder().build())
            .send()
            .await;
-        assert!(put_logging.is_err(), "PutBucketLogging should fail for missing bucket");
-        let put_logging_err = put_logging.err().unwrap();
-        let put_logging_code = put_logging_err.as_service_error().and_then(|e| e.code());
-        assert!(
-            matches!(put_logging_code, Some("NoSuchBucket")),
-            "Unexpected PutBucketLogging error code: {:?}, err: {:?}",
-            put_logging_code,
-            put_logging_err
-        );
+        assert_s3_error(put_logging, 404, "NoSuchBucket", "PutBucketLogging for a missing bucket");

        let get_accelerate = client
            .get_bucket_accelerate_configuration()
            .bucket(missing_bucket)
            .send()
            .await;
-        assert!(get_accelerate.is_err(), "GetBucketAccelerateConfiguration should fail for missing bucket");
-        let get_accelerate_err = get_accelerate.err().unwrap();
-        let get_accelerate_code = get_accelerate_err.as_service_error().and_then(|e| e.code());
-        assert!(
-            matches!(get_accelerate_code, Some("NoSuchBucket")),
-            "Unexpected GetBucketAccelerateConfiguration error code: {:?}, err: {:?}",
-            get_accelerate_code,
-            get_accelerate_err
+        assert_s3_error(
+            get_accelerate,
+            404,
+            "NoSuchBucket",
+            "GetBucketAccelerateConfiguration for a missing bucket",
        );

        let get_request_payment = client.get_bucket_request_payment().bucket(missing_bucket).send().await;
-        assert!(get_request_payment.is_err(), "GetBucketRequestPayment should fail for missing bucket");
-        let get_request_payment_err = get_request_payment.err().unwrap();
-        let get_request_payment_code = get_request_payment_err.as_service_error().and_then(|e| e.code());
-        assert!(
-            matches!(get_request_payment_code, Some("NoSuchBucket")),
-            "Unexpected GetBucketRequestPayment error code: {:?}, err: {:?}",
-            get_request_payment_code,
-            get_request_payment_err
-        );
+        assert_s3_error(get_request_payment, 404, "NoSuchBucket", "GetBucketRequestPayment for a missing bucket");

        let put_accelerate = client
            .put_bucket_accelerate_configuration()
@@ -307,14 +296,11 @@ mod tests {
            )
            .send()
            .await;
-        assert!(put_accelerate.is_err(), "PutBucketAccelerateConfiguration should fail for missing bucket");
-        let put_accelerate_err = put_accelerate.err().unwrap();
-        let put_accelerate_code = put_accelerate_err.as_service_error().and_then(|e| e.code());
-        assert!(
-            matches!(put_accelerate_code, Some("NoSuchBucket")),
-            "Unexpected PutBucketAccelerateConfiguration error code: {:?}, err: {:?}",
-            put_accelerate_code,
-            put_accelerate_err
+        assert_s3_error(
+            put_accelerate,
+            404,
+            "NoSuchBucket",
+            "PutBucketAccelerateConfiguration for a missing bucket",
        );

        let put_request_payment = client
@@ -328,15 +314,7 @@ mod tests {
            )
            .send()
            .await;
-        assert!(put_request_payment.is_err(), "PutBucketRequestPayment should fail for missing bucket");
-        let put_request_payment_err = put_request_payment.err().unwrap();
-        let put_request_payment_code = put_request_payment_err.as_service_error().and_then(|e| e.code());
-        assert!(
-            matches!(put_request_payment_code, Some("NoSuchBucket")),
-            "Unexpected PutBucketRequestPayment error code: {:?}, err: {:?}",
-            put_request_payment_code,
-            put_request_payment_err
-        );
+        assert_s3_error(put_request_payment, 404, "NoSuchBucket", "PutBucketRequestPayment for a missing bucket");

        let put_website = client
            .put_bucket_website()
@@ -353,37 +331,13 @@ mod tests {
            )
            .send()
            .await;
-        assert!(put_website.is_err(), "PutBucketWebsite should fail for missing bucket");
-        let put_website_err = put_website.err().unwrap();
-        let put_website_code = put_website_err.as_service_error().and_then(|e| e.code());
-        assert!(
-            matches!(put_website_code, Some("NoSuchBucket")),
-            "Unexpected PutBucketWebsite error code: {:?}, err: {:?}",
-            put_website_code,
-            put_website_err
-        );
+        assert_s3_error(put_website, 404, "NoSuchBucket", "PutBucketWebsite for a missing bucket");

        let get_website = client.get_bucket_website().bucket(missing_bucket).send().await;
-        assert!(get_website.is_err(), "GetBucketWebsite should fail for missing bucket");
-        let get_website_err = get_website.err().unwrap();
-        let get_website_code = get_website_err.as_service_error().and_then(|e| e.code());
-        assert!(
-            matches!(get_website_code, Some("NoSuchBucket")),
-            "Unexpected GetBucketWebsite error code: {:?}, err: {:?}",
-            get_website_code,
-            get_website_err
-        );
+        assert_s3_error(get_website, 404, "NoSuchBucket", "GetBucketWebsite for a missing bucket");

        let delete_website = client.delete_bucket_website().bucket(missing_bucket).send().await;
-        assert!(delete_website.is_err(), "DeleteBucketWebsite should fail for missing bucket");
-        let delete_website_err = delete_website.err().unwrap();
-        let delete_website_code = delete_website_err.as_service_error().and_then(|e| e.code());
-        assert!(
-            matches!(delete_website_code, Some("NoSuchBucket")),
-            "Unexpected DeleteBucketWebsite error code: {:?}, err: {:?}",
-            delete_website_code,
-            delete_website_err
-        );
+        assert_s3_error(delete_website, 404, "NoSuchBucket", "DeleteBucketWebsite for a missing bucket");

        env.stop_server();
    }
@@ -15,38 +15,23 @@
 //! Regression test for Issue #1423
 //! Verifies that Bucket Policies are honored for Authenticated Users.

-use crate::common::{RustFSTestEnvironment, init_logging};
-use aws_sdk_s3::config::{Credentials, Region};
-use aws_sdk_s3::{Client, Config};
+use crate::common::{AdminTransport, RustFSTestEnvironment, admin_create_user_via, init_logging};
+use aws_sdk_s3::Client;
+use aws_sdk_s3::error::ProvideErrorMetadata;
 use tracing::info;

+/// This suite deliberately drives the admin API through the external `awscurl`
+/// binary, so user creation pins `AdminTransport::Awscurl`.
 async fn create_user(
    env: &RustFSTestEnvironment,
    username: &str,
    password: &str,
 ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-    let create_user_body = serde_json::json!({
-        "secretKey": password,
-        "status": "enabled"
-    })
-    .to_string();
-
-    let create_user_url = format!("{}/rustfs/admin/v3/add-user?accessKey={}", env.url, username);
-    crate::common::awscurl_put(&create_user_url, &create_user_body, &env.access_key, &env.secret_key).await?;
-    Ok(())
+    admin_create_user_via(AdminTransport::Awscurl, &env.url, &env.access_key, &env.secret_key, username, password).await
 }

 fn create_user_client(env: &RustFSTestEnvironment, access_key: &str, secret_key: &str) -> Client {
-    let credentials = Credentials::new(access_key, secret_key, None, None, "test-user");
-    let config = Config::builder()
-        .credentials_provider(credentials)
-        .region(Region::new("us-east-1"))
-        .endpoint_url(&env.url)
-        .force_path_style(true)
-        .behavior_version_latest()
-        .build();
-
-    Client::from_conf(config)
+    env.create_s3_client_with_credentials(access_key, secret_key)
 }

 #[tokio::test]
@@ -73,10 +58,14 @@ async fn test_bucket_policy_authenticated_user() -> Result<(), Box<dyn std::erro
    let user_client = create_user_client(&env, user_access, user_secret);

    // 4. Verify Access Denied initially (No Policy)
-    let result = user_client.list_objects_v2().bucket(bucket_name).send().await;
-    if result.is_ok() {
-        return Err("Should be Access Denied initially".into());
-    }
+    let denied = user_client
+        .list_objects_v2()
+        .bucket(bucket_name)
+        .send()
+        .await
+        .expect_err("a user without a bucket policy must be denied");
+    assert_eq!(denied.raw_response().map(|response| response.status().as_u16()), Some(403));
+    assert_eq!(denied.as_service_error().and_then(ProvideErrorMetadata::code), Some("AccessDenied"));

    // 5. Apply Bucket Policy Allowed User
    let policy_json = serde_json::json!({
@@ -20,10 +20,10 @@ mod tests {
    use crate::common::{RustFSTestEnvironment, init_logging};
    use aws_sdk_s3::Client;
    use aws_sdk_s3::config::{Credentials, Region, RequestChecksumCalculation};
+    use aws_sdk_s3::error::ProvideErrorMetadata;
    use aws_sdk_s3::primitives::ByteStream;
    use aws_sdk_s3::types::{ChecksumAlgorithm, ChecksumMode, CompletedMultipartUpload, CompletedPart};
    use aws_smithy_http_client::Builder as SmithyHttpClientBuilder;
-    use base64::Engine;
    use md5::{Digest as Md5Digest, Md5};
    use rustfs_rio::{Checksum, ChecksumType as RioChecksumType};
    use sha2::Sha256;
@@ -73,12 +73,12 @@ mod tests {
        let mut hasher = Md5::new();
        hasher.update(body);
        let digest = hasher.finalize();
-        base64::engine::general_purpose::STANDARD.encode(digest.as_slice())
+        base64_simd::STANDARD.encode_to_string(digest.as_slice())
    }

    fn checksum_sha256_base64(body: &[u8]) -> String {
        let digest = Sha256::digest(body);
-        base64::engine::general_purpose::STANDARD.encode(digest.as_slice())
+        base64_simd::STANDARD.encode_to_string(digest.as_slice())
    }

    fn checksum_crc64nvme_base64(body: &[u8]) -> String {
@@ -186,21 +186,31 @@ mod tests {
            .send()
            .await;

-        assert!(
-            result.is_err(),
-            "PutObject with a mismatched SHA256 must be rejected, but it succeeded (issue #4341)"
+        let error = result.expect_err("PutObject with a mismatched SHA256 must be rejected (issue #4341)");
+        assert_eq!(
+            error.raw_response().map(|response| response.status().as_u16()),
+            Some(400),
+            "Mismatched SHA256 must return HTTP 400, got {error:?}"
        );
-        let err = result.err().unwrap();
-        let msg = format!("{err:?}");
-        info!("PutObject correctly rejected mismatched checksum: {msg}");
-        assert!(
-            msg.contains("BadDigest") || msg.to_lowercase().contains("digest") || msg.to_lowercase().contains("checksum"),
-            "Expected a BadDigest/checksum error, got: {msg}"
+        assert_eq!(
+            error.as_service_error().and_then(ProvideErrorMetadata::code),
+            Some("BadDigest"),
+            "Mismatched SHA256 must return BadDigest, got {error:?}"
        );

        // And the object must not have been stored.
-        let head = client.head_object().bucket(bucket).key(key).send().await;
-        assert!(head.is_err(), "Object must not exist after a rejected mismatched-checksum PutObject");
+        let error = client
+            .head_object()
+            .bucket(bucket)
+            .key(key)
+            .send()
+            .await
+            .expect_err("Object must not exist after a rejected mismatched-checksum PutObject");
+        assert_eq!(
+            error.raw_response().map(|response| response.status().as_u16()),
+            Some(404),
+            "Rejected mismatched-checksum PutObject absence probe must return HTTP 404, got {error:?}"
+        );
        info!("PASSED: PutObject rejects mismatched SHA256 and stores nothing");
    }

@@ -546,14 +556,29 @@ mod tests {
                })
                .send()
                .await;
-            assert!(put_bad.is_err(), "{header}: a mismatched checksum must be rejected");
-            let msg = format!("{:?}", put_bad.err().unwrap());
-            assert!(
-                msg.contains("BadDigest") || msg.to_lowercase().contains("digest") || msg.to_lowercase().contains("checksum"),
-                "{header}: expected a BadDigest/checksum error, got: {msg}"
+            let error = put_bad.expect_err("a mismatched checksum must be rejected");
+            assert_eq!(
+                error.raw_response().map(|response| response.status().as_u16()),
+                Some(400),
+                "{header}: mismatched checksum must return HTTP 400, got {error:?}"
+            );
+            assert_eq!(
+                error.as_service_error().and_then(ProvideErrorMetadata::code),
+                Some("BadDigest"),
+                "{header}: mismatched checksum must return BadDigest, got {error:?}"
+            );
+            let error = client
+                .head_object()
+                .bucket(bucket)
+                .key(&bad_key)
+                .send()
+                .await
+                .expect_err("nothing must be stored after a rejected PutObject");
+            assert_eq!(
+                error.raw_response().map(|response| response.status().as_u16()),
+                Some(404),
+                "{header}: rejected PutObject absence probe must return HTTP 404, got {error:?}"
            );
-            let head = client.head_object().bucket(bucket).key(&bad_key).send().await;
-            assert!(head.is_err(), "{header}: nothing must be stored after a rejected PutObject");

            info!("PASSED additional-checksum verify-on-write: {header}");
        }
@@ -310,6 +310,17 @@ pub fn rustfs_binary_path() -> PathBuf {
    rustfs_binary_path_with_features(requested_rustfs_build_features().as_deref())
 }

+fn resolve_rustfs_binary_path(workspace: &Path, configured_target_dir: Option<&Path>) -> PathBuf {
+    let mut path = match configured_target_dir {
+        Some(path) if path.is_absolute() => path.to_path_buf(),
+        Some(path) => workspace.join(path),
+        None => workspace.join("target"),
+    };
+    path.push(if cfg!(debug_assertions) { "debug" } else { "release" });
+    path.push(format!("rustfs{}", std::env::consts::EXE_SUFFIX));
+    path
+}
+
 /// Resolve the RustFS binary relative to the workspace, optionally requesting build features.
 pub fn rustfs_binary_path_with_features(requested_features: Option<&str>) -> PathBuf {
    if let Some(path) = std::env::var_os("CARGO_BIN_EXE_rustfs") {
@@ -317,11 +328,9 @@ pub fn rustfs_binary_path_with_features(requested_features: Option<&str>) -> Pat
    }
    let requested_features = requested_features.and_then(normalize_rustfs_build_features);

-    let mut binary_path = workspace_root();
-    binary_path.push("target");
-    let profile_dir = if cfg!(debug_assertions) { "debug" } else { "release" };
-    binary_path.push(profile_dir);
-    binary_path.push(format!("rustfs{}", std::env::consts::EXE_SUFFIX));
+    let workspace = workspace_root();
+    let configured_target_dir = std::env::var_os("CARGO_TARGET_DIR").map(PathBuf::from);
+    let binary_path = resolve_rustfs_binary_path(&workspace, configured_target_dir.as_deref());

    let features_match = binary_features_match(&binary_path, requested_features.as_deref());
    let source_is_newer = workspace_sources_newer_than_binary(&binary_path);
@@ -338,7 +347,7 @@ pub fn rustfs_binary_path_with_features(requested_features: Option<&str>) -> Pat
    }

    info!("Building RustFS binary to ensure it's up to date...");
-    build_rustfs_binary(requested_features.as_deref());
+    build_rustfs_binary(requested_features.as_deref(), &binary_path);

    info!("Using RustFS binary at {:?}", binary_path);
    binary_path
@@ -440,7 +449,7 @@ fn path_is_newer_than(binary_modified: std::time::SystemTime, path: &Path) -> bo
 }

 /// Build the RustFS binary using cargo
-fn build_rustfs_binary(requested_features: Option<&str>) {
+fn build_rustfs_binary(requested_features: Option<&str>, binary_path: &Path) {
    let workspace = workspace_root();
    info!("Building RustFS binary from workspace: {:?}", workspace);

@@ -476,11 +485,7 @@ fn build_rustfs_binary(requested_features: Option<&str>) {
        panic!("Failed to build RustFS binary. Error: {stderr}");
    }

-    let mut binary_path = workspace;
-    binary_path.push("target");
-    binary_path.push(if cfg!(debug_assertions) { "debug" } else { "release" });
-    binary_path.push(format!("rustfs{}", std::env::consts::EXE_SUFFIX));
-    let stamp_path = rustfs_binary_features_stamp_path(&binary_path);
+    let stamp_path = rustfs_binary_features_stamp_path(binary_path);
    if let Err(err) = stdfs::write(&stamp_path, requested_features.unwrap_or_default()) {
        warn!("Failed to write RustFS feature stamp {:?}: {}", stamp_path, err);
    }
@@ -633,6 +638,18 @@ impl RustFSTestEnvironment {
        extra_args: Vec<&str>,
        extra_env: &[(&str, &str)],
        cleanup_existing: bool,
+    ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+        let binary_path = rustfs_binary_path();
+        self.start_rustfs_server_inner_with_binary(&binary_path, extra_args, extra_env, cleanup_existing)
+            .await
+    }
+
+    async fn start_rustfs_server_inner_with_binary(
+        &mut self,
+        binary_path: &Path,
+        extra_args: Vec<&str>,
+        extra_env: &[(&str, &str)],
+        cleanup_existing: bool,
    ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
        if cleanup_existing {
            self.cleanup_existing_processes().await?;
@@ -642,8 +659,7 @@ impl RustFSTestEnvironment {

        info!("Starting RustFS server with args: {:?}", args);

-        let binary_path = rustfs_binary_path();
-        let mut command = Command::new(&binary_path);
+        let mut command = Command::new(binary_path);
        command.env("RUST_LOG", "rustfs=info,rustfs_notify=debug");
        // The embedded console would bind the fixed default port :9001, which
        // collides with unrelated local services (e.g. Docker Desktop). Tests
@@ -663,6 +679,19 @@ impl RustFSTestEnvironment {
        Ok(())
    }

+    /// Start a specific RustFS binary against this environment's isolated
+    /// data directory. Upgrade tests use this to seed an old on-disk format
+    /// before restarting the same environment with the workspace binary.
+    pub async fn start_rustfs_server_from_binary(
+        &mut self,
+        binary_path: &Path,
+        extra_args: Vec<&str>,
+        extra_env: &[(&str, &str)],
+    ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+        self.start_rustfs_server_inner_with_binary(binary_path, extra_args, extra_env, true)
+            .await
+    }
+
    /// Start RustFS server with basic configuration
    pub async fn start_rustfs_server(&mut self, extra_args: Vec<&str>) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
        self.start_rustfs_server_inner(extra_args, &[], true).await
@@ -1715,30 +1744,128 @@ pub(crate) async fn admin_create_user(
    username: &str,
    secret_key: &str,
 ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-    let url = format!("{}/rustfs/admin/v3/add-user?accessKey={}", env.url, username);
-    let body = serde_json::json!({
-        "secretKey": secret_key,
-        "status": "enabled"
-    });
-    let response = signed_request(
-        http::Method::PUT,
-        &url,
-        &env.access_key,
-        &env.secret_key,
-        Some(body.to_string().into_bytes()),
-        Some("application/json"),
-    )
-    .await?;
+    admin_create_user_via(AdminTransport::Signed, &env.url, &env.access_key, &env.secret_key, username, secret_key).await
+}

-    if response.status() != reqwest::StatusCode::OK {
-        let status = response.status();
-        let body = response.text().await.unwrap_or_default();
-        return Err(format!("create user failed: {status} {body}").into());
+/// Transport used by the shared admin-API helpers: in-process SigV4 signing
+/// via [`signed_request`], or the external `awscurl` binary (an independent
+/// SigV4 implementation exercised by the awscurl-gated suites).
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub(crate) enum AdminTransport {
+    Signed,
+    Awscurl,
+}
+
+/// Execute an admin-API request against `base_url` with admin credentials over
+/// the chosen transport, failing on any non-success response.
+pub(crate) async fn admin_execute_at(
+    transport: AdminTransport,
+    method: http::Method,
+    base_url: &str,
+    admin_access_key: &str,
+    admin_secret_key: &str,
+    path_and_query: &str,
+    body: Option<&str>,
+) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+    let url = format!("{base_url}{path_and_query}");
+    match transport {
+        AdminTransport::Signed => {
+            let content_type = match body {
+                Some(body) if !body.is_empty() => Some("application/json"),
+                _ => None,
+            };
+            let response = signed_request(
+                method.clone(),
+                &url,
+                admin_access_key,
+                admin_secret_key,
+                body.map(|body| body.as_bytes().to_vec()),
+                content_type,
+            )
+            .await?;
+            if !response.status().is_success() {
+                let status = response.status();
+                let text = response.text().await.unwrap_or_default();
+                return Err(format!("{method} {path_and_query} failed: {status} {text}").into());
+            }
+        }
+        AdminTransport::Awscurl => {
+            execute_awscurl(&url, method.as_str(), body, admin_access_key, admin_secret_key).await?;
+        }
    }
-
    Ok(())
 }

+/// Create a new IAM user via the admin API over the chosen transport.
+pub(crate) async fn admin_create_user_via(
+    transport: AdminTransport,
+    base_url: &str,
+    admin_access_key: &str,
+    admin_secret_key: &str,
+    username: &str,
+    secret_key: &str,
+) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+    let path = format!("/rustfs/admin/v3/add-user?accessKey={username}");
+    let body = serde_json::json!({"secretKey": secret_key, "status": "enabled"}).to_string();
+    admin_execute_at(
+        transport,
+        http::Method::PUT,
+        base_url,
+        admin_access_key,
+        admin_secret_key,
+        &path,
+        Some(&body),
+    )
+    .await
+}
+
+/// Install a canned policy via the admin API over the chosen transport.
+pub(crate) async fn admin_add_canned_policy_via(
+    transport: AdminTransport,
+    base_url: &str,
+    admin_access_key: &str,
+    admin_secret_key: &str,
+    policy_name: &str,
+    policy_json: &str,
+) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+    let path = format!("/rustfs/admin/v3/add-canned-policy?name={policy_name}");
+    admin_execute_at(
+        transport,
+        http::Method::PUT,
+        base_url,
+        admin_access_key,
+        admin_secret_key,
+        &path,
+        Some(policy_json),
+    )
+    .await
+}
+
+/// Attach a canned policy to a user via the admin API over the chosen transport.
+pub(crate) async fn admin_attach_user_policy_via(
+    transport: AdminTransport,
+    base_url: &str,
+    admin_access_key: &str,
+    admin_secret_key: &str,
+    policy_name: &str,
+    username: &str,
+) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+    let path = format!("/rustfs/admin/v3/set-user-or-group-policy?policyName={policy_name}&userOrGroup={username}&isGroup=false");
+    // `Some("")` preserves the historical wire shape on both transports: awscurl
+    // keeps sending `-d ''` and the signed path attaches an empty body with no
+    // content type.
+    admin_execute_at(
+        transport,
+        http::Method::PUT,
+        base_url,
+        admin_access_key,
+        admin_secret_key,
+        &path,
+        Some(""),
+    )
+    .await
+}
+
 #[cfg(test)]
 mod tests {
    use super::*;
@@ -1776,6 +1903,27 @@ mod tests {
        );
    }

+    #[test]
+    fn resolves_rustfs_binary_in_configured_cargo_target_directory() {
+        let workspace = Path::new("workspace");
+        let profile = if cfg!(debug_assertions) { "debug" } else { "release" };
+        let binary = format!("rustfs{}", std::env::consts::EXE_SUFFIX);
+        assert_eq!(
+            resolve_rustfs_binary_path(workspace, None),
+            workspace.join("target").join(profile).join(&binary)
+        );
+        assert_eq!(
+            resolve_rustfs_binary_path(workspace, Some(Path::new("custom-target"))),
+            workspace.join("custom-target").join(profile).join(&binary)
+        );
+
+        let absolute = std::env::temp_dir().join("rustfs-e2e-custom-target");
+        assert_eq!(
+            resolve_rustfs_binary_path(workspace, Some(&absolute)),
+            absolute.join(profile).join(binary)
+        );
+    }
+
    #[test]
    fn full_feature_enables_any_required_feature() {
        assert!(rustfs_build_feature_enabled(Some("full"), "sftp"));
@@ -4,7 +4,8 @@ use crate::common::{RustFSTestEnvironment, init_logging, rustfs_binary_path};
 use aws_sdk_s3::primitives::ByteStream;
 use aws_sdk_s3::types::{CompletedMultipartUpload, CompletedPart};
 use std::fs;
-use std::path::PathBuf;
+use std::io;
+use std::path::{Path, PathBuf};
 use std::process::Command;
 use tracing::info;

@@ -31,30 +32,58 @@ fn generate_high_ratio_binary_data(size: usize, seed: u8) -> Vec<u8> {
        .collect()
 }

-fn find_part_files(temp_dir: &str, bucket: &str, object_key: &str) -> Vec<PathBuf> {
+fn find_part_files(temp_dir: &str, bucket: &str, object_key: &str) -> io::Result<Vec<PathBuf>> {
    let bucket_path = PathBuf::from(temp_dir).join(bucket);
    let mut part_files = Vec::new();

-    fn scan_dir(dir: &PathBuf, target: &str, results: &mut Vec<PathBuf>) {
-        if let Ok(entries) = fs::read_dir(dir) {
-            for entry in entries.flatten() {
-                let path = entry.path();
-                if path.is_dir() {
-                    scan_dir(&path, target, results);
-                } else if path
-                    .file_name()
-                    .map(|n| n.to_string_lossy().starts_with("part."))
-                    .unwrap_or(false)
-                    && path.to_string_lossy().contains(target)
-                {
-                    results.push(path);
+    fn scan_dir(dir: &Path, target: &str, results: &mut Vec<PathBuf>) -> io::Result<()> {
+        let entries = fs::read_dir(dir)
+            .map_err(|error| io::Error::new(error.kind(), format!("failed to read {}: {error}", dir.display())))?;
+        for entry in entries {
+            let entry = entry
+                .map_err(|error| io::Error::new(error.kind(), format!("failed to read entry in {}: {error}", dir.display())))?;
+            let path = entry.path();
+            let file_type = entry
+                .file_type()
+                .map_err(|error| io::Error::new(error.kind(), format!("failed to inspect {}: {error}", path.display())))?;
+            if file_type.is_dir() {
+                scan_dir(&path, target, results)?;
+            } else if path
+                .file_name()
+                .map(|n| n.to_string_lossy().starts_with("part."))
+                .unwrap_or(false)
+                && path.to_string_lossy().contains(target)
+            {
+                if !file_type.is_file() {
+                    return Err(io::Error::new(
+                        io::ErrorKind::InvalidData,
+                        format!("expected regular part file at {}", path.display()),
+                    ));
                }
+                results.push(path);
            }
        }
+        Ok(())
    }

-    scan_dir(&bucket_path, object_key, &mut part_files);
-    part_files
+    scan_dir(&bucket_path, object_key, &mut part_files)?;
+    Ok(part_files)
+}
+
+fn part_files_total_size(part_files: &[PathBuf]) -> io::Result<u64> {
+    part_files.iter().try_fold(0_u64, |total, path| {
+        let metadata = fs::symlink_metadata(path)
+            .map_err(|error| io::Error::new(error.kind(), format!("failed to stat {}: {error}", path.display())))?;
+        if !metadata.file_type().is_file() {
+            return Err(io::Error::new(
+                io::ErrorKind::InvalidData,
+                format!("expected regular part file at {}", path.display()),
+            ));
+        }
+        total
+            .checked_add(metadata.len())
+            .ok_or_else(|| io::Error::new(io::ErrorKind::InvalidData, "on-disk part size overflow"))
+    })
 }

 async fn start_rustfs_with_compression(env: &mut RustFSTestEnvironment) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
@@ -123,8 +152,9 @@ async fn test_compression_roundtrip() -> Result<(), Box<dyn std::error::Error +
    let content_length = head_response.content_length().unwrap_or(0);
    assert_eq!(content_length as usize, original_size, "Content-Length should be original size");

-    let part_files = find_part_files(&env.temp_dir, COMPRESSION_TEST_BUCKET, object_key);
-    let total_physical_size: u64 = part_files.iter().filter_map(|p| fs::metadata(p).ok()).map(|m| m.len()).sum();
+    let part_files = find_part_files(&env.temp_dir, COMPRESSION_TEST_BUCKET, object_key)?;
+    assert!(!part_files.is_empty(), "expected on-disk part files for the compressed object");
+    let total_physical_size = part_files_total_size(&part_files)?;

    assert!(
        total_physical_size < original_size as u64,
@@ -246,9 +276,9 @@ async fn test_compression_multipart_roundtrip() -> Result<(), Box<dyn std::error
        "Content-Length should be the logical object size"
    );

-    let part_files = find_part_files(&env.temp_dir, MULTIPART_COMPRESSION_BUCKET, object_key);
+    let part_files = find_part_files(&env.temp_dir, MULTIPART_COMPRESSION_BUCKET, object_key)?;
    assert!(!part_files.is_empty(), "expected on-disk part files for the multipart object");
-    let total_physical_size: u64 = part_files.iter().filter_map(|p| fs::metadata(p).ok()).map(|m| m.len()).sum();
+    let total_physical_size = part_files_total_size(&part_files)?;
    assert!(
        total_physical_size < (total_size / 2) as u64,
        "Physical size {total_physical_size} should be well below original size {total_size} (multipart compression applied)"
@@ -366,9 +396,9 @@ async fn test_compression_multipart_high_ratio_binary_roundtrip() -> Result<(),

    // This pattern compresses to roughly 1/50 of its logical size, so a comfortably loose 2x
    // margin still proves the parts were stored compressed rather than raw or double-encoded.
-    let part_files = find_part_files(&env.temp_dir, MPU_HIGH_RATIO_BUCKET, object_key);
+    let part_files = find_part_files(&env.temp_dir, MPU_HIGH_RATIO_BUCKET, object_key)?;
    assert!(!part_files.is_empty(), "expected on-disk part files for the multipart object");
-    let total_physical_size: u64 = part_files.iter().filter_map(|p| fs::metadata(p).ok()).map(|m| m.len()).sum();
+    let total_physical_size = part_files_total_size(&part_files)?;
    assert!(
        total_physical_size < (total_size as u64) / 2,
        "Physical size {total_physical_size} should be far below the logical size {total_size} for high-ratio data"
@@ -522,9 +552,9 @@ async fn test_compression_multipart_upload_part_copy_roundtrip() -> Result<(), B
        "Content-Length should be the logical object size"
    );

-    let part_files = find_part_files(&env.temp_dir, MPU_COPY_COMPRESSION_BUCKET, target_key);
+    let part_files = find_part_files(&env.temp_dir, MPU_COPY_COMPRESSION_BUCKET, target_key)?;
    assert!(!part_files.is_empty(), "expected on-disk part files for the copied object");
-    let total_physical_size: u64 = part_files.iter().filter_map(|p| fs::metadata(p).ok()).map(|m| m.len()).sum();
+    let total_physical_size = part_files_total_size(&part_files)?;
    assert!(
        total_physical_size < (total_size / 2) as u64,
        "Physical size {total_physical_size} should be well below original size {total_size} (copied part compression applied)"
@@ -585,9 +615,9 @@ async fn test_compression_multipart_three_parts_part_number_gets() -> Result<(),
        "Content-Length should be the logical object size"
    );

-    let part_files = find_part_files(&env.temp_dir, MPU_THREE_PARTS_BUCKET, object_key);
+    let part_files = find_part_files(&env.temp_dir, MPU_THREE_PARTS_BUCKET, object_key)?;
    assert!(!part_files.is_empty(), "expected on-disk part files for the multipart object");
-    let total_physical_size: u64 = part_files.iter().filter_map(|p| fs::metadata(p).ok()).map(|m| m.len()).sum();
+    let total_physical_size = part_files_total_size(&part_files)?;
    assert!(
        total_physical_size < (total_size / 2) as u64,
        "Physical size {total_physical_size} should be well below original size {total_size} (multipart compression applied)"
@@ -622,11 +652,10 @@ const MPU_SSE_COMPRESSION_BUCKET: &str = "compression-mpu-sse-bucket";
 async fn start_rustfs_with_compression_and_sse(
    env: &mut RustFSTestEnvironment,
 ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-    use base64::Engine;
    env.cleanup_existing_processes().await?;

    let binary_path = rustfs_binary_path();
-    let master_key = base64::engine::general_purpose::STANDARD.encode([0x42u8; 32]);
+    let master_key = base64_simd::STANDARD.encode_to_string([0x42u8; 32]);
    // Server output goes to a file inside the per-test temp dir so a failing
    // run can be diagnosed from the child's logs.
    let server_log = std::fs::File::create(format!("{}/server.log", env.temp_dir))?;
@@ -734,9 +763,9 @@ async fn test_compression_multipart_sse_s3_roundtrip() -> Result<(), Box<dyn std
        "HEAD must report SSE-S3"
    );

-    let part_files = find_part_files(&env.temp_dir, MPU_SSE_COMPRESSION_BUCKET, object_key);
+    let part_files = find_part_files(&env.temp_dir, MPU_SSE_COMPRESSION_BUCKET, object_key)?;
    assert!(!part_files.is_empty(), "expected on-disk part files for the multipart object");
-    let total_physical_size: u64 = part_files.iter().filter_map(|p| fs::metadata(p).ok()).map(|m| m.len()).sum();
+    let total_physical_size = part_files_total_size(&part_files)?;
    assert!(
        total_physical_size < (total_size / 2) as u64,
        "Physical size {total_physical_size} should be well below original size {total_size} (compress-then-encrypt applied)"
@@ -27,8 +27,7 @@ mod tests {
        VersioningConfiguration,
    };
    use aws_smithy_http_client::Builder as SmithyHttpClientBuilder;
-    use base64::Engine as _;
-    use base64::engine::general_purpose::STANDARD as BASE64;
+    use base64_simd::STANDARD as BASE64;
    use rustfs_rio::{Checksum, ChecksumType as RioChecksumType};
    use sha2::{Digest, Sha256};
    use tracing::info;
@@ -465,7 +464,7 @@ mod tests {
        create_versioned_bucket(&client, dst_bucket).await;

        let content = b"deterministic synthetic payload for copy-object checksum #4996";
-        let expected_sha256 = BASE64.encode(Sha256::digest(content));
+        let expected_sha256 = BASE64.encode_to_string(Sha256::digest(content));

        client
            .put_object()
@@ -534,7 +533,7 @@ mod tests {
        create_versioned_bucket(&client, dst_bucket).await;

        let content = b"another deterministic payload whose source checksum must survive the copy";
-        let expected_sha256 = BASE64.encode(Sha256::digest(content));
+        let expected_sha256 = BASE64.encode_to_string(Sha256::digest(content));

        // Store the source WITH a SHA-256 checksum so it has one to preserve.
        let put_src = client
@@ -614,7 +613,7 @@ mod tests {
        create_versioned_bucket(&client, dst_bucket).await;

        let content = b"payload whose copy must be re-checksummed with a different algorithm";
-        let expected_sha256 = BASE64.encode(Sha256::digest(content));
+        let expected_sha256 = BASE64.encode_to_string(Sha256::digest(content));

        // Source is stored WITH a SHA-256 checksum.
        client
@@ -59,7 +59,6 @@ where
 /// Regression test for data usage accuracy (issue #1012).
 /// Launches rustfs, writes 1000 objects, then asserts admin data usage reports the full count.
 #[tokio::test(flavor = "multi_thread")]
-#[ignore = "Starts a rustfs server and requires awscurl; enable when running full E2E"]
 async fn data_usage_reports_all_objects() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    init_logging();

@@ -86,28 +85,20 @@ async fn data_usage_reports_all_objects() -> Result<(), Box<dyn std::error::Erro
        usage
            .buckets_usage
            .get(TEST_BUCKET)
-            .map(|bucket_usage| usage.objects_total_count >= 1000 && bucket_usage.objects_count >= 1000)
+            .map(|bucket_usage| usage.objects_total_count == 1000 && bucket_usage.objects_count == 1000)
            .unwrap_or(false)
    })
    .await?;

-    // Assert total object count and per-bucket count are not truncated
+    // Assert total object count and per-bucket count are exact.
    let bucket_usage = usage
        .buckets_usage
        .get(TEST_BUCKET)
        .cloned()
        .expect("bucket usage should exist");

-    assert!(
-        usage.objects_total_count >= 1000,
-        "total object count should be at least 1000, got {}",
-        usage.objects_total_count
-    );
-    assert!(
-        bucket_usage.objects_count >= 1000,
-        "bucket object count should be at least 1000, got {}",
-        bucket_usage.objects_count
-    );
+    assert_eq!(usage.objects_total_count, 1000, "total object count should be exact");
+    assert_eq!(bucket_usage.objects_count, 1000, "bucket object count should be exact");

    env.stop_server();
    Ok(())
@@ -116,7 +107,6 @@ async fn data_usage_reports_all_objects() -> Result<(), Box<dyn std::error::Erro
 /// Regression test for issue #3898.
 /// Versioned buckets should expose versions and delete markers through admin data usage.
 #[tokio::test(flavor = "multi_thread")]
-#[ignore = "Starts a rustfs server and requires awscurl; enable when running full E2E"]
 async fn data_usage_reports_versioned_objects_and_delete_markers() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    init_logging();

@@ -23,8 +23,8 @@
 //! It was fixed in three layers on `main`, each with its own *unit* regression:
 //!   * rustfs#4594 — `GetObjectStreamingReader::poll_read` now returns
 //!     `UnexpectedEof` on a short body instead of a clean `Ok(())`
-//!     (`rustfs/src/app/object_usecase.rs`,
-//!     `app::object_usecase::tests::get_object_streaming_reader_errors_on_short_eof`).
+//!     (`rustfs/src/app/object/get.rs`,
+//!     `app::object::get::tests::get_object_streaming_reader_errors_on_short_eof`).
 //!   * rustfs#4560 — the lazy multipart codec reader degrades a later part to
 //!     the legacy per-part decode in place, and surfaces reconstruction errors
 //!     instead of silently truncating
@@ -155,8 +155,13 @@ mod tests {
            .key(key)
            .version_id(version_id)
            .send()
-            .await;
-        assert!(get_deleted_version.is_err(), "explicitly deleted version should no longer be readable");
+            .await
+            .expect_err("explicitly deleted version should no longer be readable");
+        assert_eq!(
+            get_deleted_version.raw_response().map(|response| response.status().as_u16()),
+            Some(404),
+            "explicitly deleted version absence probe must return HTTP 404, got {get_deleted_version:?}"
+        );

        Ok(())
    }
@@ -117,9 +117,18 @@ mod tests {
        );

        // Verify HEAD returns 404
-        let head = client.head_object().bucket(bucket).key("to-delete.txt").send().await;
-
-        assert!(head.is_err(), "RT-05 FAIL: HEAD on deleted object should return error, got success");
+        let error = client
+            .head_object()
+            .bucket(bucket)
+            .key("to-delete.txt")
+            .send()
+            .await
+            .expect_err("RT-05 FAIL: HEAD on deleted object should return 404, got success");
+        assert_eq!(
+            error.raw_response().map(|response| response.status().as_u16()),
+            Some(404),
+            "RT-05 FAIL: HEAD on deleted object must return HTTP 404, got {error:?}"
+        );

        info!("RT-05 PASS: delete correctly removes object from LIST and HEAD");
        Ok(())
@@ -414,9 +423,18 @@ mod tests {

        // All HEAD requests should return 404
        for key in &keys {
-            let head = client.head_object().bucket(bucket).key(*key).send().await;
-
-            assert!(head.is_err(), "RT-05f FAIL: HEAD on deleted key '{key}' should return error");
+            let error = client
+                .head_object()
+                .bucket(bucket)
+                .key(*key)
+                .send()
+                .await
+                .expect_err("RT-05f FAIL: HEAD on deleted key should return 404, got success");
+            assert_eq!(
+                error.raw_response().map(|response| response.status().as_u16()),
+                Some(404),
+                "RT-05f FAIL: HEAD on deleted key '{key}' must return HTTP 404, got {error:?}"
+            );
        }

        // LIST should be empty
@@ -16,36 +16,29 @@
 //! session policy** (`Policy` parameter) via `awscurl --service sts` with explicit
 //! `Content-Type: application/x-www-form-urlencoded` on `POST /`.

-use crate::common::{RustFSTestEnvironment, awscurl_delete, awscurl_post_sts_form_urlencoded, awscurl_put, init_logging};
-use aws_sdk_s3::config::{Credentials, Region};
+use crate::common::{
+    AdminTransport, RustFSTestEnvironment, admin_add_canned_policy_via, admin_attach_user_policy_via, admin_create_user_via,
+    awscurl_delete, awscurl_post_sts_form_urlencoded, build_test_s3_config, init_logging,
+};
+use aws_sdk_s3::Client;
+use aws_sdk_s3::error::ProvideErrorMetadata;
 use aws_sdk_s3::primitives::ByteStream;
 use aws_sdk_s3::types::{Delete, ObjectIdentifier, Tag, Tagging};
-use aws_sdk_s3::{Client, Config};
 use tracing::info;
 use uuid::Uuid;

 fn user_client(env: &RustFSTestEnvironment, access_key: &str, secret_key: &str) -> Client {
-    let credentials = Credentials::new(access_key, secret_key, None, None, "e2e-existing-tag");
-    let config = Config::builder()
-        .credentials_provider(credentials)
-        .region(Region::new("us-east-1"))
-        .endpoint_url(&env.url)
-        .force_path_style(true)
-        .behavior_version_latest()
-        .build();
-    Client::from_conf(config)
+    env.create_s3_client_with_credentials(access_key, secret_key)
 }

 fn sts_session_client(env: &RustFSTestEnvironment, access_key: &str, secret_key: &str, session_token: &str) -> Client {
-    let credentials = Credentials::new(access_key, secret_key, Some(session_token.into()), None, "e2e-sts-session");
-    let config = Config::builder()
-        .credentials_provider(credentials)
-        .region(Region::new("us-east-1"))
-        .endpoint_url(&env.url)
-        .force_path_style(true)
-        .behavior_version_latest()
-        .build();
-    Client::from_conf(config)
+    Client::from_conf(build_test_s3_config(
+        &env.url,
+        access_key,
+        secret_key,
+        Some(session_token),
+        "e2e-sts-session",
+    ))
 }

 fn extract_xml_tag(xml: &str, tag: &str) -> Option<String> {
@@ -76,15 +69,16 @@ async fn assume_role_with_session_policy(
    parse_assume_role_credentials(&xml)
 }

+// This suite deliberately drives the admin API through the external `awscurl`
+// binary (an independent SigV4 implementation), so the wrappers below pin
+// `AdminTransport::Awscurl`.
+
 async fn admin_create_user(
    env: &RustFSTestEnvironment,
    username: &str,
    password: &str,
 ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-    let body = serde_json::json!({ "secretKey": password, "status": "enabled" }).to_string();
-    let url = format!("{}/rustfs/admin/v3/add-user?accessKey={}", env.url, username);
-    awscurl_put(&url, &body, &env.access_key, &env.secret_key).await?;
-    Ok(())
+    admin_create_user_via(AdminTransport::Awscurl, &env.url, &env.access_key, &env.secret_key, username, password).await
 }

 async fn admin_add_canned_policy(
@@ -92,9 +86,15 @@ async fn admin_add_canned_policy(
    policy_name: &str,
    policy_json: &str,
 ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-    let url = format!("{}/rustfs/admin/v3/add-canned-policy?name={}", env.url, policy_name);
-    awscurl_put(&url, policy_json, &env.access_key, &env.secret_key).await?;
-    Ok(())
+    admin_add_canned_policy_via(
+        AdminTransport::Awscurl,
+        &env.url,
+        &env.access_key,
+        &env.secret_key,
+        policy_name,
+        policy_json,
+    )
+    .await
 }

 async fn admin_attach_policy_to_user(
@@ -102,12 +102,7 @@ async fn admin_attach_policy_to_user(
    policy_name: &str,
    username: &str,
 ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-    let url = format!(
-        "{}/rustfs/admin/v3/set-user-or-group-policy?policyName={}&userOrGroup={}&isGroup=false",
-        env.url, policy_name, username
-    );
-    awscurl_put(&url, "", &env.access_key, &env.secret_key).await?;
-    Ok(())
+    admin_attach_user_policy_via(AdminTransport::Awscurl, &env.url, &env.access_key, &env.secret_key, policy_name, username).await
 }

 async fn admin_remove_user(env: &RustFSTestEnvironment, username: &str) {
@@ -208,10 +203,17 @@ async fn test_e2e_iam_policy_existing_object_tag_get_object() -> Result<(), Box<
    let _ = out.body.collect().await?;

    put_object_tag_kv(&admin, &bucket, key, "security", "private").await?;
-    let denied = uclient.get_object().bucket(&bucket).key(key).send().await;
-    assert!(
-        denied.is_err(),
-        "GetObject must be denied when ExistingObjectTag no longer matches IAM policy"
+    let denied = uclient
+        .get_object()
+        .bucket(&bucket)
+        .key(key)
+        .send()
+        .await
+        .expect_err("GetObject must be denied when ExistingObjectTag no longer matches IAM policy");
+    assert_eq!(
+        denied.as_service_error().and_then(ProvideErrorMetadata::code),
+        Some("AccessDenied"),
+        "IAM ExistingObjectTag mismatch must return AccessDenied: {denied:?}"
    );

    cleanup_bucket_and_object(&admin, &bucket, key).await;
@@ -245,8 +247,13 @@ async fn test_e2e_bucket_policy_existing_object_tag_get_object() -> Result<(), B
        .bucket(&bucket)
        .key(key)
        .send()
-        .await;
-    assert!(deny_before.is_err(), "without bucket policy, user must be denied");
+        .await
+        .expect_err("without bucket policy, user must be denied");
+    assert_eq!(
+        deny_before.as_service_error().and_then(ProvideErrorMetadata::code),
+        Some("AccessDenied"),
+        "missing bucket policy must return AccessDenied: {deny_before:?}"
+    );

    let bp = serde_json::json!({
        "Version": "2012-10-17",
@@ -268,8 +275,18 @@ async fn test_e2e_bucket_policy_existing_object_tag_get_object() -> Result<(), B
    let _ = ok.body.collect().await?;

    put_object_tag_kv(&admin, &bucket, key, "security", "private").await?;
-    let denied = uclient.get_object().bucket(&bucket).key(key).send().await;
-    assert!(denied.is_err(), "GetObject must fail when tag no longer satisfies bucket policy");
+    let denied = uclient
+        .get_object()
+        .bucket(&bucket)
+        .key(key)
+        .send()
+        .await
+        .expect_err("GetObject must fail when tag no longer satisfies bucket policy");
+    assert_eq!(
+        denied.as_service_error().and_then(ProvideErrorMetadata::code),
+        Some("AccessDenied"),
+        "bucket-policy ExistingObjectTag mismatch must return AccessDenied: {denied:?}"
+    );

    cleanup_bucket_and_object(&admin, &bucket, key).await;
    admin_remove_user(&env, &user).await;
@@ -335,10 +352,17 @@ async fn test_e2e_sts_assume_role_session_policy_existing_object_tag() -> Result
    let _ = ok.body.collect().await?;

    put_object_tag_kv(&parent_client, &bucket, key, "security", "private").await?;
-    let denied = session_client.get_object().bucket(&bucket).key(key).send().await;
-    assert!(
-        denied.is_err(),
-        "session policy must deny GetObject when ExistingObjectTag no longer matches"
+    let denied = session_client
+        .get_object()
+        .bucket(&bucket)
+        .key(key)
+        .send()
+        .await
+        .expect_err("session policy must deny GetObject when ExistingObjectTag no longer matches");
+    assert_eq!(
+        denied.as_service_error().and_then(ProvideErrorMetadata::code),
+        Some("AccessDenied"),
+        "STS ExistingObjectTag mismatch must return AccessDenied: {denied:?}"
    );

    cleanup_bucket_and_object(&admin, &bucket, key).await;
@@ -433,8 +457,18 @@ async fn test_e2e_sts_session_policy_delete_objects_object_prefix_only() -> Resu
    assert_eq!(error.key(), Some(denied_key));
    assert_eq!(error.code(), Some("AccessDenied"));

-    let allowed_head = parent_client.head_object().bucket(&bucket).key(allowed_key).send().await;
-    assert!(allowed_head.is_err(), "allowed-prefix object should have been deleted");
+    let allowed_head = parent_client
+        .head_object()
+        .bucket(&bucket)
+        .key(allowed_key)
+        .send()
+        .await
+        .expect_err("allowed-prefix object should have been deleted");
+    assert_eq!(
+        allowed_head.raw_response().map(|response| response.status().as_u16()),
+        Some(404),
+        "allowed-prefix object absence probe must return HTTP 404, got {allowed_head:?}"
+    );

    parent_client
        .head_object()
@@ -1113,7 +1113,7 @@ fn md5_bytes(input: impl AsRef<[u8]>) -> [u8; 16] {
 fn md5_hex(input: impl AsRef<[u8]>) -> String {
    let mut hasher = Md5::new();
    hasher.update(input.as_ref());
-    hex::encode(hasher.finalize())
+    hex_simd::encode_to_string(hasher.finalize(), hex_simd::AsciiCase::Lower)
 }

 fn ensure_store_budget(state: &StoreState, removed_bytes: usize, added_bytes: usize, adds_version: bool) -> S3Result {
@@ -1375,7 +1375,7 @@ impl S3 for FakeBackend {
            Some(value) => value,
            None => {
                let (digest, _body_permit) = md5_digest(body.clone(), _body_permit).await?;
-                hex::encode(digest)
+                hex_simd::encode_to_string(digest, hex_simd::AsciiCase::Lower)
            }
        };
        let version = ObjectVersion {
@@ -1660,7 +1660,7 @@ impl S3 for FakeBackend {
        }
        let body = collect_stream(input.body, input.content_length, fault.as_ref(), &self.control).await?;
        let (digest, _body_permit) = md5_digest(body.clone(), _body_permit).await?;
-        let e_tag = hex::encode(digest);
+        let e_tag = hex_simd::encode_to_string(digest, hex_simd::AsciiCase::Lower);
        let mut state = lock(&self.store);
        let existing_bytes = state
            .uploads
@@ -15,20 +15,11 @@
 //! E2E tests for group management (fixes #2028).

 use crate::common::{RustFSTestEnvironment, admin_ok, admin_request, init_logging};
-use aws_sdk_s3::config::{Credentials, Region};
-use aws_sdk_s3::{Client, Config};
+use aws_sdk_s3::Client;
 use tracing::info;

 fn create_user_s3_client(env: &RustFSTestEnvironment, access_key: &str, secret_key: &str) -> Client {
-    let credentials = Credentials::new(access_key, secret_key, None, None, "e2e-group-test");
-    let config = Config::builder()
-        .credentials_provider(credentials)
-        .region(Region::new("us-east-1"))
-        .endpoint_url(&env.url)
-        .force_path_style(true)
-        .behavior_version_latest()
-        .build();
-    Client::from_conf(config)
+    env.create_s3_client_with_credentials(access_key, secret_key)
 }

 #[tokio::test(flavor = "multi_thread")]
@@ -15,23 +15,22 @@
 //! Four-node EC regression gate for inline storage and the inline GET reader.
 //!
 //! The storage decision is based on shard bytes (256 KiB / 32 KiB objects for
-//! the default EC 2+2 geometry), while the GET fast path has its own object-size
-//! limits (128 KiB / 16 KiB). A local OTLP/HTTP collector observes the existing
-//! reader-path counter without adding a scrape endpoint or production logging.
+//! the default EC 2+2 geometry), and the GET fast path follows the persisted
+//! inline marker. A local OTLP/HTTP collector observes the existing reader-path
+//! counter without adding a scrape endpoint or production logging.
 //! One S3 GET can select readers on multiple EC nodes, so the counter tracks
 //! distributed reader selection rather than HTTP request count.

-use crate::common::{RustFSTestClusterEnvironment, RustFSTestEnvironment, init_logging, local_http_client};
+use crate::common::{RustFSTestClusterEnvironment, RustFSTestEnvironment, init_logging};
 use aws_sdk_s3::Client;
 use aws_sdk_s3::primitives::ByteStream;
 use aws_sdk_s3::types::{
    BucketLifecycleConfiguration, BucketVersioningStatus, CompletedMultipartUpload, CompletedPart, ExpirationStatus,
    LifecycleRule, LifecycleRuleFilter, ServerSideEncryption, Transition, TransitionStorageClass, VersioningConfiguration,
 };
-use base64::Engine;
 use bytes::Bytes;
 use flate2::read::GzDecoder;
-use http::header::{CONTENT_ENCODING, HOST};
+use http::header::CONTENT_ENCODING;
 use http::{Method, Request, Response, StatusCode};
 use http_body_util::{BodyExt, Full};
 use hyper::body::Incoming;
@@ -43,9 +42,6 @@ use opentelemetry_proto::tonic::metrics::v1::{
    Metric, NumberDataPoint, ResourceMetrics, ScopeMetrics, Sum, metric, number_data_point,
 };
 use prost::Message;
-use rustfs_signer::constants::UNSIGNED_PAYLOAD;
-use rustfs_signer::sign_v4;
-use s3s::Body;
 use std::collections::BTreeMap;
 use std::convert::Infallible;
 use std::error::Error;
@@ -93,6 +89,7 @@ const MPU_PART_1_SIZE: usize = 5 * 1024 * 1024;
 const MPU_PART_2_SIZE: usize = 16 * KIB;
 const TIER_BUCKET: &str = "inline-fallback-cold-tier";
 const TIER_PREFIX: &str = "tiered";
+const ALLOW_LOOPBACK_TIER_ENDPOINT_ENV: &str = "RUSTFS_TIER_RUSTFS_ALLOW_LOOPBACK_ENDPOINT";
 const MSGPACK_FALLBACK_CONTROL_SERIES: [(&str, &str); 4] = [
    (FALLBACK_REQUEST_DIRECTION, "ReadMultipleReq"),
    (FALLBACK_RESPONSE_DIRECTION, "ReadMultipleResp"),
@@ -795,12 +792,12 @@ fn metric_attribute(key: &str, value: &str) -> KeyValue {
 }

 fn boundary_cases(state: VersionState) -> Vec<BoundaryCase> {
-    let (fast_limit, storage_limit) = match state {
-        VersionState::Enabled => (16 * KIB, 32 * KIB),
-        VersionState::Unversioned => (128 * KIB, 256 * KIB),
+    let storage_limit = match state {
+        VersionState::Enabled => 32 * KIB,
+        VersionState::Unversioned => 256 * KIB,
        // A suspended bucket stores its null version using the unversioned
        // shard threshold, while ObjectInfo keeps version-aware GET semantics.
-        VersionState::Suspended => (16 * KIB, 256 * KIB),
+        VersionState::Suspended => 256 * KIB,
    };
    let mut sizes = vec![0, 16 * KIB - 1, 16 * KIB, 16 * KIB + 1, 32 * KIB - 1, 32 * KIB, 32 * KIB + 1];
    if !matches!(state, VersionState::Enabled) {
@@ -821,7 +818,7 @@ fn boundary_cases(state: VersionState) -> Vec<BoundaryCase> {
            stored_inline: size <= storage_limit,
            expected_reader_path: if size == 0 {
                EMPTY
-            } else if size <= fast_limit {
+            } else if size <= storage_limit {
                INLINE_DIRECT
            } else {
                LEGACY_DUPLEX
@@ -1263,6 +1260,8 @@ async fn put_two_part_multipart(client: &Client, bucket: &str, key: &str) -> Tes
    Ok((body, part2, complete.e_tag().map(str::to_owned)))
 }

+/// Thin wrapper over [`crate::common::admin_request`], kept local so the call
+/// sites below keep their `Option<&str>` body shape.
 async fn signed_admin_request(
    base_url: &str,
    method: Method,
@@ -1271,30 +1270,7 @@ async fn signed_admin_request(
    access_key: &str,
    secret_key: &str,
 ) -> TestResult<(reqwest::StatusCode, String)> {
-    let url = format!("{base_url}{path}");
-    let uri = url.parse::<http::Uri>()?;
-    let authority = uri.authority().ok_or("request URL missing authority")?.to_string();
-    let body_bytes = body.map(|value| value.as_bytes().to_vec()).unwrap_or_default();
-
-    let request = http::Request::builder()
-        .method(method.clone())
-        .uri(uri)
-        .header(HOST, authority)
-        .header("x-amz-content-sha256", UNSIGNED_PAYLOAD);
-    let signed = sign_v4(request.body(Body::empty())?, 0, access_key, secret_key, "", "us-east-1");
-
-    let client = local_http_client();
-    let mut request_builder = client.request(method, url.as_str());
-    for (name, value) in signed.headers() {
-        request_builder = request_builder.header(name, value);
-    }
-    if !body_bytes.is_empty() {
-        request_builder = request_builder.body(body_bytes);
-    }
-    let response = request_builder.send().await?;
-    let status = response.status();
-    let text = response.text().await?;
-    Ok((status, text))
+    crate::common::admin_request(base_url, method, path, body.map(str::to_string), access_key, secret_key).await
 }

 fn unique_tier_name() -> String {
@@ -1808,7 +1784,7 @@ async fn four_node_inline_fallback_controls() -> TestResult {
    let collector = OtlpMetricCollector::start().await?;
    let mut cluster = RustFSTestClusterEnvironment::new(4).await?;
    configure_reader_metric_cluster(&mut cluster, &collector);
-    let sse_master_key = base64::engine::general_purpose::STANDARD.encode([0x42u8; 32]);
+    let sse_master_key = base64_simd::STANDARD.encode_to_string([0x42u8; 32]);
    cluster.set_env("RUSTFS_SSE_S3_MASTER_KEY", &sse_master_key);
    cluster.start().await?;

@@ -2017,7 +1993,7 @@ async fn four_node_mixed_msgpack_compat_mode_preserves_fallback_controls() -> Te

    let collector = OtlpMetricCollector::start().await?;
    let mut cluster = RustFSTestClusterEnvironment::new(4).await?;
-    let sse_master_key = base64::engine::general_purpose::STANDARD.encode([0x42u8; 32]);
+    let sse_master_key = base64_simd::STANDARD.encode_to_string([0x42u8; 32]);
    cluster.set_env("RUSTFS_SSE_S3_MASTER_KEY", sse_master_key);
    cluster.set_env("RUSTFS_COMPRESSION_ENABLED", "true");
    cluster.set_env("RUSTFS_COMPRESSION_MULTIPART_ENABLED", "true");
@@ -2125,6 +2101,7 @@ async fn four_node_add_tier_converges() -> TestResult {
    cold.create_s3_client().create_bucket().bucket(TIER_BUCKET).send().await?;

    let mut hot = RustFSTestClusterEnvironment::new(4).await?;
+    hot.set_env(ALLOW_LOOPBACK_TIER_ENDPOINT_ENV, "true");
    hot.start().await?;

    let tier_name = unique_tier_name();
@@ -2143,6 +2120,7 @@ async fn four_node_add_tier_converges_after_offline_node_restart_without_second_
    cold.create_s3_client().create_bucket().bucket(TIER_BUCKET).send().await?;

    let mut hot = RustFSTestClusterEnvironment::new(4).await?;
+    hot.set_env(ALLOW_LOOPBACK_TIER_ENDPOINT_ENV, "true");
    hot.start().await?;

    let tier_name = unique_tier_name();
@@ -2239,6 +2217,7 @@ async fn four_node_manual_transition_distributed_admission_conflict_reports_stat
    cold_client.create_bucket().bucket(TIER_BUCKET).send().await?;

    let mut hot = RustFSTestClusterEnvironment::new(4).await?;
+    hot.set_env(ALLOW_LOOPBACK_TIER_ENDPOINT_ENV, "true");
    hot.set_env("RUSTFS_SCANNER_ENABLED", "false");
    hot.set_env("RUSTFS_SCANNER_CYCLE", "3600");
    hot.set_env("RUSTFS_MAX_TRANSITION_WORKERS", "1");
@@ -2381,6 +2360,7 @@ async fn four_node_manual_transition_rollout_non_empty_restart_readback() -> Tes
    cold_client.create_bucket().bucket(TIER_BUCKET).send().await?;

    let mut hot = RustFSTestClusterEnvironment::new(4).await?;
+    hot.set_env(ALLOW_LOOPBACK_TIER_ENDPOINT_ENV, "true");
    hot.set_env("RUSTFS_SCANNER_ENABLED", "false");
    hot.set_env("RUSTFS_SCANNER_CYCLE", "3600");
    hot.set_env("RUSTFS_MAX_TRANSITION_WORKERS", "2");
@@ -2485,11 +2465,12 @@ async fn four_node_mixed_msgpack_compat_mode_preserves_fallback_controls_during_

    let collector = OtlpMetricCollector::start().await?;
    let mut hot = RustFSTestClusterEnvironment::new(4).await?;
+    hot.set_env(ALLOW_LOOPBACK_TIER_ENDPOINT_ENV, "true");
    configure_mixed_msgpack_cluster(&mut hot, &collector)?;
    hot.set_env("RUSTFS_SCANNER_CYCLE", "1");
    hot.set_env("RUSTFS_ILM_PROCESS_TIME", "1");

-    let sse_master_key = base64::engine::general_purpose::STANDARD.encode([0x42u8; 32]);
+    let sse_master_key = base64_simd::STANDARD.encode_to_string([0x42u8; 32]);
    hot.set_env("RUSTFS_SSE_S3_MASTER_KEY", sse_master_key);
    hot.set_env("RUSTFS_COMPRESSION_ENABLED", "true");
    hot.start().await?;
@@ -2596,6 +2577,7 @@ async fn four_node_transitioned_inline_fallback() -> TestResult {

    let collector = OtlpMetricCollector::start().await?;
    let mut hot = RustFSTestClusterEnvironment::new(4).await?;
+    hot.set_env(ALLOW_LOOPBACK_TIER_ENDPOINT_ENV, "true");
    configure_reader_metric_cluster(&mut hot, &collector);
    hot.set_env("RUSTFS_SCANNER_CYCLE", "1");
    hot.set_env("RUSTFS_ILM_PROCESS_TIME", "1");
@@ -24,9 +24,10 @@

 use crate::common::{RustFSTestEnvironment, awscurl_get, awscurl_post, init_logging as common_init_logging, local_http_client};
 use aws_sdk_s3::Client;
+use aws_sdk_s3::error::{ProvideErrorMetadata, SdkError};
 use aws_sdk_s3::primitives::ByteStream;
 use aws_sdk_s3::types::ServerSideEncryption;
-use base64::{Engine, engine::general_purpose::STANDARD as BASE64};
+use base64_simd::STANDARD as BASE64;
 use http::header::{CONTENT_TYPE, HOST};
 use md5::{Digest as Md5Digest, Md5};
 use rustfs_signer::constants::UNSIGNED_PAYLOAD;
@@ -50,6 +51,9 @@ pub const VAULT_TOKEN: &str = "dev-root-token";
 pub const VAULT_TRANSIT_PATH: &str = "transit";
 pub const VAULT_KEY_NAME: &str = "rustfs-master-key";
 pub const ENV_TEST_VAULT_BIN: &str = "RUSTFS_TEST_VAULT_BIN";
+pub const SSE_C_KEY_MISMATCH_MESSAGE: &str =
+    "The provided encryption parameters did not match the ones used originally to encrypt the object.";
+pub const SSE_C_MISSING_PARAMETERS_MESSAGE: &str = "The object was stored using a form of Server Side Encryption. The correct parameters must be provided to retrieve the object.";

 /// Initialize tracing for KMS tests with KMS-specific log levels
 pub fn init_logging() {
@@ -60,7 +64,25 @@ pub fn init_logging() {
 pub fn sse_customer_key_md5_base64(key: &str) -> String {
    let mut hasher = Md5::new();
    hasher.update(key.as_bytes());
-    BASE64.encode(hasher.finalize())
+    BASE64.encode_to_string(hasher.finalize())
+}
+
+pub fn assert_s3_error<T, E>(result: Result<T, SdkError<E>>, status: u16, code: &str, message: &str, context: &str)
+where
+    T: std::fmt::Debug,
+    E: ProvideErrorMetadata + std::fmt::Debug,
+{
+    let error = result.expect_err(context);
+    assert_eq!(
+        error.raw_response().map(|response| response.status().as_u16()),
+        Some(status),
+        "{context}: unexpected HTTP status: {error:?}"
+    );
+    let service_error = error
+        .as_service_error()
+        .expect("request failure should retain an S3 service error");
+    assert_eq!(service_error.code(), Some(code), "{context}: unexpected error code: {error:?}");
+    assert_eq!(service_error.message(), Some(message), "{context}: unexpected error message: {error:?}");
 }

 pub async fn kms_admin_request(
@@ -343,7 +365,7 @@ pub async fn create_key_with_specific_id(key_dir: &str, key_id: &str) -> Result<
        "created_at": format!("{}[UTC]", chrono::Utc::now().to_rfc3339()),
        "rotated_at": serde_json::Value::Null,
        "created_by": "e2e-test",
-        "encrypted_key_material": BASE64.encode(key_data),
+        "encrypted_key_material": BASE64.encode_to_string(key_data),
        "nonce": Vec::<u8>::new()
    });

@@ -361,7 +383,7 @@ pub async fn test_sse_c_encryption(s3_client: &Client, bucket: &str) -> Result<(
    info!("Testing SSE-C encryption");

    let test_key = "01234567890123456789012345678901"; // 32-byte key
-    let test_key_b64 = base64::engine::general_purpose::STANDARD.encode(test_key);
+    let test_key_b64 = base64_simd::STANDARD.encode_to_string(test_key);
    let test_key_md5 = sse_customer_key_md5_base64(test_key);
    let test_data = b"Hello, KMS SSE-C World!";
    let object_key = "test-sse-c-object";
@@ -529,8 +551,8 @@ pub async fn test_error_scenarios(s3_client: &Client, bucket: &str) -> Result<()
    // Test SSE-C with wrong key for download
    let test_key = "01234567890123456789012345678901";
    let wrong_key = "98765432109876543210987654321098";
-    let test_key_b64 = base64::engine::general_purpose::STANDARD.encode(test_key);
-    let wrong_key_b64 = base64::engine::general_purpose::STANDARD.encode(wrong_key);
+    let test_key_b64 = base64_simd::STANDARD.encode_to_string(test_key);
+    let wrong_key_b64 = base64_simd::STANDARD.encode_to_string(wrong_key);
    let test_key_md5 = sse_customer_key_md5_base64(test_key);
    let wrong_key_md5 = sse_customer_key_md5_base64(wrong_key);
    let test_data = b"Test data for error scenarios";
@@ -559,7 +581,13 @@ pub async fn test_error_scenarios(s3_client: &Client, bucket: &str) -> Result<()
        .send()
        .await;

-    assert!(wrong_key_result.is_err(), "Download with wrong SSE-C key should fail");
+    assert_s3_error(
+        wrong_key_result,
+        400,
+        "InvalidRequest",
+        SSE_C_KEY_MISMATCH_MESSAGE,
+        "download with a wrong SSE-C key must be rejected",
+    );
    info!("✅ Correctly rejected download with wrong SSE-C key");

    info!("Error scenario tests completed successfully");
@@ -779,7 +807,7 @@ pub async fn test_multipart_upload_with_config(
    // Prepare encryption parameters
    let (sse_c_key_b64, sse_c_key_md5) = match &config.encryption_type {
        EncryptionType::SSEC { key, key_md5 } => {
-            let key_b64 = base64::engine::general_purpose::STANDARD.encode(key);
+            let key_b64 = base64_simd::STANDARD.encode_to_string(key);
            (Some(key_b64), Some(key_md5.clone()))
        }
        _ => (None, None),
@@ -432,7 +432,6 @@ async fn test_configured_local_kms_admin_and_versioned_cleanup() -> TestResult {
 }

 #[tokio::test]
-#[ignore = "requires a Vault binary"]
 async fn test_configured_vault_kms_admin_and_versioned_cleanup() -> TestResult {
    let mut env = VaultTestEnvironment::new().await?;
    env.start_vault().await?;
@@ -0,0 +1,140 @@
+// Copyright 2024 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+//! Ranged GETs over encrypted single-part objects.
+//!
+//! Byte-exactness must hold on every frame layout the server can write:
+//! legacy v1 (variable frames, conservative full read) and, when
+//! `RUSTFS_ENCRYPTION_FRAME_V2=true` reaches the server under test, the
+//! fixed-frame v2 layout whose marker enables the closed-form frame seek.
+//! The matrix crosses frame boundaries, starts mid-frame, and ends inside
+//! the final short frame, so a mispositioned seek cannot pass.
+
+use super::common::LocalKMSTestEnvironment;
+use crate::common::{TEST_BUCKET, init_logging};
+use aws_sdk_s3::primitives::ByteStream;
+use aws_sdk_s3::types::ServerSideEncryption;
+use tracing::info;
+
+const FRAME_PLAINTEXT: usize = 8 * 1024;
+
+#[tokio::test]
+async fn sse_s3_single_part_ranged_gets_are_byte_exact() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+    init_logging();
+    info!("Testing ranged GETs over an SSE-S3 single-part object");
+
+    let mut kms_env = LocalKMSTestEnvironment::new().await?;
+    let _default_key_id = kms_env.start_rustfs_for_local_kms().await?;
+    kms_env.wait_for_kms_ready().await?;
+
+    let s3_client = kms_env.base_env.create_s3_client();
+    kms_env.base_env.create_test_bucket(TEST_BUCKET).await?;
+
+    let test_key = "encrypted-range-get";
+    let body: Vec<u8> = (0..3 * FRAME_PLAINTEXT + 500).map(|i| (i % 251) as u8).collect();
+
+    let put = s3_client
+        .put_object()
+        .bucket(TEST_BUCKET)
+        .key(test_key)
+        .server_side_encryption(ServerSideEncryption::Aes256)
+        .body(ByteStream::from(body.clone()))
+        .send()
+        .await?;
+    assert_eq!(
+        put.server_side_encryption(),
+        Some(&ServerSideEncryption::Aes256),
+        "the object under test must actually be encrypted"
+    );
+
+    let cases: &[(usize, usize)] = &[
+        // Head range inside frame 0.
+        (0, 99),
+        // Crossing the first frame boundary.
+        (FRAME_PLAINTEXT - 1, FRAME_PLAINTEXT),
+        // Starting exactly on a frame boundary.
+        (FRAME_PLAINTEXT, FRAME_PLAINTEXT + 9),
+        // Mid-object, mid-frame on both ends.
+        (2 * FRAME_PLAINTEXT + 5, 3 * FRAME_PLAINTEXT + 100),
+        // Tail range ending inside the final short frame.
+        (3 * FRAME_PLAINTEXT + 100, 3 * FRAME_PLAINTEXT + 499),
+    ];
+
+    for &(start, end) in cases {
+        let response = s3_client
+            .get_object()
+            .bucket(TEST_BUCKET)
+            .key(test_key)
+            .range(format!("bytes={start}-{end}"))
+            .send()
+            .await?;
+        assert_eq!(
+            response.content_length(),
+            Some((end - start + 1) as i64),
+            "range {start}-{end} content length"
+        );
+        let data = response.body.collect().await?.into_bytes();
+        assert_eq!(data.as_ref(), &body[start..=end], "range {start}-{end} must be byte-exact");
+    }
+
+    // A suffix range exercises the offset resolution path as well.
+    let response = s3_client
+        .get_object()
+        .bucket(TEST_BUCKET)
+        .key(test_key)
+        .range("bytes=-123")
+        .send()
+        .await?;
+    let data = response.body.collect().await?.into_bytes();
+    assert_eq!(data.as_ref(), &body[body.len() - 123..], "suffix range must be byte-exact");
+
+    // The unranged body still round-trips.
+    let response = s3_client.get_object().bucket(TEST_BUCKET).key(test_key).send().await?;
+    let data = response.body.collect().await?.into_bytes();
+    assert_eq!(data.as_ref(), body.as_slice(), "full body must round-trip");
+
+    // A block-aligned object ends in an empty authenticated final frame under
+    // the v2 layout; tail ranges touching the last plaintext byte must not be
+    // misread as truncation.
+    let aligned_key = "encrypted-range-get-aligned";
+    let aligned_body: Vec<u8> = (0..3 * FRAME_PLAINTEXT).map(|i| ((i + 3) % 251) as u8).collect();
+    s3_client
+        .put_object()
+        .bucket(TEST_BUCKET)
+        .key(aligned_key)
+        .server_side_encryption(ServerSideEncryption::Aes256)
+        .body(ByteStream::from(aligned_body.clone()))
+        .send()
+        .await?;
+    for (start, end) in [
+        (2 * FRAME_PLAINTEXT + 10, 3 * FRAME_PLAINTEXT - 1),
+        (3 * FRAME_PLAINTEXT - 1, 3 * FRAME_PLAINTEXT - 1),
+    ] {
+        let response = s3_client
+            .get_object()
+            .bucket(TEST_BUCKET)
+            .key(aligned_key)
+            .range(format!("bytes={start}-{end}"))
+            .send()
+            .await?;
+        let data = response.body.collect().await?.into_bytes();
+        assert_eq!(
+            data.as_ref(),
+            &aligned_body[start..=end],
+            "aligned range {start}-{end} must be byte-exact"
+        );
+    }
+
+    Ok(())
+}
@@ -0,0 +1,220 @@
+// Copyright 2024 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+//! Anonymous access to SSE-KMS objects under per-key authorization.
+//!
+//! Locks both halves of the anonymous contract decided in backlog#2028 (D4):
+//!
+//! - **Enforcement on**: anonymous requests hold no `kms` grants, so a public
+//!   bucket policy does not let them read SSE-KMS objects or write through an
+//!   SSE-KMS default-encryption rule. Both fail with `AccessDenied`.
+//! - **Enforcement off** (the default): bucket policy alone governs anonymous
+//!   access, matching the pre-enforcement behavior — public SSE-KMS objects are
+//!   decrypted and served, and anonymous writes are encrypted under the default
+//!   key.
+//!
+//! The denial today is emergent — an empty-account principal falling through to
+//! the IAM default deny — so without this file a refactor of principal
+//! construction or policy evaluation could silently flip it. Each test carries a
+//! plaintext-object positive control: a denial proves nothing while the bucket
+//! policy has not propagated.
+
+use super::common::{LocalKMSTestEnvironment, create_key_with_specific_id};
+use crate::common::{init_logging, local_http_client};
+use aws_sdk_s3::primitives::ByteStream;
+use aws_sdk_s3::types::{
+    ServerSideEncryption, ServerSideEncryptionByDefault, ServerSideEncryptionConfiguration, ServerSideEncryptionRule,
+};
+use std::time::Duration;
+
+type TestResult = Result<(), Box<dyn std::error::Error + Send + Sync>>;
+
+const DEFAULT_KEY: &str = "kms-anon-default-key";
+const BUCKET: &str = "kms-anon-enforcement";
+const PLAIN_OBJECT: &str = "plain.txt";
+const ENCRYPTED_OBJECT: &str = "encrypted.txt";
+const PAYLOAD: &[u8] = b"kms anonymous enforcement payload";
+
+/// How long a bucket policy change may take to reach the request path.
+const POLICY_PROPAGATION: Duration = Duration::from_secs(20);
+
+/// Start a local-KMS server and build the public-bucket fixture.
+///
+/// The bucket holds a plaintext object (the positive control), an SSE-KMS
+/// object, an SSE-KMS default-encryption rule, and a bucket policy opening
+/// `GetObject`/`PutObject` to everyone. The enforcement switch defaults to off,
+/// so the enforcing case has to set it explicitly.
+async fn start_public_sse_kms_bucket(env: &mut LocalKMSTestEnvironment, enforce: bool) -> TestResult {
+    create_key_with_specific_id(&env.kms_keys_dir, DEFAULT_KEY).await?;
+
+    let key_dir = env.kms_keys_dir.clone();
+    let args = vec![
+        "--kms-enable",
+        "--kms-backend",
+        "local",
+        "--kms-key-dir",
+        key_dir.as_str(),
+        "--kms-default-key-id",
+        DEFAULT_KEY,
+    ];
+    let mut envs = vec![("RUSTFS_KMS_ALLOW_INSECURE_DEV_DEFAULTS", "true")];
+    if enforce {
+        envs.push(("RUSTFS_KMS_ENFORCE_SSE_KEY_POLICY", "true"));
+    }
+    env.base_env.start_rustfs_server_with_env(args, &envs).await?;
+    env.base_env.create_test_bucket(BUCKET).await?;
+
+    let owner = env.base_env.create_s3_client();
+
+    owner
+        .put_object()
+        .bucket(BUCKET)
+        .key(PLAIN_OBJECT)
+        .body(ByteStream::from_static(PAYLOAD))
+        .send()
+        .await?;
+    owner
+        .put_object()
+        .bucket(BUCKET)
+        .key(ENCRYPTED_OBJECT)
+        .body(ByteStream::from_static(PAYLOAD))
+        .server_side_encryption(ServerSideEncryption::AwsKms)
+        .ssekms_key_id(DEFAULT_KEY)
+        .send()
+        .await?;
+
+    let encryption_config = ServerSideEncryptionConfiguration::builder()
+        .rules(
+            ServerSideEncryptionRule::builder()
+                .apply_server_side_encryption_by_default(
+                    ServerSideEncryptionByDefault::builder()
+                        .sse_algorithm(ServerSideEncryption::AwsKms)
+                        .kms_master_key_id(DEFAULT_KEY)
+                        .build()?,
+                )
+                .build(),
+        )
+        .build()?;
+    owner
+        .put_bucket_encryption()
+        .bucket(BUCKET)
+        .server_side_encryption_configuration(encryption_config)
+        .send()
+        .await?;
+
+    let policy = serde_json::json!({
+        "Version": "2012-10-17",
+        "Statement": [{
+            "Sid": "PublicReadWrite",
+            "Effect": "Allow",
+            "Principal": "*",
+            "Action": ["s3:GetObject", "s3:PutObject"],
+            "Resource": [format!("arn:aws:s3:::{BUCKET}/*")]
+        }]
+    })
+    .to_string();
+    owner.put_bucket_policy().bucket(BUCKET).policy(&policy).send().await?;
+    let _ = owner.delete_public_access_block().bucket(BUCKET).send().await;
+
+    Ok(())
+}
+
+fn object_url(env: &LocalKMSTestEnvironment, key: &str) -> String {
+    format!("{}/{BUCKET}/{key}", env.base_env.url)
+}
+
+async fn anonymous_get(env: &LocalKMSTestEnvironment, key: &str) -> Result<reqwest::Response, reqwest::Error> {
+    local_http_client().get(object_url(env, key)).send().await
+}
+
+async fn anonymous_put(env: &LocalKMSTestEnvironment, key: &str) -> Result<reqwest::Response, reqwest::Error> {
+    local_http_client().put(object_url(env, key)).body(PAYLOAD).send().await
+}
+
+/// Retry the plaintext read until the public bucket policy is live.
+async fn wait_for_public_read(env: &LocalKMSTestEnvironment) -> TestResult {
+    let deadline = tokio::time::Instant::now() + POLICY_PROPAGATION;
+    loop {
+        let status = anonymous_get(env, PLAIN_OBJECT).await?.status();
+        if status.as_u16() == 200 {
+            return Ok(());
+        }
+        if tokio::time::Instant::now() >= deadline {
+            return Err(format!("positive control never became readable: anonymous GET {PLAIN_OBJECT} -> {status}").into());
+        }
+        tokio::time::sleep(Duration::from_millis(500)).await;
+    }
+}
+
+async fn assert_anonymous_denied(response: reqwest::Response, what: &str) -> TestResult {
+    let status = response.status().as_u16();
+    let body = response.text().await?;
+    assert_eq!(status, 403, "{what} must be denied, got {status}: {body}");
+    assert!(body.contains("AccessDenied"), "{what} must carry AccessDenied: {body}");
+    Ok(())
+}
+
+/// Enforcement on: a public bucket policy does not exempt anonymous requests
+/// from per-key authorization, on either the read or the default-encryption
+/// write path.
+#[tokio::test(flavor = "multi_thread")]
+async fn anonymous_sse_kms_denied_under_enforcement() -> TestResult {
+    init_logging();
+
+    let mut env = LocalKMSTestEnvironment::new().await?;
+    start_public_sse_kms_bucket(&mut env, true).await?;
+    wait_for_public_read(&env).await?;
+
+    let read = anonymous_get(&env, ENCRYPTED_OBJECT).await?;
+    assert_anonymous_denied(read, "anonymous GET of an SSE-KMS object").await?;
+
+    let write = anonymous_put(&env, "anon-write.txt").await?;
+    assert_anonymous_denied(write, "anonymous PUT through an SSE-KMS default-encryption rule").await?;
+
+    Ok(())
+}
+
+/// Enforcement off (the default): bucket policy alone governs anonymous access,
+/// and the default-encryption rule still encrypts anonymous writes.
+#[tokio::test(flavor = "multi_thread")]
+async fn anonymous_sse_kms_governed_by_bucket_policy_without_enforcement() -> TestResult {
+    init_logging();
+
+    let mut env = LocalKMSTestEnvironment::new().await?;
+    start_public_sse_kms_bucket(&mut env, false).await?;
+    wait_for_public_read(&env).await?;
+
+    let read = anonymous_get(&env, ENCRYPTED_OBJECT).await?;
+    assert_eq!(read.status().as_u16(), 200, "anonymous GET of a public SSE-KMS object must succeed");
+    assert_eq!(read.bytes().await?.as_ref(), PAYLOAD, "the object must be served decrypted");
+
+    let write = anonymous_put(&env, "anon-write.txt").await?;
+    assert_eq!(write.status().as_u16(), 200, "anonymous PUT to a public bucket must succeed");
+
+    let stored = env
+        .base_env
+        .create_s3_client()
+        .head_object()
+        .bucket(BUCKET)
+        .key("anon-write.txt")
+        .send()
+        .await?;
+    assert_eq!(
+        stored.server_side_encryption(),
+        Some(&ServerSideEncryption::AwsKms),
+        "the anonymous write must be encrypted by the bucket default rule"
+    );
+
+    Ok(())
+}
@@ -19,9 +19,9 @@
 //! complex workflows.

 use super::common::{
-    EncryptionType, LocalKMSTestEnvironment, MultipartTestConfig, create_sse_c_config, sse_customer_key_md5_base64,
-    test_all_multipart_encryption_types, test_kms_key_management, test_multipart_upload_with_config, test_sse_c_encryption,
-    test_sse_kms_encryption, test_sse_s3_encryption,
+    EncryptionType, LocalKMSTestEnvironment, MultipartTestConfig, SSE_C_KEY_MISMATCH_MESSAGE, assert_s3_error,
+    create_sse_c_config, sse_customer_key_md5_base64, test_all_multipart_encryption_types, test_kms_key_management,
+    test_multipart_upload_with_config, test_sse_c_encryption, test_sse_kms_encryption, test_sse_s3_encryption,
 };
 use crate::common::{TEST_BUCKET, init_logging};
 use tracing::info;
@@ -177,7 +177,7 @@ async fn test_comprehensive_key_isolation() -> Result<(), Box<dyn std::error::Er
    // Verify that files cannot be read with wrong keys
    info!("🔒 Verify key isolation");
    let wrong_key = "11111111111111111111111111111111";
-    let wrong_key_b64 = base64::Engine::encode(&base64::engine::general_purpose::STANDARD, wrong_key);
+    let wrong_key_b64 = base64_simd::STANDARD.encode_to_string(wrong_key);
    let wrong_key_md5 = sse_customer_key_md5_base64(wrong_key);

    // Try to read file encrypted with key1 using wrong key
@@ -191,7 +191,13 @@ async fn test_comprehensive_key_isolation() -> Result<(), Box<dyn std::error::Er
        .send()
        .await;

-    assert!(wrong_read_result.is_err(), "The encrypted file should not be readable with the wrong key");
+    assert_s3_error(
+        wrong_read_result,
+        400,
+        "InvalidRequest",
+        SSE_C_KEY_MISMATCH_MESSAGE,
+        "multipart SSE-C object GET with a wrong key must be rejected",
+    );
    info!("✅ Confirm that key isolation is working correctly");

    kms_env.base_env.delete_test_bucket(TEST_BUCKET).await?;
@@ -21,21 +21,13 @@
 //! - Concurrent encryption operations
 //! - Security validation tests

-use super::common::{LocalKMSTestEnvironment, sse_customer_key_md5_base64};
+use super::common::{LocalKMSTestEnvironment, SSE_C_KEY_MISMATCH_MESSAGE, assert_s3_error, sse_customer_key_md5_base64};
 use crate::common::{TEST_BUCKET, init_logging};
 use aws_sdk_s3::types::ServerSideEncryption;
-use base64::Engine;
-use md5::{Digest as Md5Digest, Md5};
 use std::sync::Arc;
 use tokio::sync::Semaphore;
 use tracing::{info, warn};

-fn md5_hex(input: impl AsRef<[u8]>) -> String {
-    let mut hasher = Md5::new();
-    hasher.update(input.as_ref());
-    hex::encode(hasher.finalize())
-}
-
 /// Test encryption of zero-byte files (empty files)
 #[tokio::test]
 async fn test_kms_zero_byte_file_encryption() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
@@ -75,7 +67,7 @@ async fn test_kms_zero_byte_file_encryption() -> Result<(), Box<dyn std::error::
    // Test SSE-C with zero-byte file
    info!("📤 Testing SSE-C with zero-byte file");
    let test_key = "01234567890123456789012345678901";
-    let test_key_b64 = base64::engine::general_purpose::STANDARD.encode(test_key);
+    let test_key_b64 = base64_simd::STANDARD.encode_to_string(test_key);
    let test_key_md5 = sse_customer_key_md5_base64(test_key);
    let object_key_c = "zero-byte-sse-c";

@@ -168,7 +160,7 @@ async fn test_kms_single_byte_file_encryption() -> Result<(), Box<dyn std::error
    // Test SSE-C with single byte
    info!("📤 Testing SSE-C with single-byte file");
    let test_key = "01234567890123456789012345678901";
-    let test_key_b64 = base64::engine::general_purpose::STANDARD.encode(test_key);
+    let test_key_b64 = base64_simd::STANDARD.encode_to_string(test_key);
    let test_key_md5 = sse_customer_key_md5_base64(test_key);
    let object_key_c = "single-byte-sse-c";

@@ -294,8 +286,8 @@ async fn test_kms_invalid_key_scenarios() -> Result<(), Box<dyn std::error::Erro
    // Test 1: Invalid key length for SSE-C
    info!("🔍 Testing invalid SSE-C key length");
    let invalid_short_key = "short"; // Too short
-    let invalid_key_b64 = base64::engine::general_purpose::STANDARD.encode(invalid_short_key);
-    let invalid_key_md5 = md5_hex(invalid_short_key);
+    let invalid_key_b64 = base64_simd::STANDARD.encode_to_string(invalid_short_key);
+    let invalid_key_md5 = sse_customer_key_md5_base64(invalid_short_key);

    let invalid_key_result = s3_client
        .put_object()
@@ -308,14 +300,32 @@ async fn test_kms_invalid_key_scenarios() -> Result<(), Box<dyn std::error::Erro
        .send()
        .await;

-    assert!(invalid_key_result.is_err(), "Should reject invalid key length");
+    assert_s3_error(
+        invalid_key_result,
+        400,
+        "InvalidRequest",
+        "SSE-C key must be 32 bytes (256 bits), got 5 bytes.",
+        "invalid SSE-C key length must be rejected",
+    );
+    assert_s3_error(
+        s3_client
+            .get_object()
+            .bucket(TEST_BUCKET)
+            .key("test-invalid-key-length")
+            .send()
+            .await,
+        404,
+        "NoSuchKey",
+        "The specified key does not exist.",
+        "rejected invalid-key PUT must not create an object",
+    );
    info!("✅ Correctly rejected invalid key length");

    // Test 2: Mismatched MD5 for SSE-C
    info!("🔍 Testing mismatched MD5 for SSE-C key");
    let valid_key = "01234567890123456789012345678901";
-    let valid_key_b64 = base64::engine::general_purpose::STANDARD.encode(valid_key);
-    let wrong_md5 = "wrongmd5hash12345678901234567890"; // Wrong MD5
+    let valid_key_b64 = base64_simd::STANDARD.encode_to_string(valid_key);
+    let wrong_md5 = sse_customer_key_md5_base64("98765432109876543210987654321098");

    let wrong_md5_result = s3_client
        .put_object()
@@ -324,11 +334,24 @@ async fn test_kms_invalid_key_scenarios() -> Result<(), Box<dyn std::error::Erro
        .body(aws_sdk_s3::primitives::ByteStream::from(test_data.to_vec()))
        .sse_customer_algorithm("AES256")
        .sse_customer_key(&valid_key_b64)
-        .sse_customer_key_md5(wrong_md5)
+        .sse_customer_key_md5(&wrong_md5)
        .send()
        .await;

-    assert!(wrong_md5_result.is_err(), "Should reject mismatched MD5");
+    assert_s3_error(
+        wrong_md5_result,
+        400,
+        "InvalidRequest",
+        "The calculated MD5 hash of the key did not match the hash that was provided.",
+        "mismatched SSE-C key MD5 must be rejected",
+    );
+    assert_s3_error(
+        s3_client.get_object().bucket(TEST_BUCKET).key("test-wrong-md5").send().await,
+        404,
+        "NoSuchKey",
+        "The specified key does not exist.",
+        "rejected mismatched-MD5 PUT must not create an object",
+    );
    info!("✅ Correctly rejected mismatched MD5");

    // Test 3: Try to access SSE-C object without providing key
@@ -355,7 +378,28 @@ async fn test_kms_invalid_key_scenarios() -> Result<(), Box<dyn std::error::Erro
        .send()
        .await;

-    assert!(no_key_result.is_err(), "Should require SSE-C key for access");
+    assert_s3_error(
+        no_key_result,
+        400,
+        "InvalidRequest",
+        "The object was stored using a form of Server Side Encryption. The correct parameters must be provided to retrieve the object.",
+        "SSE-C object GET without a customer key must be rejected",
+    );
+
+    let recovered = s3_client
+        .get_object()
+        .bucket(TEST_BUCKET)
+        .key("test-sse-c-no-key-access")
+        .sse_customer_algorithm("AES256")
+        .sse_customer_key(&valid_key_b64)
+        .sse_customer_key_md5(&valid_key_md5)
+        .send()
+        .await?
+        .body
+        .collect()
+        .await?
+        .into_bytes();
+    assert_eq!(recovered.as_ref(), test_data, "failed GET must not corrupt the SSE-C object");
    info!("✅ Correctly required SSE-C key for access");

    kms_env.base_env.delete_test_bucket(TEST_BUCKET).await?;
@@ -420,7 +464,7 @@ async fn test_kms_concurrent_encryption() -> Result<(), Box<dyn std::error::Erro
                2 => {
                    // SSE-C
                    let key = format!("testkey{i:026}"); // 32-byte key
-                    let key_b64 = base64::engine::general_purpose::STANDARD.encode(&key);
+                    let key_b64 = base64_simd::STANDARD.encode_to_string(&key);
                    let key_md5 = sse_customer_key_md5_base64(&key);

                    client
@@ -490,8 +534,8 @@ async fn test_kms_key_validation_security() -> Result<(), Box<dyn std::error::Er
    let key1 = "key1key1key1key1key1key1key1key1"; // 32 bytes
    let key2 = "key2key2key2key2key2key2key2key2"; // 32 bytes

-    let key1_b64 = base64::engine::general_purpose::STANDARD.encode(key1);
-    let key2_b64 = base64::engine::general_purpose::STANDARD.encode(key2);
+    let key1_b64 = base64_simd::STANDARD.encode_to_string(key1);
+    let key2_b64 = base64_simd::STANDARD.encode_to_string(key2);
    let key1_md5 = sse_customer_key_md5_base64(key1);
    let key2_md5 = sse_customer_key_md5_base64(key2);

@@ -563,7 +607,13 @@ async fn test_kms_key_validation_security() -> Result<(), Box<dyn std::error::Er
        .send()
        .await;

-    assert!(wrong_key_result.is_err(), "Should not be able to decrypt with wrong key");
+    assert_s3_error(
+        wrong_key_result,
+        400,
+        "InvalidRequest",
+        SSE_C_KEY_MISMATCH_MESSAGE,
+        "SSE-C object GET with the wrong customer key must be rejected",
+    );
    info!("✅ Key isolation verified - wrong key cannot decrypt data");

    kms_env.base_env.delete_test_bucket(TEST_BUCKET).await?;
@@ -23,6 +23,7 @@

 use super::common::LocalKMSTestEnvironment;
 use crate::common::{TEST_BUCKET, init_logging};
+use aws_sdk_s3::error::ProvideErrorMetadata;
 use aws_sdk_s3::types::ServerSideEncryption;
 use std::fs;
 use std::time::Duration;
@@ -77,12 +78,25 @@ async fn test_kms_key_directory_unavailable() -> Result<(), Box<dyn std::error::
        .send()
        .await;

-    // This should fail, but the server should still be responsive
-    if put_result2.is_err() {
-        info!("✅ Upload correctly failed when key directory unavailable");
-    } else {
-        warn!("⚠️ Upload succeeded despite unavailable key directory (may be using cached keys)");
-    }
+    let unavailable_error = put_result2.expect_err("a missing Local KMS key directory must reject encrypted writes");
+    assert_eq!(unavailable_error.raw_response().map(|response| response.status().as_u16()), Some(500));
+    assert_eq!(
+        unavailable_error.as_service_error().and_then(ProvideErrorMetadata::code),
+        Some("InternalError")
+    );
+    let unavailable_absence = s3_client
+        .get_object()
+        .bucket(TEST_BUCKET)
+        .key(object_key2)
+        .send()
+        .await
+        .expect_err("a write rejected by unavailable KMS must not publish an object");
+    assert_eq!(unavailable_absence.raw_response().map(|response| response.status().as_u16()), Some(404));
+    assert_eq!(
+        unavailable_absence.as_service_error().and_then(ProvideErrorMetadata::code),
+        Some("NoSuchKey")
+    );
+    info!("✅ Upload correctly failed when key directory unavailable");

    // Restore the key directory
    info!("🔧 Restoring key directory");
@@ -107,6 +121,11 @@ async fn test_kms_key_directory_unavailable() -> Result<(), Box<dyn std::error::

    assert_eq!(put_response3.server_side_encryption(), Some(&ServerSideEncryption::Aes256));

+    let get_response3 = s3_client.get_object().bucket(TEST_BUCKET).key(object_key3).send().await?;
+    assert_eq!(get_response3.server_side_encryption(), Some(&ServerSideEncryption::Aes256));
+    let downloaded_data3 = get_response3.body.collect().await?.into_bytes();
+    assert_eq!(downloaded_data3.as_ref(), test_data3);
+
    // Verify we can still access the original file
    info!("📥 Verifying access to original encrypted file");
    let get_response = s3_client.get_object().bucket(TEST_BUCKET).key(object_key).send().await?;
@@ -174,12 +193,22 @@ async fn test_kms_corrupted_key_files() -> Result<(), Box<dyn std::error::Error
        .send()
        .await;

-    // This might succeed if KMS uses cached keys, but should eventually fail
-    if put_result2.is_err() {
-        info!("✅ Upload correctly failed with corrupted key");
-    } else {
-        warn!("⚠️ Upload succeeded despite corrupted key (likely using cached key)");
-    }
+    let corrupt_error = put_result2.expect_err("corrupt Local KMS key material must reject encrypted writes");
+    assert_eq!(corrupt_error.raw_response().map(|response| response.status().as_u16()), Some(500));
+    assert_eq!(
+        corrupt_error.as_service_error().and_then(ProvideErrorMetadata::code),
+        Some("InternalError")
+    );
+    let corrupt_absence = s3_client
+        .get_object()
+        .bucket(TEST_BUCKET)
+        .key(object_key2)
+        .send()
+        .await
+        .expect_err("a write rejected by corrupt KMS material must not publish an object");
+    assert_eq!(corrupt_absence.raw_response().map(|response| response.status().as_u16()), Some(404));
+    assert_eq!(corrupt_absence.as_service_error().and_then(ProvideErrorMetadata::code), Some("NoSuchKey"));
+    info!("✅ Upload correctly failed with corrupted key");

    // Restore the original key file
    info!("🔧 Restoring original key file");
@@ -205,6 +234,11 @@ async fn test_kms_corrupted_key_files() -> Result<(), Box<dyn std::error::Error

    assert_eq!(put_response3.server_side_encryption(), Some(&ServerSideEncryption::Aes256));

+    let get_response3 = s3_client.get_object().bucket(TEST_BUCKET).key(object_key3).send().await?;
+    assert_eq!(get_response3.server_side_encryption(), Some(&ServerSideEncryption::Aes256));
+    let downloaded_data3 = get_response3.body.collect().await?.into_bytes();
+    assert_eq!(downloaded_data3.as_ref(), test_data3);
+
    kms_env.base_env.delete_test_bucket(TEST_BUCKET).await?;
    info!("✅ Corrupted key files test completed successfully");
    Ok(())
@@ -280,18 +314,14 @@ async fn test_kms_multipart_upload_interruption() -> Result<(), Box<dyn std::err
    info!("🔧 Simulating upload interruption");

    // Abort the multipart upload
-    let abort_result = s3_client
+    s3_client
        .abort_multipart_upload()
        .bucket(TEST_BUCKET)
        .key(object_key)
        .upload_id(upload_id)
        .send()
-        .await;
-
-    match abort_result {
-        Ok(_) => info!("✅ Multipart upload aborted successfully"),
-        Err(e) => warn!("⚠️ Failed to abort multipart upload: {}", e),
-    }
+        .await?;
+    info!("✅ Multipart upload aborted successfully");

    // Try to complete the aborted upload - this should fail
    info!("🔍 Attempting to complete aborted upload");
@@ -310,18 +340,38 @@ async fn test_kms_multipart_upload_interruption() -> Result<(), Box<dyn std::err
        .set_parts(Some(completed_parts))
        .build();

-    let complete_result = s3_client
+    let complete_error = s3_client
        .complete_multipart_upload()
        .bucket(TEST_BUCKET)
        .key(object_key)
        .upload_id(upload_id)
        .multipart_upload(completed_multipart_upload)
        .send()
-        .await;
-
-    assert!(complete_result.is_err(), "Should not be able to complete aborted upload");
+        .await
+        .expect_err("an aborted multipart upload must not be completable");
+    assert_eq!(complete_error.raw_response().map(|response| response.status().as_u16()), Some(404));
+    assert_eq!(
+        complete_error.as_service_error().and_then(ProvideErrorMetadata::code),
+        Some("NoSuchUpload")
+    );
+    assert_eq!(
+        complete_error.as_service_error().and_then(ProvideErrorMetadata::message),
+        Some(
+            "The specified multipart upload does not exist. The upload ID may be invalid, or the upload may have been aborted or completed."
+        )
+    );
    info!("✅ Correctly failed to complete aborted upload");

+    let missing_object = s3_client
+        .get_object()
+        .bucket(TEST_BUCKET)
+        .key(object_key)
+        .send()
+        .await
+        .expect_err("aborting a multipart upload must not publish an object");
+    assert_eq!(missing_object.raw_response().map(|response| response.status().as_u16()), Some(404));
+    assert_eq!(missing_object.as_service_error().and_then(ProvideErrorMetadata::code), Some("NoSuchKey"));
+
    // Start a new multipart upload and complete it successfully
    info!("📤 Starting new multipart upload");
    let create_multipart_output2 = s3_client
@@ -393,11 +443,10 @@ async fn test_kms_multipart_upload_interruption() -> Result<(), Box<dyn std::err
    Ok(())
 }

-/// Test KMS resilience to temporary resource constraints
+/// Test concurrent KMS encryption requests
 #[tokio::test]
-async fn test_kms_resource_constraints() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+async fn test_kms_concurrent_encryption_requests() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    init_logging();
-    info!("🧪 Testing KMS behavior under resource constraints");

    let mut kms_env = LocalKMSTestEnvironment::new().await?;
    let _default_key_id = kms_env.start_rustfs_for_local_kms().await?;
@@ -431,29 +480,27 @@ async fn test_kms_resource_constraints() -> Result<(), Box<dyn std::error::Error
    }

    // Wait for all uploads to complete
-    let mut successful_uploads = 0;
-    let mut failed_uploads = 0;
+    let mut failures = Vec::new();

    for task in upload_tasks {
-        let (object_key, result) = task.await.unwrap();
+        let (object_key, result) = task.await?;
        match result {
            Ok(_) => {
-                successful_uploads += 1;
                info!("✅ Rapid upload {} succeeded", object_key);
            }
            Err(e) => {
-                failed_uploads += 1;
                warn!("❌ Rapid upload {} failed: {}", object_key, e);
+                failures.push(format!("{object_key}: {e}"));
            }
        }
    }

-    info!("📊 Rapid upload results: {} succeeded, {} failed", successful_uploads, failed_uploads);
-
-    // We expect most uploads to succeed even under load
-    assert!(successful_uploads >= 7, "Expected at least 7/10 rapid uploads to succeed");
+    assert!(
+        failures.is_empty(),
+        "all 10 concurrent KMS uploads must succeed; failures: {}",
+        failures.join("; ")
+    );

    kms_env.base_env.delete_test_bucket(TEST_BUCKET).await?;
-    info!("✅ Resource constraints test completed successfully");
    Ok(())
 }
@@ -66,6 +66,7 @@ const SURVIVOR_KEY: &str = "keep/object.bin";
 const TIER_NAME: &str = "KMSCOLD";
 const TIER_BUCKET: &str = "kms-ilm-cold-tier";
 const TIER_PREFIX: &str = "tiered";
+const ALLOW_LOOPBACK_TIER_ENDPOINT_ENV: (&str, &str) = ("RUSTFS_TIER_RUSTFS_ALLOW_LOOPBACK_ENDPOINT", "true");
 const TRANSITION_BUCKET: &str = "kms-ilm-transition";
 const TRANSITION_KEY: &str = "tier/object.bin";

@@ -80,7 +81,7 @@ const ILM_DEADLINE: StdDuration = StdDuration::from_secs(90);
 /// `--kms-default-key-id`, insecure dev defaults). The lifecycle env matches
 /// `reliant/lifecycle.rs::fast_lifecycle_env` plus `RUSTFS_ILM_DEBUG_DAY_SECS=2`,
 /// so a `Days=1` rule is due about two seconds after the write.
-async fn start_enforcing_ilm_server(env: &mut LocalKMSTestEnvironment) -> TestResult {
+async fn start_enforcing_ilm_server(env: &mut LocalKMSTestEnvironment, extra_env: &[(&str, &str)]) -> TestResult {
    create_key_with_specific_id(&env.kms_keys_dir, SSE_KEY).await?;

    let key_dir = env.kms_keys_dir.clone();
@@ -94,13 +95,14 @@ async fn start_enforcing_ilm_server(env: &mut LocalKMSTestEnvironment) -> TestRe
        SSE_KEY,
    ];

-    let envs = [
+    let mut envs = vec![
        ("RUSTFS_KMS_ALLOW_INSECURE_DEV_DEFAULTS", "true"),
        ("RUSTFS_KMS_ENFORCE_SSE_KEY_POLICY", "true"),
        ("RUSTFS_SCANNER_CYCLE", "1"),
        ("RUSTFS_ILM_PROCESS_TIME", "1"),
        ("RUSTFS_ILM_DEBUG_DAY_SECS", "2"),
    ];
+    envs.extend_from_slice(extra_env);

    env.base_env.start_rustfs_server_with_env(args, &envs).await?;
    Ok(())
@@ -427,7 +429,7 @@ async fn ilm_expiration_on_sse_kms_bucket_under_enforcement() -> TestResult {
    init_logging();

    let mut env = LocalKMSTestEnvironment::new().await?;
-    start_enforcing_ilm_server(&mut env).await?;
+    start_enforcing_ilm_server(&mut env, &[]).await?;
    env.base_env.create_test_bucket(EXPIRY_BUCKET).await?;

    let client = env.base_env.create_s3_client();
@@ -499,7 +501,7 @@ async fn ilm_transition_on_sse_kms_bucket_under_enforcement_reads_back() -> Test

    // Hot server: Local KMS + enforcement + accelerated lifecycle clock.
    let mut env = LocalKMSTestEnvironment::new().await?;
-    start_enforcing_ilm_server(&mut env).await?;
+    start_enforcing_ilm_server(&mut env, &[ALLOW_LOOPBACK_TIER_ENDPOINT_ENV]).await?;
    let hot_client = env.base_env.create_s3_client();

    add_rustfs_tier(&env.base_env, &cold.base_env).await?;
@@ -20,7 +20,8 @@
 //! - Complete encryption/decryption lifecycle

 use super::common::{
-    LocalKMSTestEnvironment, get_kms_status, sse_customer_key_md5_base64, test_kms_key_management, test_sse_c_encryption,
+    LocalKMSTestEnvironment, SSE_C_KEY_MISMATCH_MESSAGE, assert_s3_error, get_kms_status, sse_customer_key_md5_base64,
+    test_kms_key_management, test_sse_c_encryption,
 };
 use crate::common::{TEST_BUCKET, init_logging};
 use tracing::{error, info};
@@ -137,8 +138,8 @@ async fn test_local_kms_key_isolation() {
    // Test that different SSE-C keys create isolated encrypted objects
    let key1 = "01234567890123456789012345678901";
    let key2 = "98765432109876543210987654321098";
-    let key1_b64 = base64::Engine::encode(&base64::engine::general_purpose::STANDARD, key1);
-    let key2_b64 = base64::Engine::encode(&base64::engine::general_purpose::STANDARD, key2);
+    let key1_b64 = base64_simd::STANDARD.encode_to_string(key1);
+    let key2_b64 = base64_simd::STANDARD.encode_to_string(key2);
    let key1_md5 = sse_customer_key_md5_base64(key1);
    let key2_md5 = sse_customer_key_md5_base64(key2);

@@ -196,7 +197,13 @@ async fn test_local_kms_key_isolation() {
        .send()
        .await;

-    assert!(wrong_key_result.is_err(), "Should not be able to decrypt object1 with key2");
+    assert_s3_error(
+        wrong_key_result,
+        400,
+        "InvalidRequest",
+        SSE_C_KEY_MISMATCH_MESSAGE,
+        "local SSE-C object GET with a wrong key must be rejected",
+    );

    kms_env
        .base_env
@@ -558,7 +565,7 @@ async fn test_multipart_upload_with_sse_c(

    // SSE-C encryption key
    let encryption_key = "01234567890123456789012345678901";
-    let key_b64 = base64::Engine::encode(&base64::engine::general_purpose::STANDARD, encryption_key);
+    let key_b64 = base64_simd::STANDARD.encode_to_string(encryption_key);
    let key_md5 = sse_customer_key_md5_base64(encryption_key);

    // Generate test data
@@ -0,0 +1,191 @@
+// Copyright 2024 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+//! Bulk DEK rekey sweep over stored objects.
+//!
+//! The full loop — rotate the master key, sweep, prove convergence — runs
+//! against Vault Transit, whose context-bound envelopes exercise the
+//! decrypt + re-encrypt rewrap route end to end. The capability refusal runs
+//! against the Local backend, which supports no rewrap at all.
+
+use super::common::{
+    LocalKMSTestEnvironment, VAULT_KEY_NAME, VaultTestEnvironment, kms_admin_request, start_kms, wait_for_kms_ready,
+};
+use crate::common::{TEST_BUCKET, init_logging};
+use aws_sdk_s3::primitives::ByteStream;
+use aws_sdk_s3::types::ServerSideEncryption;
+use std::time::Duration;
+use tracing::info;
+
+async fn rekey_status(
+    base_url: &str,
+    access_key: &str,
+    secret_key: &str,
+) -> Result<serde_json::Value, Box<dyn std::error::Error + Send + Sync>> {
+    let body = kms_admin_request(
+        base_url,
+        http::Method::GET,
+        "/rustfs/admin/v3/kms/keys/rekey/status",
+        None,
+        access_key,
+        secret_key,
+    )
+    .await?;
+    Ok(serde_json::from_str(&body)?)
+}
+
+/// Start a sweep and poll it to a terminal state.
+async fn run_rekey_to_completion(
+    base_url: &str,
+    access_key: &str,
+    secret_key: &str,
+    request_body: &str,
+) -> Result<serde_json::Value, Box<dyn std::error::Error + Send + Sync>> {
+    kms_admin_request(
+        base_url,
+        http::Method::POST,
+        "/rustfs/admin/v3/kms/keys/rekey",
+        Some(request_body),
+        access_key,
+        secret_key,
+    )
+    .await?;
+
+    for _ in 0..120 {
+        let status = rekey_status(base_url, access_key, secret_key).await?;
+        if status["state"] != "running" {
+            return Ok(status);
+        }
+        tokio::time::sleep(Duration::from_millis(500)).await;
+    }
+    Err("rekey sweep did not reach a terminal state in time".into())
+}
+
+#[tokio::test]
+async fn kms_rekey_sweep_rewraps_rotated_envelopes_and_converges() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+    init_logging();
+    info!("Testing the bulk rekey sweep against Vault Transit");
+
+    let mut env = VaultTestEnvironment::new().await?;
+    env.start_vault().await?;
+    env.setup_vault_transit().await?;
+    env.start_rustfs_for_vault().await?;
+    env.configure_vault_transit_kms().await?;
+    start_kms(&env.base_env.url, &env.base_env.access_key, &env.base_env.secret_key).await?;
+    wait_for_kms_ready(&env.base_env.url, &env.base_env.access_key, &env.base_env.secret_key).await?;
+
+    let base_url = env.base_env.url.clone();
+    let access_key = env.base_env.access_key.clone();
+    let secret_key = env.base_env.secret_key.clone();
+
+    let s3_client = env.base_env.create_s3_client();
+    env.base_env.create_test_bucket(TEST_BUCKET).await?;
+
+    // Three encrypted objects the sweep must rewrap, one plaintext object it
+    // must leave alone.
+    let encrypted_keys = ["rekey/alpha", "rekey/beta", "rekey/gamma"];
+    let mut bodies = Vec::new();
+    for (index, key) in encrypted_keys.iter().enumerate() {
+        let body: Vec<u8> = (0..2048).map(|i| ((i + index * 7) % 251) as u8).collect();
+        s3_client
+            .put_object()
+            .bucket(TEST_BUCKET)
+            .key(*key)
+            .server_side_encryption(ServerSideEncryption::AwsKms)
+            .ssekms_key_id(VAULT_KEY_NAME)
+            .body(ByteStream::from(body.clone()))
+            .send()
+            .await?;
+        bodies.push(body);
+    }
+    s3_client
+        .put_object()
+        .bucket(TEST_BUCKET)
+        .key("rekey/plaintext")
+        .body(ByteStream::from(b"unencrypted".to_vec()))
+        .send()
+        .await?;
+
+    // Rotate the master key so the stored envelopes fall behind Vault's
+    // latest version.
+    kms_admin_request(
+        &base_url,
+        http::Method::POST,
+        "/rustfs/admin/v3/kms/keys/rotate",
+        Some(&format!(r#"{{"key_id":"{VAULT_KEY_NAME}"}}"#)),
+        &access_key,
+        &secret_key,
+    )
+    .await?;
+
+    let status =
+        run_rekey_to_completion(&base_url, &access_key, &secret_key, &format!(r#"{{"buckets":["{TEST_BUCKET}"]}}"#)).await?;
+    assert_eq!(status["state"], "completed", "first sweep must complete: {status}");
+    assert_eq!(status["failed"], 0, "no object may fail: {status}");
+    assert_eq!(
+        status["rewrapped"],
+        encrypted_keys.len(),
+        "every rotated envelope must be rewrapped: {status}"
+    );
+    assert!(
+        status["not_applicable"].as_u64().unwrap_or(0) >= 1,
+        "the plaintext object must be reported not applicable: {status}"
+    );
+
+    // The rewrapped objects still serve their exact bytes.
+    for (key, expected) in encrypted_keys.iter().zip(&bodies) {
+        let response = s3_client.get_object().bucket(TEST_BUCKET).key(*key).send().await?;
+        let data = response.body.collect().await?.into_bytes();
+        assert_eq!(data.as_ref(), expected.as_slice(), "object {key} must be byte-exact after the rewrap");
+    }
+
+    // Convergence: a second sweep finds everything current and writes nothing.
+    let status =
+        run_rekey_to_completion(&base_url, &access_key, &secret_key, &format!(r#"{{"buckets":["{TEST_BUCKET}"]}}"#)).await?;
+    assert_eq!(status["state"], "completed", "second sweep must complete: {status}");
+    assert_eq!(status["rewrapped"], 0, "a converged sweep must write nothing: {status}");
+    assert_eq!(status["failed"], 0, "{status}");
+    assert_eq!(
+        status["already_current"],
+        encrypted_keys.len(),
+        "every envelope must now be current: {status}"
+    );
+
+    env.base_env.delete_test_bucket(TEST_BUCKET).await?;
+    Ok(())
+}
+
+#[tokio::test]
+async fn kms_rekey_refuses_a_backend_without_rewrap_support() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+    init_logging();
+    info!("Testing that the rekey sweep refuses the Local backend up front");
+
+    let mut kms_env = LocalKMSTestEnvironment::new().await?;
+    let _default_key_id = kms_env.start_rustfs_for_local_kms().await?;
+    kms_env.wait_for_kms_ready().await?;
+
+    let error = kms_admin_request(
+        &kms_env.base_env.url,
+        http::Method::POST,
+        "/rustfs/admin/v3/kms/keys/rekey",
+        Some("{}"),
+        &kms_env.base_env.access_key,
+        &kms_env.base_env.secret_key,
+    )
+    .await
+    .expect_err("a backend without rewrap support must be refused up front");
+    assert!(error.to_string().contains("501"), "the refusal must be 501 Not Implemented, got: {error}");
+
+    Ok(())
+}
@@ -22,9 +22,9 @@ use crate::common::{TEST_BUCKET, init_logging};
 use tracing::{error, info};

 use super::common::{
-    VAULT_KEY_NAME, VaultTestEnvironment, get_kms_status, sse_customer_key_md5_base64, start_kms,
-    test_all_multipart_encryption_types, test_error_scenarios, test_kms_key_management, test_sse_c_encryption,
-    test_sse_kms_encryption, test_sse_s3_encryption,
+    SSE_C_KEY_MISMATCH_MESSAGE, VAULT_KEY_NAME, VaultTestEnvironment, assert_s3_error, get_kms_status,
+    sse_customer_key_md5_base64, start_kms, test_all_multipart_encryption_types, test_error_scenarios, test_kms_key_management,
+    test_sse_c_encryption, test_sse_kms_encryption, test_sse_s3_encryption,
 };

 /// Helper that brings up Vault, configures RustFS, and starts the KMS service.
@@ -127,8 +127,8 @@ async fn test_vault_kms_key_isolation() -> Result<(), Box<dyn std::error::Error

    let key1 = "01234567890123456789012345678901";
    let key2 = "98765432109876543210987654321098";
-    let key1_b64 = base64::Engine::encode(&base64::engine::general_purpose::STANDARD, key1);
-    let key2_b64 = base64::Engine::encode(&base64::engine::general_purpose::STANDARD, key2);
+    let key1_b64 = base64_simd::STANDARD.encode_to_string(key1);
+    let key2_b64 = base64_simd::STANDARD.encode_to_string(key2);
    let key1_md5 = sse_customer_key_md5_base64(key1);
    let key2_md5 = sse_customer_key_md5_base64(key2);

@@ -182,7 +182,13 @@ async fn test_vault_kms_key_isolation() -> Result<(), Box<dyn std::error::Error
        .sse_customer_key_md5(&key2_md5)
        .send()
        .await;
-    assert!(wrong_key.is_err(), "Object1 should not decrypt with key2");
+    assert_s3_error(
+        wrong_key,
+        400,
+        "InvalidRequest",
+        SSE_C_KEY_MISMATCH_MESSAGE,
+        "Vault-backed SSE-C object GET with a wrong key must be rejected",
+    );

    context
        .base_env()
@@ -48,14 +48,23 @@ mod encryption_metadata_test;
 #[cfg(test)]
 mod copy_object_self_copy_sse_test;

+#[cfg(test)]
+mod encrypted_range_get_test;
+
 #[cfg(test)]
 mod copy_object_version_restore_sse_test;

 #[cfg(test)]
 mod configured_roundtrip_test;

+#[cfg(test)]
+mod kms_anonymous_enforcement_test;
+
 #[cfg(test)]
 mod kms_authorization_negative_matrix_test;

 #[cfg(test)]
 mod kms_ilm_sse_kms_test;
+
+#[cfg(test)]
+mod kms_rekey_sweep_test;
@@ -497,7 +497,7 @@ async fn test_multipart_encryption_type(
    // Prepare SSE-C keys when required
    let (sse_c_key, sse_c_md5) = if matches!(encryption_type, EncryptionType::SSEC) {
        let key = "01234567890123456789012345678901";
-        let key_b64 = base64::Engine::encode(&base64::engine::general_purpose::STANDARD, key);
+        let key_b64 = base64_simd::STANDARD.encode_to_string(key);
        let key_md5 = sse_customer_key_md5_base64(key);
        (Some(key_b64), Some(key_md5))
    } else {
@@ -131,8 +131,18 @@ mod tests {

        // DELETE through the raw key removes the normalized object.
        client.delete_object().bucket(bucket).key("//keyname").send().await?;
-        let result = client.get_object().bucket(bucket).key("keyname").send().await;
-        assert!(result.is_err(), "object must be gone after DELETE with raw key");
+        let error = client
+            .get_object()
+            .bucket(bucket)
+            .key("keyname")
+            .send()
+            .await
+            .expect_err("object must be gone after DELETE with raw key");
+        assert_eq!(
+            error.raw_response().map(|response| response.status().as_u16()),
+            Some(404),
+            "GET after DELETE with raw key must return HTTP 404, got {error:?}"
+        );

        env.stop_server();
        info!("Test completed successfully");
@@ -61,6 +61,10 @@ mod get_codec_streaming_compat_test;
 #[cfg(test)]
 mod version_id_regression_test;

+// Pinned previous-release -> current-build on-disk compatibility.
+#[cfg(test)]
+mod upgrade_compatibility_test;
+
 // Receiver-side replication LWW (rustfs/backlog#1953): stale inbound
 // replication metadata must not overwrite a newer local category state.
 #[cfg(test)]
@@ -280,6 +284,7 @@ mod console_smoke_test;
 // plus non-admin 403 probes per endpoint (sec-4 pattern).
 #[cfg(test)]
 mod admin_iam_crud_test;
+mod admin_mfa_test;

 #[cfg(test)]
 mod admin_pools_test;
@@ -22,7 +22,6 @@ use aws_sdk_s3::primitives::ByteStream;
 use aws_sdk_s3::types::{
    ServerSideEncryption, ServerSideEncryptionByDefault, ServerSideEncryptionConfiguration, ServerSideEncryptionRule,
 };
-use base64::Engine;
 use chrono::{Duration as ChronoDuration, Utc};
 use flate2::{Compression, write::GzEncoder};
 use http::HeaderValue;
@@ -47,19 +46,19 @@ fn encode_post_policy(conditions: Vec<serde_json::Value>) -> String {
        "conditions": conditions,
    });

-    base64::engine::general_purpose::STANDARD.encode(policy.to_string())
+    base64_simd::STANDARD.encode_to_string(policy.to_string())
 }

 fn sse_customer_key_md5_base64(key: &str) -> String {
    let mut hasher = Md5::new();
    hasher.update(key.as_bytes());
-    base64::engine::general_purpose::STANDARD.encode(hasher.finalize())
+    base64_simd::STANDARD.encode_to_string(hasher.finalize())
 }

 fn md5_hex(input: impl AsRef<[u8]>) -> String {
    let mut hasher = Md5::new();
    hasher.update(input.as_ref());
-    hex::encode(hasher.finalize())
+    hex_simd::encode_to_string(hasher.finalize(), hex_simd::AsciiCase::Lower)
 }

 async fn create_restricted_user(
@@ -97,7 +96,7 @@ fn restricted_user_client(env: &RustFSTestEnvironment, username: &str, secret_ke
 const LOCAL_SSE_MASTER_KEY_ENV: &str = "RUSTFS_SSE_S3_MASTER_KEY";

 fn local_sse_master_key_value() -> String {
-    base64::engine::general_purpose::STANDARD.encode([0x42u8; 32])
+    base64_simd::STANDARD.encode_to_string([0x42u8; 32])
 }

 async fn make_tar(files: &[(&str, &[u8])], dirs: &[&str]) -> Vec<u8> {
@@ -1887,7 +1886,7 @@ async fn test_anonymous_post_object_allows_sse_c_fields_outside_policy_condition
    let object_key = "sse-c-object.txt";
    let expected_body = b"anonymous-post-sse-c".to_vec();
    let customer_key = "01234567890123456789012345678901";
-    let customer_key_b64 = base64::engine::general_purpose::STANDARD.encode(customer_key);
+    let customer_key_b64 = base64_simd::STANDARD.encode_to_string(customer_key);
    let customer_key_md5 = sse_customer_key_md5_base64(customer_key);

    let admin_client = env.create_s3_client();
@@ -1941,7 +1940,7 @@ async fn test_anonymous_post_object_allows_sse_c_fields_outside_policy_condition
        .bucket(bucket)
        .key(object_key)
        .sse_customer_algorithm("AES256")
-        .sse_customer_key(base64::engine::general_purpose::STANDARD.encode(customer_key))
+        .sse_customer_key(base64_simd::STANDARD.encode_to_string(customer_key))
        .sse_customer_key_md5(customer_key_md5)
        .send()
        .await?;
@@ -1963,8 +1962,8 @@ async fn test_anonymous_post_object_rejects_sse_c_exact_policy_mismatch() -> Res
    let object_key = "sse-c-mismatch-object.txt";
    let policy_key = "01234567890123456789012345678901";
    let request_key = "abcdefghijklmnopqrstuvwxyzABCDEF";
-    let policy_key_b64 = base64::engine::general_purpose::STANDARD.encode(policy_key);
-    let request_key_b64 = base64::engine::general_purpose::STANDARD.encode(request_key);
+    let policy_key_b64 = base64_simd::STANDARD.encode_to_string(policy_key);
+    let request_key_b64 = base64_simd::STANDARD.encode_to_string(request_key);

    let admin_client = env.create_s3_client();
    admin_client.create_bucket().bucket(bucket).send().await?;
@@ -3526,7 +3525,7 @@ async fn test_signed_put_object_extract_preserves_sse_s3_and_redirect() -> Resul
    init_logging();

    let mut env = RustFSTestEnvironment::new().await?;
-    let sse_master_key = base64::engine::general_purpose::STANDARD.encode([0x42u8; 32]);
+    let sse_master_key = base64_simd::STANDARD.encode_to_string([0x42u8; 32]);
    env.start_rustfs_server_with_env(vec![], &[("RUSTFS_SSE_S3_MASTER_KEY", sse_master_key.as_str())])
        .await?;

@@ -3799,7 +3798,7 @@ async fn test_signed_put_object_extract_uses_bucket_default_sse_s3() -> Result<(
    init_logging();

    let mut env = RustFSTestEnvironment::new().await?;
-    let sse_master_key = base64::engine::general_purpose::STANDARD.encode([0x42u8; 32]);
+    let sse_master_key = base64_simd::STANDARD.encode_to_string([0x42u8; 32]);
    env.start_rustfs_server_with_env(vec![], &[("RUSTFS_SSE_S3_MASTER_KEY", sse_master_key.as_str())])
        .await?;

@@ -3925,7 +3924,7 @@ async fn test_signed_put_object_extract_preserves_sse_c() -> Result<(), Box<dyn
    let extracted_key = "nested/file.txt";
    let expected_body = b"extract-sse-c-body".to_vec();
    let customer_key = "01234567890123456789012345678901";
-    let customer_key_b64 = base64::engine::general_purpose::STANDARD.encode(customer_key);
+    let customer_key_b64 = base64_simd::STANDARD.encode_to_string(customer_key);
    let customer_key_md5 = sse_customer_key_md5_base64(customer_key);

    let client = env.create_s3_client();
@@ -34,6 +34,7 @@
 //! rejected header-SigV4 requests.

 use crate::common::{RustFSTestEnvironment, init_logging, local_http_client};
+use aws_sdk_s3::error::ProvideErrorMetadata;
 use aws_sdk_s3::primitives::ByteStream;
 use rustfs_signer::constants::UNSIGNED_PAYLOAD;
 use rustfs_signer::request_signature_v4::{SIGN_V4_ALGORITHM, get_scope, get_signature, get_signing_key};
@@ -280,7 +281,8 @@ async fn tampered_payload_is_rejected() -> Result<(), Box<dyn std::error::Error
    let mut env = RustFSTestEnvironment::new().await?;
    setup(&mut env).await?;

-    let path = format!("/{BUCKET}/tampered-payload.txt");
+    let key = "tampered-payload.txt";
+    let path = format!("/{BUCKET}/{key}");
    let claimed_body = b"the-body-i-claim-to-send";
    let actual_body = b"the-body-i-really-send!!";
    assert_eq!(claimed_body.len(), actual_body.len(), "keep content-length stable for the mismatch");
@@ -295,17 +297,82 @@ async fn tampered_payload_is_rejected() -> Result<(), Box<dyn std::error::Error
        Ok(resp) => {
            let status = resp.status();
            let body = resp.text().await.unwrap_or_default();
-            assert_ne!(status.as_u16(), 200, "payload mismatch must not succeed, body:\n{body}");
            assert!(
-                status.is_client_error() || status.is_server_error(),
-                "payload mismatch must be an error status, got {status}, body:\n{body}"
+                status.is_client_error(),
+                "payload mismatch must be rejected with a client error, got {status}, body:\n{body}"
            );
            info!(%status, "tampered payload rejected with error status");
        }
        // A mid-stream hash-mismatch abort surfacing as a transport error is
        // also a valid rejection (definitely not a 200 success).
-        Err(err) => info!(%err, "tampered payload rejected via transport error"),
+        Err(err) => {
+            assert!(!err.is_connect(), "connection failure is not proof of payload rejection: {err}");
+            assert!(!err.is_timeout(), "request timeout is not proof of payload rejection: {err}");
+            info!(%err, "tampered payload rejected via mid-stream transport error");
+        }
    }
+
+    let absent = env
+        .create_s3_client()
+        .get_object()
+        .bucket(BUCKET)
+        .key(key)
+        .send()
+        .await
+        .expect_err("a tampered payload must not publish an object");
+    assert_eq!(absent.raw_response().map(|response| response.status().as_u16()), Some(404));
+    assert_eq!(absent.as_service_error().and_then(ProvideErrorMetadata::code), Some("NoSuchKey"));
+    Ok(())
+}
+
+/// A signed UploadPart body must pass the same payload-hash gate as PutObject.
+/// Rejection must happen before the part is published into the multipart upload.
+#[tokio::test]
+async fn tampered_upload_part_payload_is_rejected() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+    init_logging();
+    let mut env = RustFSTestEnvironment::new().await?;
+    setup(&mut env).await?;
+
+    let key = "tampered-upload-part.bin";
+    let client = env.create_s3_client();
+    let upload = client.create_multipart_upload().bucket(BUCKET).key(key).send().await?;
+    let upload_id = upload.upload_id().ok_or("create multipart upload omitted upload_id")?;
+
+    let path = format!("/{BUCKET}/{key}");
+    let canonical_query = format!("partNumber=1&uploadId={}", urlencoding::encode(upload_id));
+    let request_target = format!("{path}?{canonical_query}");
+    let claimed_body = b"the-part-i-claim-to-send";
+    let actual_body = b"the-part-i-really-send!!";
+    assert_eq!(claimed_body.len(), actual_body.len(), "keep content-length stable for the mismatch");
+
+    let signer = SigV4::new(&env);
+    let headers = signer.sign("PUT", &path, &canonical_query, &sha256_hex(claimed_body));
+    let resp = send_signed(&env, reqwest::Method::PUT, &request_target, &headers, Some(actual_body.to_vec())).await?;
+    let status = resp.status();
+    let body = resp.text().await.unwrap_or_default();
+    assert_eq!(
+        status,
+        reqwest::StatusCode::BAD_REQUEST,
+        "multipart payload mismatch must be rejected as BadDigest, body:\n{body}"
+    );
+    assert_error_code(&body, "BadDigest");
+
+    let parts = client
+        .list_parts()
+        .bucket(BUCKET)
+        .key(key)
+        .upload_id(upload_id)
+        .send()
+        .await?;
+    assert!(parts.parts().is_empty(), "a tampered UploadPart must not publish a part");
+
+    client
+        .abort_multipart_upload()
+        .bucket(BUCKET)
+        .key(key)
+        .upload_id(upload_id)
+        .send()
+        .await?;
    Ok(())
 }

@@ -38,14 +38,10 @@ use aws_sdk_s3::types::{
    NotificationConfiguration, NotificationConfigurationFilter, ObjectIdentifier, QueueConfiguration, S3KeyFilter,
    VersioningConfiguration,
 };
-use http::header::{CONTENT_TYPE, HOST};
 use local_ip_address::local_ip;
 use reqwest::StatusCode;
-use rustfs_signer::constants::UNSIGNED_PAYLOAD;
-use rustfs_signer::sign_v4;
 use rustfs_utils::egress::ENV_OUTBOUND_ALLOW_ORIGINS;
 use rustfs_utils::http::headers::{AMZ_REQUEST_ID, REQUEST_ID_HEADER};
-use s3s::Body;
 use serde_json::Value;
 use std::error::Error;
 use std::io::Cursor;
@@ -415,42 +411,16 @@ async fn collect_until(
 // Admin target configuration (signed admin HTTP)
 // ---------------------------------------------------------------------------

+/// Thin wrapper over [`crate::common::signed_request`] with this suite's
+/// root credentials; a `Some` body is always JSON here.
 async fn signed_admin_request(
    env: &RustFSTestEnvironment,
    method: http::Method,
    url: &str,
    body: Option<Vec<u8>>,
 ) -> Result<reqwest::Response, BoxError> {
-    let uri = url.parse::<http::Uri>()?;
-    let authority = uri.authority().ok_or("admin URL missing authority")?.to_string();
-    let mut builder = http::Request::builder()
-        .method(method.clone())
-        .uri(uri)
-        .header(HOST, authority)
-        .header("x-amz-content-sha256", UNSIGNED_PAYLOAD);
-    if body.is_some() {
-        builder = builder.header(CONTENT_TYPE, "application/json");
-    }
-
-    let content_len = body.as_ref().map(|b| b.len() as i64).unwrap_or_default();
-    let signed = sign_v4(
-        builder.body(Body::empty())?,
-        content_len,
-        &env.access_key,
-        &env.secret_key,
-        "",
-        "us-east-1",
-    );
-
-    let reqwest_method = reqwest::Method::from_bytes(method.as_str().as_bytes())?;
-    let mut request = crate::common::local_http_client().request(reqwest_method, url);
-    for (name, value) in signed.headers() {
-        request = request.header(name, value);
-    }
-    if let Some(body) = body {
-        request = request.body(body);
-    }
-    Ok(request.send().await?)
+    let content_type = body.is_some().then_some("application/json");
+    crate::common::signed_request(method, url, &env.access_key, &env.secret_key, body, content_type).await
 }

 async fn enable_notify_module(env: &RustFSTestEnvironment) -> TestResult {
@@ -21,6 +21,9 @@
 //! - Bypass governance retention header handling

 use aws_sdk_s3::Client;
+use aws_sdk_s3::error::SdkError;
+use aws_sdk_s3::operation::delete_object::DeleteObjectError;
+use aws_sdk_s3::operation::put_object_retention::PutObjectRetentionError;
 use aws_sdk_s3::primitives::ByteStream;
 use aws_sdk_s3::types::{
    DefaultRetention, ObjectLockConfiguration, ObjectLockEnabled, ObjectLockLegalHold, ObjectLockLegalHoldStatus, ObjectLockMode,
@@ -180,11 +183,8 @@ pub async fn put_object_retention(
    mode: ObjectLockRetentionMode,
    retain_until: DateTime<Utc>,
    bypass_governance: bool,
-) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-    // AWS SDK requires UTC time without timezone offset (e.g., "2026-01-24T11:20:14Z")
-    let retain_until_str = retain_until.format("%Y-%m-%dT%H:%M:%SZ").to_string();
-    let retain_until_datetime =
-        aws_sdk_s3::primitives::DateTime::from_str(&retain_until_str, aws_sdk_s3::primitives::DateTimeFormat::DateTime)?;
+) -> Result<(), Box<SdkError<PutObjectRetentionError>>> {
+    let retain_until_datetime = aws_sdk_s3::primitives::DateTime::from_secs(retain_until.timestamp());

    let retention = ObjectLockRetention::builder()
        .mode(mode.clone())
@@ -202,7 +202,7 @@ pub async fn put_object_retention(
        request = request.version_id(vid);
    }

-    request.send().await?;
+    request.send().await.map_err(Box::new)?;
    info!("Put object retention on {} with mode {:?}", key, mode);
    Ok(())
 }
@@ -236,7 +236,7 @@ pub async fn delete_object_with_bypass(
    key: &str,
    version_id: Option<&str>,
    bypass_governance: bool,
-) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+) -> Result<(), Box<SdkError<DeleteObjectError>>> {
    let mut request = client
        .delete_object()
        .bucket(bucket)
@@ -247,7 +247,7 @@ pub async fn delete_object_with_bypass(
        request = request.version_id(vid);
    }

-    request.send().await?;
+    request.send().await.map_err(Box::new)?;
    info!("Deleted object {} (bypass: {})", key, bypass_governance);
    Ok(())
 }
@@ -24,9 +24,11 @@
 //! - PutObjectRetention modification restrictions
 //! - Default bucket retention is applied to new objects

+use std::borrow::Borrow;
+
 use super::common::*;
 use aws_sdk_s3::Client;
-use aws_sdk_s3::error::ProvideErrorMetadata;
+use aws_sdk_s3::error::{ProvideErrorMetadata, SdkError};
 use aws_sdk_s3::primitives::{ByteStream, DateTimeFormat};
 use aws_sdk_s3::types::{
    CompletedMultipartUpload, CompletedPart, Delete, MetadataDirective, ObjectIdentifier, ObjectLockLegalHoldStatus,
@@ -70,25 +72,49 @@ fn retention_timestamp(days: i64) -> aws_sdk_s3::primitives::DateTime {
        .expect("retention timestamp should parse")
 }

-fn assert_access_denied<T, E: std::fmt::Debug>(result: Result<T, E>, context: &str) {
-    let err = match result {
-        Ok(_) => panic!("{context}"),
-        Err(err) => format!("{err:?}"),
-    };
-    assert!(
-        err.contains("AccessDenied") || err.to_lowercase().contains("access denied"),
-        "{context}: expected AccessDenied, got: {err}"
+fn assert_access_denied<T, E, R>(result: Result<T, R>, context: &str)
+where
+    T: std::fmt::Debug,
+    E: ProvideErrorMetadata + std::fmt::Debug,
+    R: Borrow<SdkError<E>> + std::fmt::Debug,
+{
+    let error = result.expect_err(context);
+    let sdk_error = error.borrow();
+    assert_eq!(
+        sdk_error.raw_response().map(|response| response.status().as_u16()),
+        Some(403),
+        "{context}: expected HTTP 403, got: {error:?}"
+    );
+    assert_eq!(
+        sdk_error.as_service_error().and_then(ProvideErrorMetadata::code),
+        Some("AccessDenied"),
+        "{context}: expected AccessDenied, got: {error:?}"
    );
 }

-fn assert_invalid_object_lock_retention_pair<T, E: std::fmt::Debug>(result: Result<T, E>, context: &str) {
-    let err = match result {
-        Ok(_) => panic!("{context}"),
-        Err(err) => format!("{err:?}"),
-    };
-    assert!(
-        err.contains("InvalidRequest") || err.contains("must both be supplied"),
-        "{context}: expected invalid paired retention headers, got: {err}"
+fn assert_invalid_object_lock_retention_pair<T, E, R>(result: Result<T, R>, context: &str)
+where
+    T: std::fmt::Debug,
+    E: ProvideErrorMetadata + std::fmt::Debug,
+    R: Borrow<SdkError<E>> + std::fmt::Debug,
+{
+    let error = result.expect_err(context);
+    let sdk_error = error.borrow();
+    assert_eq!(
+        sdk_error.raw_response().map(|response| response.status().as_u16()),
+        Some(400),
+        "{context}: expected HTTP 400, got: {error:?}"
+    );
+    let service_error = sdk_error.as_service_error().expect("expected an S3 service error");
+    assert_eq!(
+        service_error.code(),
+        Some("InvalidRequest"),
+        "{context}: expected InvalidRequest, got: {error:?}"
+    );
+    assert_eq!(
+        service_error.message(),
+        Some("x-amz-object-lock-retain-until-date and x-amz-object-lock-mode must both be supplied"),
+        "{context}: unexpected error message: {error:?}"
    );
 }

@@ -129,14 +155,15 @@ async fn test_delete_object_blocked_by_compliance_retention() {
        .unwrap();

    // Attempt to delete - should fail
-    let delete_result = delete_object_with_bypass(&client, bucket, key, Some(&version_id), false).await;
-    assert!(delete_result.is_err(), "Delete should fail for COMPLIANCE locked object");
+    assert_access_denied(
+        delete_object_with_bypass(&client, bucket, key, Some(&version_id), false).await,
+        "Delete should fail for COMPLIANCE locked object",
+    );

    // Even with bypass header, COMPLIANCE should not allow deletion
-    let delete_with_bypass_result = delete_object_with_bypass(&client, bucket, key, Some(&version_id), true).await;
-    assert!(
-        delete_with_bypass_result.is_err(),
-        "Delete with bypass should still fail for COMPLIANCE mode"
+    assert_access_denied(
+        delete_object_with_bypass(&client, bucket, key, Some(&version_id), true).await,
+        "Delete with bypass should still fail for COMPLIANCE mode",
    );

    info!("✅ Test passed: COMPLIANCE retention blocks deletion");
@@ -165,8 +192,10 @@ async fn test_delete_object_blocked_by_governance_without_bypass() {
        .unwrap();

    // Attempt to delete without bypass - should fail
-    let delete_result = delete_object_with_bypass(&client, bucket, key, Some(&version_id), false).await;
-    assert!(delete_result.is_err(), "Delete without bypass should fail for GOVERNANCE locked object");
+    assert_access_denied(
+        delete_object_with_bypass(&client, bucket, key, Some(&version_id), false).await,
+        "Delete without bypass should fail for GOVERNANCE locked object",
+    );

    info!("✅ Test passed: GOVERNANCE retention blocks deletion without bypass");
 }
@@ -198,14 +227,19 @@ async fn test_delete_object_allowed_by_governance_with_bypass() {
    assert!(delete_result.is_ok(), "Delete with bypass should succeed for GOVERNANCE mode");

    // Verify object is deleted
-    let head_result = client
+    let head_error = client
        .head_object()
        .bucket(bucket)
        .key(key)
        .version_id(&version_id)
        .send()
-        .await;
-    assert!(head_result.is_err(), "Object should be deleted");
+        .await
+        .expect_err("Object should be deleted");
+    assert_eq!(
+        head_error.raw_response().map(|response| response.status().as_u16()),
+        Some(404),
+        "deleted version should return HTTP 404: {head_error:?}"
+    );

    info!("✅ Test passed: GOVERNANCE retention allows deletion with bypass");
 }
@@ -240,17 +274,18 @@ async fn test_delete_object_creates_delete_marker_for_retained_current_version()
        .expect("delete marker should have a version id")
        .to_string();

-    let protected_delete = delete_object_with_bypass(&client, bucket, key, Some(&retained_version_id), false).await;
-    assert!(protected_delete.is_err(), "Retained version should still reject direct deletion");
+    assert_access_denied(
+        delete_object_with_bypass(&client, bucket, key, Some(&retained_version_id), false).await,
+        "Retained version should still reject direct deletion",
+    );

    delete_object_with_bypass(&client, bucket, key, Some(&delete_marker_version_id), false)
        .await
        .unwrap();

-    let still_protected = delete_object_with_bypass(&client, bucket, key, Some(&retained_version_id), false).await;
-    assert!(
-        still_protected.is_err(),
-        "Retained version should remain protected after delete marker removal"
+    assert_access_denied(
+        delete_object_with_bypass(&client, bucket, key, Some(&retained_version_id), false).await,
+        "Retained version should remain protected after delete marker removal",
    );

    delete_object_with_bypass(&client, bucket, key, Some(&retained_version_id), true)
@@ -282,12 +317,16 @@ async fn test_delete_object_blocked_by_legal_hold() {
        .unwrap();

    // Attempt to delete - should fail (legal hold cannot be bypassed)
-    let delete_result = delete_object_with_bypass(&client, bucket, key, Some(&version_id), false).await;
-    assert!(delete_result.is_err(), "Delete should fail for legal hold object");
+    assert_access_denied(
+        delete_object_with_bypass(&client, bucket, key, Some(&version_id), false).await,
+        "Delete should fail for legal hold object",
+    );

    // Even with bypass header, legal hold should block deletion
-    let delete_with_bypass_result = delete_object_with_bypass(&client, bucket, key, Some(&version_id), true).await;
-    assert!(delete_with_bypass_result.is_err(), "Delete with bypass should still fail for legal hold");
+    assert_access_denied(
+        delete_object_with_bypass(&client, bucket, key, Some(&version_id), true).await,
+        "Delete with bypass should still fail for legal hold",
+    );

    info!("✅ Test passed: Legal Hold blocks deletion");
 }
@@ -315,14 +354,19 @@ async fn test_delete_object_allowed_with_legal_hold_off() {
    let delete_result = delete_object_with_bypass(&client, bucket, key, Some(&version_id), false).await;
    assert!(delete_result.is_ok(), "Delete should succeed when legal hold is OFF");

-    let head_result = client
+    let head_error = client
        .head_object()
        .bucket(bucket)
        .key(key)
        .version_id(&version_id)
        .send()
-        .await;
-    assert!(head_result.is_err(), "Object should be deleted when legal hold is OFF");
+        .await
+        .expect_err("Object should be deleted when legal hold is OFF");
+    assert_eq!(
+        head_error.raw_response().map(|response| response.status().as_u16()),
+        Some(404),
+        "deleted version should return HTTP 404: {head_error:?}"
+    );

    info!("✅ Test passed: Legal Hold OFF allows deletion");
 }
@@ -545,8 +589,10 @@ async fn test_put_object_overwrite_creates_new_version_under_legal_hold() {
        "held version must keep its legal hold after the overwrite"
    );

-    let delete_result = delete_object_with_bypass(&client, bucket, key, Some(&held_version_id), false).await;
-    assert!(delete_result.is_err(), "held version must stay delete-protected after the overwrite");
+    assert_access_denied(
+        delete_object_with_bypass(&client, bucket, key, Some(&held_version_id), false).await,
+        "held version must stay delete-protected after the overwrite",
+    );
 }

 #[tokio::test]
@@ -768,8 +814,10 @@ async fn test_copy_object_overwrite_creates_new_version_under_legal_hold() {
        "held destination version must keep its legal hold after the copy"
    );

-    let delete_result = delete_object_with_bypass(&client, bucket, dst_key, Some(&held_version_id), false).await;
-    assert!(delete_result.is_err(), "held destination version must stay delete-protected");
+    assert_access_denied(
+        delete_object_with_bypass(&client, bucket, dst_key, Some(&held_version_id), false).await,
+        "held destination version must stay delete-protected",
+    );
 }

 #[tokio::test]
@@ -909,10 +957,9 @@ async fn test_create_multipart_upload_creates_new_version_under_compliance_reten

    // COMPLIANCE retention on the previous version survives the overwrite and
    // cannot be bypassed.
-    let delete_result = delete_object_with_bypass(&client, bucket, key, Some(&retained_version_id), true).await;
-    assert!(
-        delete_result.is_err(),
-        "retained version must stay delete-protected even with governance bypass"
+    assert_access_denied(
+        delete_object_with_bypass(&client, bucket, key, Some(&retained_version_id), true).await,
+        "retained version must stay delete-protected even with governance bypass",
    );
 }

@@ -971,8 +1018,10 @@ async fn test_delete_completed_multipart_object_blocked_by_legal_hold() {
        .unwrap();

    let version_id = complete_output.version_id().expect("multipart object should be versioned");
-    let delete_result = delete_object_with_bypass(&client, bucket, key, Some(version_id), false).await;
-    assert!(delete_result.is_err(), "Delete should fail for multipart object protected by legal hold");
+    assert_access_denied(
+        delete_object_with_bypass(&client, bucket, key, Some(version_id), false).await,
+        "Delete should fail for multipart object protected by legal hold",
+    );
 }

 #[tokio::test]
@@ -1032,8 +1081,10 @@ async fn test_delete_completed_multipart_object_blocked_by_retention() {
        .unwrap();

    let version_id = complete_output.version_id().expect("multipart object should be versioned");
-    let delete_result = delete_object_with_bypass(&client, bucket, key, Some(version_id), false).await;
-    assert!(delete_result.is_err(), "Delete should fail for multipart object protected by retention");
+    assert_access_denied(
+        delete_object_with_bypass(&client, bucket, key, Some(version_id), false).await,
+        "Delete should fail for multipart object protected by retention",
+    );
 }

 #[tokio::test]
@@ -1111,8 +1162,10 @@ async fn test_complete_multipart_upload_creates_new_version_under_legal_hold() {
        "held version must keep its legal hold after multipart completion"
    );

-    let delete_result = delete_object_with_bypass(&client, bucket, key, Some(&held_version_id), false).await;
-    assert!(delete_result.is_err(), "held version must stay delete-protected");
+    assert_access_denied(
+        delete_object_with_bypass(&client, bucket, key, Some(&held_version_id), false).await,
+        "held version must stay delete-protected",
+    );
 }

 #[tokio::test]
@@ -1181,10 +1234,9 @@ async fn test_complete_multipart_upload_creates_new_version_under_compliance_ret

    // COMPLIANCE retention on the previous version survives the overwrite and
    // cannot be bypassed.
-    let delete_result = delete_object_with_bypass(&client, bucket, key, Some(&retained_version_id), true).await;
-    assert!(
-        delete_result.is_err(),
-        "retained version must stay delete-protected even with governance bypass"
+    assert_access_denied(
+        delete_object_with_bypass(&client, bucket, key, Some(&retained_version_id), true).await,
+        "retained version must stay delete-protected even with governance bypass",
    );
 }

@@ -1438,7 +1490,7 @@ async fn test_put_retention_compliance_cannot_shorten() {
    )
    .await;

-    assert!(shorten_result.is_err(), "Shortening COMPLIANCE retention should fail");
+    assert_access_denied(shorten_result, "Shortening COMPLIANCE retention should fail");

    info!("✅ Test passed: Cannot shorten COMPLIANCE retention");
 }
@@ -1561,10 +1613,7 @@ async fn test_put_retention_governance_shorten_requires_bypass() {
    )
    .await;

-    assert!(
-        shorten_without_bypass.is_err(),
-        "Shortening GOVERNANCE retention without bypass should fail"
-    );
+    assert_access_denied(shorten_without_bypass, "Shortening GOVERNANCE retention without bypass should fail");

    // Shorten with bypass - should succeed
    let shorten_with_bypass = put_object_retention(
@@ -1621,8 +1670,10 @@ async fn test_default_retention_applied_to_new_objects() {
    let version_id = response.version_id().unwrap();

    // Try to delete without bypass - should fail due to default retention
-    let delete_result = delete_object_with_bypass(&client, bucket, key, Some(version_id), false).await;
-    assert!(delete_result.is_err(), "Delete should fail for object with default retention applied");
+    assert_access_denied(
+        delete_object_with_bypass(&client, bucket, key, Some(version_id), false).await,
+        "Delete should fail for object with default retention applied",
+    );

    let retention = client
        .get_object_retention()
@@ -1710,8 +1761,10 @@ async fn test_delete_object_creates_delete_marker_for_default_retained_current_v
        .expect("delete marker should have a version id")
        .to_string();

-    let protected_delete = delete_object_with_bypass(&client, bucket, key, Some(&retained_version_id), false).await;
-    assert!(protected_delete.is_err(), "Default-retained version should still reject direct deletion");
+    assert_access_denied(
+        delete_object_with_bypass(&client, bucket, key, Some(&retained_version_id), false).await,
+        "Default-retained version should still reject direct deletion",
+    );

    let retention_after_delete_marker = client
        .get_object_retention()
@@ -2011,11 +2064,9 @@ async fn test_copy_object_retention_uses_destination_policy() {

    // COMPLIANCE retention on the previous destination version survives the
    // overwrite and cannot be bypassed.
-    let delete_result =
-        delete_object_with_bypass(&client, dst_bucket, "locked-destination", Some(&retained_version_id), true).await;
-    assert!(
-        delete_result.is_err(),
-        "retained destination version must stay delete-protected even with governance bypass"
+    assert_access_denied(
+        delete_object_with_bypass(&client, dst_bucket, "locked-destination", Some(&retained_version_id), true).await,
+        "retained destination version must stay delete-protected even with governance bypass",
    );
 }

@@ -2268,9 +2319,9 @@ async fn test_versioning_auto_enabled_with_object_lock() {
 // ============================================================================

 #[tokio::test]
-async fn test_error_message_distinguishes_legal_hold_from_retention() {
+async fn test_legal_hold_and_retention_delete_errors_are_exact_and_non_mutating() {
    init_logging();
-    info!("🧪 Test: Error messages distinguish Legal Hold from Retention");
+    info!("🧪 Test: Legal Hold and Retention reject deletes without mutating objects");

    let mut env = ObjectLockTestEnvironment::new().await.unwrap();
    env.start_rustfs().await.unwrap();
@@ -2295,7 +2346,6 @@ async fn test_error_message_distinguishes_legal_hold_from_retention() {
            .await
            .unwrap();

-    // Delete legal hold object - check error
    let lh_delete_result = client
        .delete_object()
        .bucket(bucket)
@@ -2303,18 +2353,8 @@ async fn test_error_message_distinguishes_legal_hold_from_retention() {
        .version_id(&lh_version)
        .send()
        .await;
+    assert_access_denied(lh_delete_result, "Legal Hold must reject deleting the protected version");

-    if let Err(e) = lh_delete_result {
-        let error_str = format!("{:?}", e);
-        info!("Legal hold delete error: {}", error_str);
-        // Error should mention legal hold
-        assert!(
-            error_str.to_lowercase().contains("legal") || error_str.to_lowercase().contains("hold"),
-            "Error should mention legal hold"
-        );
-    }
-
-    // Delete retention object - check error
    let ret_delete_result = client
        .delete_object()
        .bucket(bucket)
@@ -2322,16 +2362,24 @@ async fn test_error_message_distinguishes_legal_hold_from_retention() {
        .version_id(&ret_version)
        .send()
        .await;
+    assert_access_denied(ret_delete_result, "COMPLIANCE retention must reject deleting the protected version");

-    if let Err(e) = ret_delete_result {
-        let error_str = format!("{:?}", e);
-        info!("Retention delete error: {}", error_str);
-        // Error should mention retention
-        assert!(
-            error_str.to_lowercase().contains("retention") || error_str.to_lowercase().contains("compliance"),
-            "Error should mention retention"
-        );
+    for (key, version_id) in [(legal_hold_key, &lh_version), (retention_key, &ret_version)] {
+        let body = client
+            .get_object()
+            .bucket(bucket)
+            .key(key)
+            .version_id(version_id)
+            .send()
+            .await
+            .expect("rejected delete must leave the protected version readable")
+            .body
+            .collect()
+            .await
+            .expect("protected version body should remain readable")
+            .into_bytes();
+        assert_eq!(body.as_ref(), b"data", "rejected delete mutated protected object {key}");
    }

-    info!("✅ Test passed: Error messages distinguish lock types");
+    info!("✅ Test passed: protected deletes are exact and non-mutating");
 }
@@ -11,29 +11,20 @@ The tests cover the following AWS policy variable scenarios:
 3. **Variable concatenation** - Combining variables with static text like `prefix-${aws:username}-suffix`
 4. **Nested variables** - Complex nested variable patterns like `${${aws:username}-test}`
 5. **Deny scenarios** - Testing deny policies with variables
+6. **STS credentials** - Variable resolution inherited by temporary credentials

 ## Prerequisites

- RustFS server binary
 - `awscurl` utility for admin API calls
 - AWS SDK for Rust (included in the project)

 ## Running Tests

-### Run All Policy Tests Using Unified Test Runner
-
-```bash
-# Run all policy tests with comprehensive reporting
-# Note: Requires a RustFS server running on localhost:9000
-cargo test -p e2e_test policy::test_runner::test_policy_full_suite -- --nocapture --ignored  --test-threads=1
-
-# Run only critical policy tests
-cargo test -p e2e_test policy::test_runner::test_policy_critical_suite -- --nocapture --ignored --test-threads=1
-```
-
 ### Run All Policy Tests

 ```bash
 # From the project root directory
-cargo test -p e2e_test policy:: -- --nocapture --ignored --test-threads=1
-```
+cargo test -p e2e_test policy:: -- --nocapture
+```
+
+Each test starts an isolated RustFS server on a dynamically allocated local port and cleans it up afterward.
@@ -18,5 +18,3 @@
 //! including single-value, multi-value, and nested variable scenarios.

 mod policy_variables_test;
-mod test_env;
-mod test_runner;
@@ -14,65 +14,52 @@

 //! Tests for AWS IAM policy variables with single-value, multi-value, and nested scenarios

-use crate::common::{awscurl_delete, awscurl_put, init_logging};
-use crate::policy::test_env::PolicyTestEnvironment;
+use crate::common::{
+    AdminTransport, RustFSTestEnvironment, admin_add_canned_policy_via, admin_attach_user_policy_via, admin_create_user_via,
+    awscurl_delete, awscurl_put, build_test_s3_config, build_test_sts_client, init_logging,
+};
+use aws_sdk_s3::Client;
+use aws_sdk_s3::error::ProvideErrorMetadata;
 use aws_sdk_s3::primitives::ByteStream;
 use tracing::info;

-/// Helper function to create a regular user with given credentials
+/// Helper function to create a regular user with given credentials.
+///
+/// This suite deliberately drives the admin API through the external `awscurl`
+/// binary, so the shared helpers are pinned to `AdminTransport::Awscurl`.
 async fn create_user(
-    env: &PolicyTestEnvironment,
+    env: &RustFSTestEnvironment,
    username: &str,
    password: &str,
 ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-    let create_user_body = serde_json::json!({
-        "secretKey": password,
-        "status": "enabled"
-    })
-    .to_string();
-
-    let create_user_url = format!("{}/rustfs/admin/v3/add-user?accessKey={}", env.url, username);
-    awscurl_put(&create_user_url, &create_user_body, &env.access_key, &env.secret_key).await?;
-    Ok(())
-}
-
-/// Helper function to create an STS user with given credentials
-async fn create_sts_user(
-    env: &PolicyTestEnvironment,
-    username: &str,
-    password: &str,
-) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-    // For STS, we create a regular user first, then use it to assume roles
-    create_user(env, username, password).await?;
-    Ok(())
+    admin_create_user_via(AdminTransport::Awscurl, &env.url, &env.access_key, &env.secret_key, username, password).await
 }

 /// Helper function to create and attach a policy
 async fn create_and_attach_policy(
-    env: &PolicyTestEnvironment,
+    env: &RustFSTestEnvironment,
    policy_name: &str,
    username: &str,
    policy_document: serde_json::Value,
 ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-    let policy_string = policy_document.to_string();
-
-    // Create policy
-    let add_policy_url = format!("{}/rustfs/admin/v3/add-canned-policy?name={}", env.url, policy_name);
-    awscurl_put(&add_policy_url, &policy_string, &env.access_key, &env.secret_key).await?;
-
-    // Attach policy to user
-    let attach_policy_url = format!(
-        "{}/rustfs/admin/v3/set-user-or-group-policy?policyName={}&userOrGroup={}&isGroup=false",
-        env.url, policy_name, username
-    );
-    awscurl_put(&attach_policy_url, "", &env.access_key, &env.secret_key).await?;
+    admin_add_canned_policy_via(
+        AdminTransport::Awscurl,
+        &env.url,
+        &env.access_key,
+        &env.secret_key,
+        policy_name,
+        &policy_document.to_string(),
+    )
+    .await?;
+    admin_attach_user_policy_via(AdminTransport::Awscurl, &env.url, &env.access_key, &env.secret_key, policy_name, username)
+        .await?;
    Ok(())
 }

 /// Helper function to clean up test resources
-async fn cleanup_user_and_policy(env: &PolicyTestEnvironment, username: &str, policy_name: &str) {
+async fn cleanup_user_and_policy(env: &RustFSTestEnvironment, username: &str, policy_name: &str) {
    // Create admin client for cleanup
-    let admin_client = env.create_s3_client(&env.access_key, &env.secret_key);
+    let admin_client = env.create_s3_client();

    // Delete buckets that might have been created by this user
    let bucket_patterns = [
@@ -84,7 +71,7 @@ async fn cleanup_user_and_policy(env: &PolicyTestEnvironment, username: &str, po
        format!("{username}-test"),
        format!("{username}-sts-bucket"),
        format!("{username}-service-bucket"),
-        "private-test-bucket".to_string(), // For deny test
+        format!("{username}-private-bucket"),
    ];

    // Try to delete objects and buckets
@@ -121,24 +108,18 @@ async fn cleanup_user_and_policy(env: &PolicyTestEnvironment, username: &str, po

 /// Test AWS policy variables with single-value scenarios
 #[tokio::test(flavor = "multi_thread")]
-#[ignore = "Starts a rustfs server; enable when running full E2E"]
 pub async fn test_aws_policy_variables_single_value() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-    test_aws_policy_variables_single_value_impl().await
-}
-
-/// Implementation function for single-value policy variables test
-pub async fn test_aws_policy_variables_single_value_impl() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    init_logging();
    info!("Starting AWS policy variables single-value test");

-    let env = PolicyTestEnvironment::with_address("127.0.0.1:9000").await?;
+    let mut env = RustFSTestEnvironment::new().await?;
+    env.start_rustfs_server(vec![]).await?;

    test_aws_policy_variables_single_value_impl_with_env(&env).await
 }

-/// Implementation function for single-value policy variables test with shared environment
-pub async fn test_aws_policy_variables_single_value_impl_with_env(
-    env: &PolicyTestEnvironment,
+async fn test_aws_policy_variables_single_value_impl_with_env(
+    env: &RustFSTestEnvironment,
 ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    // Create test user
    let test_user = "testuser1";
@@ -198,9 +179,7 @@ pub async fn test_aws_policy_variables_single_value_impl_with_env(
    awscurl_put(&attach_policy_url, "", &env.access_key, &env.secret_key).await?;

    // Create S3 client for test user
-    let test_client = env.create_s3_client(test_user, test_password);
-
-    tokio::time::sleep(std::time::Duration::from_millis(500)).await;
+    let test_client = env.create_s3_client_with_credentials(test_user, test_password);

    // Test 1: User should be able to list buckets (allowed by policy)
    info!("Test 1: User listing buckets");
@@ -257,11 +236,13 @@ pub async fn test_aws_policy_variables_single_value_impl_with_env(
    // Test 6: User should NOT be able to create bucket NOT matching username pattern
    info!("Test 6: User attempting to create bucket NOT matching pattern");
    let other_bucket_name = "other-user-bucket";
-    let create_other_result = test_client.create_bucket().bucket(other_bucket_name).send().await;
-    if create_other_result.is_ok() {
-        cleanup().await;
-        return Err("User should NOT be able to create bucket NOT matching username pattern".into());
-    }
+    let denied = test_client
+        .create_bucket()
+        .bucket(other_bucket_name)
+        .send()
+        .await
+        .expect_err("a bucket outside the username pattern must be denied");
+    assert_eq!(denied.as_service_error().and_then(ProvideErrorMetadata::code), Some("AccessDenied"));

    // Cleanup
    info!("Cleaning up test resources");
@@ -273,24 +254,18 @@ pub async fn test_aws_policy_variables_single_value_impl_with_env(

 /// Test AWS policy variables with multi-value scenarios
 #[tokio::test(flavor = "multi_thread")]
-#[ignore = "Starts a rustfs server; enable when running full E2E"]
 pub async fn test_aws_policy_variables_multi_value() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-    test_aws_policy_variables_multi_value_impl().await
-}
-
-/// Implementation function for multi-value policy variables test
-pub async fn test_aws_policy_variables_multi_value_impl() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    init_logging();
    info!("Starting AWS policy variables multi-value test");

-    let env = PolicyTestEnvironment::with_address("127.0.0.1:9000").await?;
+    let mut env = RustFSTestEnvironment::new().await?;
+    env.start_rustfs_server(vec![]).await?;

    test_aws_policy_variables_multi_value_impl_with_env(&env).await
 }

-/// Implementation function for multi-value policy variables test with shared environment
-pub async fn test_aws_policy_variables_multi_value_impl_with_env(
-    env: &PolicyTestEnvironment,
+async fn test_aws_policy_variables_multi_value_impl_with_env(
+    env: &RustFSTestEnvironment,
 ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    // Create test user
    let test_user = "testuser2";
@@ -338,7 +313,7 @@ pub async fn test_aws_policy_variables_multi_value_impl_with_env(
    create_and_attach_policy(env, policy_name, test_user, policy_document).await?;

    // Create S3 client for test user
-    let test_client = env.create_s3_client(test_user, test_password);
+    let test_client = env.create_s3_client_with_credentials(test_user, test_password);

    // Test 1: User should be able to create buckets matching any of the multi-value patterns
    info!("Test 1: User creating first bucket matching multi-value pattern");
@@ -368,11 +343,13 @@ pub async fn test_aws_policy_variables_multi_value_impl_with_env(
    // Test 4: User should NOT be able to create bucket NOT matching any multi-value pattern
    info!("Test 4: User attempting to create bucket NOT matching any pattern");
    let other_bucket_name = format!("{test_user}-other-bucket");
-    let create_other_result = test_client.create_bucket().bucket(&other_bucket_name).send().await;
-    if create_other_result.is_ok() {
-        cleanup().await;
-        return Err("User should NOT be able to create bucket NOT matching any multi-value pattern".into());
-    }
+    let denied = test_client
+        .create_bucket()
+        .bucket(&other_bucket_name)
+        .send()
+        .await
+        .expect_err("a bucket outside all allowed patterns must be denied");
+    assert_eq!(denied.as_service_error().and_then(ProvideErrorMetadata::code), Some("AccessDenied"));

    // Test 5: User should be able to list objects in their allowed buckets
    info!("Test 5: User listing objects in allowed buckets");
@@ -398,24 +375,18 @@ pub async fn test_aws_policy_variables_multi_value_impl_with_env(

 /// Test AWS policy variables with variable concatenation
 #[tokio::test(flavor = "multi_thread")]
-#[ignore = "Starts a rustfs server; enable when running full E2E"]
 pub async fn test_aws_policy_variables_concatenation() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-    test_aws_policy_variables_concatenation_impl().await
-}
-
-/// Implementation function for concatenation policy variables test
-pub async fn test_aws_policy_variables_concatenation_impl() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    init_logging();
    info!("Starting AWS policy variables concatenation test");

-    let env = PolicyTestEnvironment::with_address("127.0.0.1:9000").await?;
+    let mut env = RustFSTestEnvironment::new().await?;
+    env.start_rustfs_server(vec![]).await?;

    test_aws_policy_variables_concatenation_impl_with_env(&env).await
 }

-/// Implementation function for concatenation policy variables test with shared environment
-pub async fn test_aws_policy_variables_concatenation_impl_with_env(
-    env: &PolicyTestEnvironment,
+async fn test_aws_policy_variables_concatenation_impl_with_env(
+    env: &RustFSTestEnvironment,
 ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    // Create test user
    let test_user = "testuser3";
@@ -455,10 +426,7 @@ pub async fn test_aws_policy_variables_concatenation_impl_with_env(
    create_and_attach_policy(env, policy_name, test_user, policy_document).await?;

    // Create S3 client for test user
-    let test_client = env.create_s3_client(test_user, test_password);
-
-    // Add a small delay to allow policy to propagate
-    tokio::time::sleep(std::time::Duration::from_millis(500)).await;
+    let test_client = env.create_s3_client_with_credentials(test_user, test_password);

    // Test: User should be able to create bucket matching concatenated pattern
    info!("Test: User creating bucket matching concatenated pattern");
@@ -487,41 +455,30 @@ pub async fn test_aws_policy_variables_concatenation_impl_with_env(

 /// Test AWS policy variables with nested scenarios
 #[tokio::test(flavor = "multi_thread")]
-#[ignore = "Starts a rustfs server; enable when running full E2E"]
 pub async fn test_aws_policy_variables_nested() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-    test_aws_policy_variables_nested_impl().await
-}
-
-/// Implementation function for nested policy variables test
-pub async fn test_aws_policy_variables_nested_impl() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    init_logging();
    info!("Starting AWS policy variables nested test");

-    let env = PolicyTestEnvironment::with_address("127.0.0.1:9000").await?;
+    let mut env = RustFSTestEnvironment::new().await?;
+    env.start_rustfs_server(vec![]).await?;

    test_aws_policy_variables_nested_impl_with_env(&env).await
 }

 /// Test AWS policy variables with STS temporary credentials
 #[tokio::test(flavor = "multi_thread")]
-#[ignore = "Starts a rustfs server; enable when running full E2E"]
 pub async fn test_aws_policy_variables_sts() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-    test_aws_policy_variables_sts_impl().await
-}
-
-/// Implementation function for STS policy variables test
-pub async fn test_aws_policy_variables_sts_impl() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    init_logging();
    info!("Starting AWS policy variables STS test");

-    let env = PolicyTestEnvironment::with_address("127.0.0.1:9000").await?;
+    let mut env = RustFSTestEnvironment::new().await?;
+    env.start_rustfs_server(vec![]).await?;

    test_aws_policy_variables_sts_impl_with_env(&env).await
 }

-/// Implementation function for nested policy variables test with shared environment
-pub async fn test_aws_policy_variables_nested_impl_with_env(
-    env: &PolicyTestEnvironment,
+async fn test_aws_policy_variables_nested_impl_with_env(
+    env: &RustFSTestEnvironment,
 ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    // Create test user
    let test_user = "testuser4";
@@ -561,10 +518,7 @@ pub async fn test_aws_policy_variables_nested_impl_with_env(
    create_and_attach_policy(env, policy_name, test_user, policy_document).await?;

    // Create S3 client for test user
-    let test_client = env.create_s3_client(test_user, test_password);
-
-    // Add a small delay to allow policy to propagate
-    tokio::time::sleep(std::time::Duration::from_millis(500)).await;
+    let test_client = env.create_s3_client_with_credentials(test_user, test_password);

    // Test nested variable resolution
    info!("Test: Nested variable resolution");
@@ -581,14 +535,14 @@ pub async fn test_aws_policy_variables_nested_impl_with_env(
        return Err(format!("User should be able to create bucket with nested variable: {e}").into());
    }

-    // Verify bucket creation fails with unresolved variable
-    let unresolved_bucket = format!("${{}}-test {test_user}");
-    let create_unresolved = test_client.create_bucket().bucket(&unresolved_bucket).send().await;
-
-    if create_unresolved.is_ok() {
-        cleanup().await;
-        return Err("User should NOT be able to create bucket with unresolved variable".into());
-    }
+    // Verify a valid bucket name outside the resolved resource is denied.
+    let denied = test_client
+        .create_bucket()
+        .bucket("other-user-test")
+        .send()
+        .await
+        .expect_err("a bucket outside the resolved nested variable must be denied");
+    assert_eq!(denied.as_service_error().and_then(ProvideErrorMetadata::code), Some("AccessDenied"));

    // Cleanup
    info!("Cleaning up test resources");
@@ -598,9 +552,8 @@ pub async fn test_aws_policy_variables_nested_impl_with_env(
    Ok(())
 }

-/// Implementation function for STS policy variables test with shared environment
-pub async fn test_aws_policy_variables_sts_impl_with_env(
-    env: &PolicyTestEnvironment,
+async fn test_aws_policy_variables_sts_impl_with_env(
+    env: &RustFSTestEnvironment,
 ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    // Create test user for STS
    let test_user = "testuser-sts";
@@ -612,8 +565,7 @@ pub async fn test_aws_policy_variables_sts_impl_with_env(
        cleanup_user_and_policy(env, test_user, policy_name).await;
    };

-    // Create STS user
-    create_sts_user(env, test_user, test_password).await?;
+    create_user(env, test_user, test_password).await?;

    // Create policy with STS-compatible variables
    let policy_document = serde_json::json!({
@@ -624,6 +576,11 @@ pub async fn test_aws_policy_variables_sts_impl_with_env(
                "Action": ["s3:ListAllMyBuckets"],
                "Resource": ["arn:aws:s3:::*"]
            },
+            {
+                "Effect": "Allow",
+                "Action": ["sts:AssumeRole"],
+                "Resource": ["arn:aws:s3:::*"]
+            },
            {
                "Effect": "Allow",
                "Action": ["s3:CreateBucket"],
@@ -631,7 +588,12 @@ pub async fn test_aws_policy_variables_sts_impl_with_env(
            },
            {
                "Effect": "Allow",
-                "Action": ["s3:ListBucket", "s3:PutObject", "s3:GetObject"],
+                "Action": ["s3:ListBucket"],
+                "Resource": [format!("arn:aws:s3:::{}-sts-bucket", "${aws:username}")]
+            },
+            {
+                "Effect": "Allow",
+                "Action": ["s3:PutObject", "s3:GetObject"],
                "Resource": [format!("arn:aws:s3:::{}-sts-bucket/*", "${aws:username}")]
            }
        ]
@@ -639,11 +601,22 @@ pub async fn test_aws_policy_variables_sts_impl_with_env(

    create_and_attach_policy(env, policy_name, test_user, policy_document).await?;

-    // Create S3 client for test user
-    let test_client = env.create_s3_client(test_user, test_password);
-
-    // Add a small delay to allow policy to propagate
-    tokio::time::sleep(std::time::Duration::from_millis(500)).await;
+    let assumed = build_test_sts_client(&env.url, test_user, test_password, None, "policy-variable-sts")
+        .assume_role()
+        .role_arn("arn:aws:iam::123456789012:role/policy-variable")
+        .role_session_name("policy-variable-e2e")
+        .send()
+        .await?;
+    let credentials = assumed
+        .credentials()
+        .ok_or("AssumeRole response should contain temporary credentials")?;
+    let test_client = Client::from_conf(build_test_s3_config(
+        &env.url,
+        credentials.access_key_id(),
+        credentials.secret_access_key(),
+        Some(credentials.session_token()),
+        "policy-variable-sts-session",
+    ));

    // Test: User should be able to create bucket matching STS pattern
    info!("Test: User creating bucket matching STS pattern");
@@ -699,24 +672,18 @@ pub async fn test_aws_policy_variables_sts_impl_with_env(

 /// Test AWS policy variables with deny scenarios
 #[tokio::test(flavor = "multi_thread")]
-#[ignore = "Starts a rustfs server; enable when running full E2E"]
 pub async fn test_aws_policy_variables_deny() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-    test_aws_policy_variables_deny_impl().await
-}
-
-/// Implementation function for deny policy variables test
-pub async fn test_aws_policy_variables_deny_impl() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    init_logging();
    info!("Starting AWS policy variables deny test");

-    let env = PolicyTestEnvironment::with_address("127.0.0.1:9000").await?;
+    let mut env = RustFSTestEnvironment::new().await?;
+    env.start_rustfs_server(vec![]).await?;

    test_aws_policy_variables_deny_impl_with_env(&env).await
 }

-/// Implementation function for deny policy variables test with shared environment
-pub async fn test_aws_policy_variables_deny_impl_with_env(
-    env: &PolicyTestEnvironment,
+async fn test_aws_policy_variables_deny_impl_with_env(
+    env: &RustFSTestEnvironment,
 ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    // Create test user
    let test_user = "testuser5";
@@ -759,10 +726,7 @@ pub async fn test_aws_policy_variables_deny_impl_with_env(
    create_and_attach_policy(env, policy_name, test_user, policy_document).await?;

    // Create S3 client for test user
-    let test_client = env.create_s3_client(test_user, test_password);
-
-    // Add a small delay to allow policy to propagate
-    tokio::time::sleep(std::time::Duration::from_millis(500)).await;
+    let test_client = env.create_s3_client_with_credentials(test_user, test_password);

    // Test 1: User should be able to create bucket matching username pattern
    info!("Test 1: User creating bucket matching username pattern");
@@ -775,12 +739,14 @@ pub async fn test_aws_policy_variables_deny_impl_with_env(

    // Test 2: User should NOT be able to create bucket with "private" in the name (deny rule)
    info!("Test 2: User attempting to create bucket with 'private' in name (should be denied)");
-    let private_bucket_name = "private-test-bucket";
-    let create_private_result = test_client.create_bucket().bucket(private_bucket_name).send().await;
-    if create_private_result.is_ok() {
-        cleanup().await;
-        return Err("User should NOT be able to create bucket with 'private' in name due to deny rule".into());
-    }
+    let private_bucket_name = format!("{test_user}-private-bucket");
+    let denied = test_client
+        .create_bucket()
+        .bucket(&private_bucket_name)
+        .send()
+        .await
+        .expect_err("the explicit deny must reject a matching bucket name");
+    assert_eq!(denied.as_service_error().and_then(ProvideErrorMetadata::code), Some("AccessDenied"));

    // Cleanup
    info!("Cleaning up test resources");
@@ -1,100 +0,0 @@
-// Copyright 2024 RustFS Team
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-//! Custom test environment for policy variables tests
-//!
-//! This module provides a custom test environment that doesn't automatically
-//! stop servers when destroyed, addressing the server stopping issue.
-
-use aws_sdk_s3::Client;
-use aws_sdk_s3::config::{Config, Credentials, Region};
-use std::net::TcpStream;
-use std::time::Duration;
-use tokio::time::sleep;
-use tracing::{info, warn};
-
-// Default credentials
-const DEFAULT_ACCESS_KEY: &str = "rustfsadmin";
-const DEFAULT_SECRET_KEY: &str = "rustfsadmin";
-
-/// Custom test environment that doesn't automatically stop servers
-pub struct PolicyTestEnvironment {
-    pub temp_dir: String,
-    pub address: String,
-    pub url: String,
-    pub access_key: String,
-    pub secret_key: String,
-}
-
-impl PolicyTestEnvironment {
-    /// Create a new test environment with specific address
-    /// This environment won't stop any server when dropped
-    pub async fn with_address(address: &str) -> Result<Self, Box<dyn std::error::Error + Send + Sync>> {
-        let temp_dir = format!("/tmp/rustfs_policy_test_{}", uuid::Uuid::new_v4());
-        tokio::fs::create_dir_all(&temp_dir).await?;
-
-        let url = format!("http://{address}");
-
-        Ok(Self {
-            temp_dir,
-            address: address.to_string(),
-            url,
-            access_key: DEFAULT_ACCESS_KEY.to_string(),
-            secret_key: DEFAULT_SECRET_KEY.to_string(),
-        })
-    }
-
-    /// Create an AWS S3 client configured for this RustFS instance
-    pub fn create_s3_client(&self, access_key: &str, secret_key: &str) -> Client {
-        let credentials = Credentials::new(access_key, secret_key, None, None, "policy-test");
-        let config = Config::builder()
-            .credentials_provider(credentials)
-            .region(Region::new("us-east-1"))
-            .endpoint_url(&self.url)
-            .force_path_style(true)
-            .behavior_version_latest()
-            .build();
-        Client::from_conf(config)
-    }
-
-    /// Wait for RustFS server to be ready by checking TCP connectivity
-    pub async fn wait_for_server_ready(&self) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-        info!("Waiting for RustFS server to be ready on {}", self.address);
-
-        for i in 0..30 {
-            if TcpStream::connect(&self.address).is_ok() {
-                info!("✅ RustFS server is ready after {} attempts", i + 1);
-                return Ok(());
-            }
-
-            if i == 29 {
-                return Err("RustFS server failed to become ready within 30 seconds".into());
-            }
-
-            sleep(Duration::from_secs(1)).await;
-        }
-
-        Ok(())
-    }
-}
-
-// Implement Drop trait that doesn't stop servers
-impl Drop for PolicyTestEnvironment {
-    fn drop(&mut self) {
-        // Clean up temp directory only, don't stop any server
-        if let Err(e) = std::fs::remove_dir_all(&self.temp_dir) {
-            warn!("Failed to clean up temp directory {}: {}", self.temp_dir, e);
-        }
-    }
-}
@@ -1,230 +0,0 @@
-// Copyright 2024 RustFS Team
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-use crate::common::init_logging;
-use crate::policy::test_env::PolicyTestEnvironment;
-use std::time::Instant;
-use tokio::time::{Duration, sleep};
-use tracing::{error, info};
-
-/// Test case definition
-#[derive(Debug, Clone)]
-pub struct TestDefinition {
-    pub name: String,
-    pub is_critical: bool,
-}
-
-impl TestDefinition {
-    pub fn new(name: impl Into<String>, is_critical: bool) -> Self {
-        Self {
-            name: name.into(),
-            is_critical,
-        }
-    }
-}
-
-/// Test result
-#[derive(Debug, Clone)]
-pub struct TestResult {
-    pub test_name: String,
-    pub success: bool,
-    pub error_message: Option<String>,
-}
-
-impl TestResult {
-    pub fn success(test_name: String) -> Self {
-        Self {
-            test_name,
-            success: true,
-            error_message: None,
-        }
-    }
-
-    pub fn failure(test_name: String, error: String) -> Self {
-        Self {
-            test_name,
-            success: false,
-            error_message: Some(error),
-        }
-    }
-}
-
-/// Test suite configuration
-#[derive(Debug, Clone, Default)]
-pub struct TestSuiteConfig {
-    pub include_critical_only: bool,
-}
-
-/// Policy test suite
-pub struct PolicyTestSuite {
-    tests: Vec<TestDefinition>,
-    config: TestSuiteConfig,
-}
-
-impl PolicyTestSuite {
-    /// Create default test suite
-    pub fn new() -> Self {
-        let tests = vec![
-            TestDefinition::new("test_aws_policy_variables_single_value", true),
-            TestDefinition::new("test_aws_policy_variables_multi_value", true),
-            TestDefinition::new("test_aws_policy_variables_concatenation", true),
-            TestDefinition::new("test_aws_policy_variables_nested", true),
-            TestDefinition::new("test_aws_policy_variables_deny", true),
-            TestDefinition::new("test_aws_policy_variables_sts", true),
-        ];
-
-        Self {
-            tests,
-            config: TestSuiteConfig::default(),
-        }
-    }
-
-    /// Configure test suite
-    pub fn with_config(mut self, config: TestSuiteConfig) -> Self {
-        self.config = config;
-        self
-    }
-
-    /// Run test suite
-    pub async fn run_test_suite(&self) -> Vec<TestResult> {
-        init_logging();
-        info!("Starting Policy Variables test suite");
-
-        let start_time = Instant::now();
-        let mut results = Vec::new();
-
-        // Create test environment
-        let env = match PolicyTestEnvironment::with_address("127.0.0.1:9000").await {
-            Ok(env) => env,
-            Err(e) => {
-                error!("Failed to create test environment: {}", e);
-                return vec![TestResult::failure("env_creation".into(), e.to_string())];
-            }
-        };
-
-        // Wait for server to be ready
-        if env.wait_for_server_ready().await.is_err() {
-            error!("Server is not ready");
-            return vec![TestResult::failure("server_check".into(), "Server not ready".into())];
-        }
-
-        // Filter tests
-        let tests_to_run: Vec<&TestDefinition> = self
-            .tests
-            .iter()
-            .filter(|test| !self.config.include_critical_only || test.is_critical)
-            .collect();
-
-        info!("Scheduled {} tests", tests_to_run.len());
-
-        // Run tests
-        for (i, test_def) in tests_to_run.iter().enumerate() {
-            info!("Running test {}/{}: {}", i + 1, tests_to_run.len(), test_def.name);
-            let test_start = Instant::now();
-
-            let result = self.run_single_test(test_def, &env).await;
-            let test_duration = test_start.elapsed();
-
-            match result {
-                Ok(_) => {
-                    info!("Test passed: {} ({:.2}s)", test_def.name, test_duration.as_secs_f64());
-                    results.push(TestResult::success(test_def.name.clone()));
-                }
-                Err(e) => {
-                    error!("Test failed: {} ({:.2}s): {}", test_def.name, test_duration.as_secs_f64(), e);
-                    results.push(TestResult::failure(test_def.name.clone(), e.to_string()));
-                }
-            }
-
-            // Delay between tests to avoid resource conflicts
-            if i < tests_to_run.len() - 1 {
-                sleep(Duration::from_secs(2)).await;
-            }
-        }
-
-        // Print summary
-        self.print_summary(&results, start_time.elapsed());
-
-        results
-    }
-
-    /// Run a single test
-    async fn run_single_test(
-        &self,
-        test_def: &TestDefinition,
-        env: &PolicyTestEnvironment,
-    ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-        match test_def.name.as_str() {
-            "test_aws_policy_variables_single_value" => {
-                super::policy_variables_test::test_aws_policy_variables_single_value_impl_with_env(env).await
-            }
-            "test_aws_policy_variables_multi_value" => {
-                super::policy_variables_test::test_aws_policy_variables_multi_value_impl_with_env(env).await
-            }
-            "test_aws_policy_variables_concatenation" => {
-                super::policy_variables_test::test_aws_policy_variables_concatenation_impl_with_env(env).await
-            }
-            "test_aws_policy_variables_nested" => {
-                super::policy_variables_test::test_aws_policy_variables_nested_impl_with_env(env).await
-            }
-            "test_aws_policy_variables_deny" => {
-                super::policy_variables_test::test_aws_policy_variables_deny_impl_with_env(env).await
-            }
-            "test_aws_policy_variables_sts" => {
-                super::policy_variables_test::test_aws_policy_variables_sts_impl_with_env(env).await
-            }
-            _ => Err(format!("Test {} not implemented", test_def.name).into()),
-        }
-    }
-
-    /// Print test summary
-    fn print_summary(&self, results: &[TestResult], total_duration: Duration) {
-        info!("=== Test Suite Summary ===");
-        info!("Total duration: {:.2}s", total_duration.as_secs_f64());
-        info!("Total tests: {}", results.len());
-
-        let passed = results.iter().filter(|r| r.success).count();
-        let failed = results.len() - passed;
-        let success_rate = (passed as f64 / results.len() as f64) * 100.0;
-
-        info!("Passed: {} | Failed: {}", passed, failed);
-        info!("Success rate: {:.1}%", success_rate);
-
-        if failed > 0 {
-            error!("Failed tests:");
-            for result in results.iter().filter(|r| !r.success) {
-                error!("  - {}: {}", result.test_name, result.error_message.as_ref().unwrap());
-            }
-        }
-    }
-}
-
-/// Test suite
-#[tokio::test]
-#[ignore = "Connects to existing rustfs server"]
-async fn test_policy_critical_suite() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-    let config = TestSuiteConfig {
-        include_critical_only: true,
-    };
-    let suite = PolicyTestSuite::new().with_config(config);
-    let results = suite.run_test_suite().await;
-
-    let failed = results.iter().filter(|r| !r.success).count();
-    if failed > 0 {
-        return Err(format!("Critical tests failed: {failed} failures").into());
-    }
-
-    info!("All critical tests passed");
-    Ok(())
-}
@@ -87,7 +87,9 @@ fn valid_config() -> PresigningConfig {
 }

 /// Flip bytes inside the `X-Amz-Signature=` query value without changing its
-/// length, producing a structurally valid but incorrect signature.
+/// length, producing a structurally valid but incorrect signature. Every hex
+/// digit is replaced by its complement (15 - v), which has no fixed point, so
+/// the tamper changes the value no matter which digits the signature contains.
 fn tamper_signature(uri: &str) -> String {
    let marker = "X-Amz-Signature=";
    let idx = uri.find(marker).expect("presigned uri must carry X-Amz-Signature") + marker.len();
@@ -96,10 +98,9 @@ fn tamper_signature(uri: &str) -> String {
    let (sig, tail) = rest.split_at(end);
    let tampered: String = sig
        .chars()
-        .map(|c| match c {
-            '0' => 'f',
-            'a' => '0',
-            other => other,
+        .map(|c| {
+            let v = c.to_digit(16).expect("X-Amz-Signature value must be hex");
+            char::from_digit(15 - v, 16).expect("complement of a hex digit is a hex digit")
        })
        .collect();
    assert_ne!(sig, tampered, "tamper must actually change the signature hex");
@@ -340,7 +341,18 @@ async fn tampered_presigned_put_returns_signature_does_not_match() -> Result<(),
    assert_error_code(&body, "SignatureDoesNotMatch");

    // The rejected write must not have created the object.
-    let head = env.create_s3_client().head_object().bucket(BUCKET).key(key).send().await;
-    assert!(head.is_err(), "tampered presigned PUT must not store the object");
+    let error = env
+        .create_s3_client()
+        .head_object()
+        .bucket(BUCKET)
+        .key(key)
+        .send()
+        .await
+        .expect_err("tampered presigned PUT must not store the object");
+    assert_eq!(
+        error.raw_response().map(|response| response.status().as_u16()),
+        Some(404),
+        "tampered presigned PUT absence probe must return HTTP 404, got {error:?}"
+    );
    Ok(())
 }
@@ -453,27 +453,28 @@ fn watch_session_lifecycle_events(child: &mut Child, counters: Arc<SessionCounte
 }

 /// Count TCP connections in CLOSE_WAIT against the given local port
-/// by shelling out to ss -tn state CLOSE-WAIT. The check is
-/// best-effort: if ss is missing on the host the function returns
-/// Ok(None) and the caller skips the assertion. The contract is zero
+/// by shelling out to ss -tn state CLOSE-WAIT. The oracle fails closed
+/// if ss is missing or cannot inspect socket state. The contract is zero
 /// CLOSE_WAIT entries attributable to the test.
 #[cfg(target_os = "linux")]
-async fn count_close_wait_on_port(port: u16) -> Result<Option<usize>> {
-    let output = match Command::new("ss").args(["-tn", "state", "CLOSE-WAIT"]).output().await {
-        Ok(o) => o,
-        Err(_) => return Ok(None),
-    };
+async fn count_close_wait_on_port(port: u16, test_id: &str) -> Result<usize> {
+    let port_filter = format!("sport = :{port}");
+    let output = Command::new("ss")
+        .args(["-H", "-t", "-n", "state", "close-wait"])
+        .arg(&port_filter)
+        .output()
+        .await
+        .map_err(|error| anyhow!("{test_id} failed to run ss CLOSE_WAIT oracle: {error}"))?;
    if !output.status.success() {
-        return Ok(None);
+        return Err(anyhow!(
+            "{test_id} ss CLOSE_WAIT oracle exited with {}: {}",
+            output.status,
+            String::from_utf8_lossy(&output.stderr)
+        ));
    }
    let stdout = String::from_utf8_lossy(&output.stdout);
-    let needle_local = format!(":{port} ");
-    let needle_local_eol = format!(":{port}\n");
-    let count = stdout
-        .lines()
-        .filter(|l| l.contains(&needle_local) || l.contains(needle_local_eol.trim_end()))
-        .count();
-    Ok(Some(count))
+    let count = stdout.lines().filter(|line| !line.trim().is_empty()).count();
+    Ok(count)
 }

 // CMPTST-01: medium-binary upload then download with SHA256 compare.
@@ -1400,7 +1401,7 @@ pub(crate) mod cmptst_24 {
    //    the JoinSet flushes finished tasks before the assertion runs.
    // 5. Assert the entered/finished session counters balance and that
    //    no CLOSE_WAIT sockets remain on the bind port (Linux ss(8)
-    //    only; the assertion skips with a warn if ss is unavailable).
+    //    only; missing or failed socket inspection is an error).
    pub(crate) async fn run_concurrent_half_close_no_leak() -> Result<()> {
        let env = ProtocolTestEnvironment::new().map_err(|e| anyhow!("{}", e))?;
        let host_key_dir = PathBuf::from(&env.temp_dir).join("sftp_host_keys");
@@ -1520,15 +1521,13 @@ pub(crate) mod cmptst_24 {
                ));
            }

-            match count_close_wait_on_port(HALF_CLOSE_SFTP_PORT).await? {
-                Some(0) => info!("{COMPLIANCE_TEST_OUTPUT_ID}: zero CLOSE_WAIT entries against port {HALF_CLOSE_SFTP_PORT}"),
-                Some(n) => {
-                    return Err(anyhow!(
-                        "{COMPLIANCE_TEST_OUTPUT_ID} {n} CLOSE_WAIT entries against port {HALF_CLOSE_SFTP_PORT}, expected 0"
-                    ));
-                }
-                None => info!("{COMPLIANCE_TEST_OUTPUT_ID}: ss(8) unavailable, skipping CLOSE_WAIT assertion"),
+            let close_wait = count_close_wait_on_port(HALF_CLOSE_SFTP_PORT, COMPLIANCE_TEST_OUTPUT_ID).await?;
+            if close_wait != 0 {
+                return Err(anyhow!(
+                    "{COMPLIANCE_TEST_OUTPUT_ID} {close_wait} CLOSE_WAIT entries against port {HALF_CLOSE_SFTP_PORT}, expected 0"
+                ));
            }
+            info!("{COMPLIANCE_TEST_OUTPUT_ID}: zero CLOSE_WAIT entries against port {HALF_CLOSE_SFTP_PORT}");

            // Drop the keepalive vector now so the test process does
            // not leave the half-closed sockets dangling past the
@@ -1947,7 +1946,7 @@ pub(crate) mod cmptst_25 {
    //    plus two 15 s ticks worst-case = 60 s) to detect CLOSE_WAIT
    //    via /proc/net/tcp and cancel the parked session.
    // 5. Assert the session task counters balance and CLOSE_WAIT count
-    //    is zero (ss(8) only; skips with a warn when ss is missing).
+    //    is zero (ss(8) only; missing or failed socket inspection is an error).
    pub(crate) async fn run_wedge_kill_after_silence_in_close_wait() -> Result<()> {
        let env = ProtocolTestEnvironment::new().map_err(|e| anyhow!("{}", e))?;
        let host_key_dir = PathBuf::from(&env.temp_dir).join("sftp_host_keys");
@@ -2056,15 +2055,13 @@ pub(crate) mod cmptst_25 {
                ));
            }

-            match count_close_wait_on_port(WEDGE_SFTP_PORT).await? {
-                Some(0) => info!("{COMPLIANCE_TEST_OUTPUT_ID}: zero CLOSE_WAIT entries against port {WEDGE_SFTP_PORT}"),
-                Some(n) => {
-                    return Err(anyhow!(
-                        "{COMPLIANCE_TEST_OUTPUT_ID} {n} CLOSE_WAIT entries against port {WEDGE_SFTP_PORT}, expected 0"
-                    ));
-                }
-                None => info!("{COMPLIANCE_TEST_OUTPUT_ID}: ss(8) unavailable, skipping CLOSE_WAIT assertion"),
+            let close_wait = count_close_wait_on_port(WEDGE_SFTP_PORT, COMPLIANCE_TEST_OUTPUT_ID).await?;
+            if close_wait != 0 {
+                return Err(anyhow!(
+                    "{COMPLIANCE_TEST_OUTPUT_ID} {close_wait} CLOSE_WAIT entries against port {WEDGE_SFTP_PORT}, expected 0"
+                ));
            }
+            info!("{COMPLIANCE_TEST_OUTPUT_ID}: zero CLOSE_WAIT entries against port {WEDGE_SFTP_PORT}");

            drop(keepalive);
            info!("PASS {COMPLIANCE_TEST_OUTPUT_ID}: wedged sessions killed by the watchdog");
@@ -31,16 +31,11 @@
 //!
 //! Advisory: <https://github.com/rustfs/rustfs/security/advisories/GHSA-3p3x-734c-h5vx>

-use crate::common::local_http_client;
 use crate::common::rustfs_binary_path_with_features;
+use crate::common::{AdminTransport, admin_add_canned_policy_via, admin_attach_user_policy_via, admin_create_user_via};
 use crate::protocols::test_env::{DEFAULT_ACCESS_KEY, DEFAULT_SECRET_KEY, ProtocolTestEnvironment};
 use anyhow::Result;
-use base64::Engine;
-use http::header::{CONTENT_TYPE, HOST};
 use reqwest::Client;
-use rustfs_signer::constants::UNSIGNED_PAYLOAD;
-use rustfs_signer::sign_v4;
-use s3s::Body;
 use tokio::process::Command;
 use tracing::info;

@@ -64,96 +59,47 @@ fn basic_auth_header() -> String {

 fn basic_auth_header_for(access_key: &str, secret_key: &str) -> String {
    let credentials = format!("{}:{}", access_key, secret_key);
-    let encoded = base64::engine::general_purpose::STANDARD.encode(credentials);
+    let encoded = base64_simd::STANDARD.encode_to_string(credentials);
    format!("Basic {}", encoded)
 }

-async fn signed_admin_request(
-    method: http::Method,
-    url: &str,
-    body: Option<Vec<u8>>,
-    content_type: Option<&str>,
-) -> Result<reqwest::Response> {
-    let uri = url.parse::<http::Uri>()?;
-    let authority = uri
-        .authority()
-        .ok_or_else(|| anyhow::anyhow!("request URL missing authority"))?
-        .to_string();
-    let mut request = http::Request::builder().method(method.clone()).uri(uri);
-    request = request.header(HOST, authority);
-    request = request.header("x-amz-content-sha256", UNSIGNED_PAYLOAD);
-    if let Some(content_type) = content_type {
-        request = request.header(CONTENT_TYPE, content_type);
-    }
-
-    let content_len = body.as_ref().map(|body| body.len() as i64).unwrap_or_default();
-    let signed = sign_v4(
-        request.body(Body::empty())?,
-        content_len,
+async fn admin_create_user(base_url: &str, username: &str, secret_key: &str) -> Result<()> {
+    admin_create_user_via(
+        AdminTransport::Signed,
+        base_url,
        DEFAULT_ACCESS_KEY,
        DEFAULT_SECRET_KEY,
-        "",
-        "us-east-1",
-    );
-
-    let reqwest_method = reqwest::Method::from_bytes(method.as_str().as_bytes())?;
-    let mut request_builder = local_http_client().request(reqwest_method, url);
-    for (name, value) in signed.headers() {
-        request_builder = request_builder.header(name, value);
-    }
-    if let Some(body) = body {
-        request_builder = request_builder.body(body);
-    }
-
-    Ok(request_builder.send().await?)
-}
-
-async fn admin_create_user(base_url: &str, username: &str, secret_key: &str) -> Result<()> {
-    let url = format!("{}/rustfs/admin/v3/add-user?accessKey={}", base_url, username);
-    let body = serde_json::json!({
-        "secretKey": secret_key,
-        "status": "enabled"
-    });
-    let response =
-        signed_admin_request(http::Method::PUT, &url, Some(body.to_string().into_bytes()), Some("application/json")).await?;
-
-    if response.status() != reqwest::StatusCode::OK {
-        let status = response.status();
-        let body = response.text().await.unwrap_or_default();
-        anyhow::bail!("create user failed: {status} {body}");
-    }
-
-    Ok(())
+        username,
+        secret_key,
+    )
+    .await
+    .map_err(|e| anyhow::anyhow!(e))
 }

 async fn admin_add_canned_policy(base_url: &str, policy_name: &str, policy: &serde_json::Value) -> Result<()> {
-    let url = format!("{}/rustfs/admin/v3/add-canned-policy?name={}", base_url, policy_name);
-    let response =
-        signed_admin_request(http::Method::PUT, &url, Some(policy.to_string().into_bytes()), Some("application/json")).await?;
-
-    if response.status() != reqwest::StatusCode::OK {
-        let status = response.status();
-        let body = response.text().await.unwrap_or_default();
-        anyhow::bail!("add canned policy failed: {status} {body}");
-    }
-
-    Ok(())
+    admin_add_canned_policy_via(
+        AdminTransport::Signed,
+        base_url,
+        DEFAULT_ACCESS_KEY,
+        DEFAULT_SECRET_KEY,
+        policy_name,
+        &policy.to_string(),
+    )
+    .await
+    .map_err(|e| anyhow::anyhow!(e))
 }

 async fn admin_attach_policy_to_user(base_url: &str, policy_name: &str, username: &str) -> Result<()> {
-    let url = format!(
-        "{}/rustfs/admin/v3/set-user-or-group-policy?policyName={}&userOrGroup={}&isGroup=false",
-        base_url, policy_name, username
-    );
-    let response = signed_admin_request(http::Method::PUT, &url, Some(Vec::new()), None).await?;
-
-    if response.status() != reqwest::StatusCode::OK {
-        let status = response.status();
-        let body = response.text().await.unwrap_or_default();
-        anyhow::bail!("attach policy failed: {status} {body}");
-    }
-
-    Ok(())
+    admin_attach_user_policy_via(
+        AdminTransport::Signed,
+        base_url,
+        DEFAULT_ACCESS_KEY,
+        DEFAULT_SECRET_KEY,
+        policy_name,
+        username,
+    )
+    .await
+    .map_err(|e| anyhow::anyhow!(e))
 }

 /// Test WebDAV: MKCOL (create bucket), PUT, GET, DELETE, PROPFIND operations
@@ -14,6 +14,7 @@

 use crate::common::{RustFSTestEnvironment, admin_request, awscurl_delete, awscurl_get, awscurl_post, awscurl_put, init_logging};
 use aws_sdk_s3::Client;
+use aws_sdk_s3::error::ProvideErrorMetadata;
 use http::{Method, StatusCode};
 use tokio::time::{Duration, sleep, timeout};
 use tracing::{debug, info};
@@ -132,19 +133,13 @@ impl QuotaTestEnv {
    pub async fn object_exists(&self, key: &str) -> Result<bool, Box<dyn std::error::Error + Send + Sync>> {
        match self.client.head_object().bucket(&self.bucket_name).key(key).send().await {
            Ok(_) => Ok(true),
-            Err(e) => {
-                // Check for any 404-related errors and return false instead of propagating
-                let error_str = e.to_string();
-                if error_str.contains("404") || error_str.contains("Not Found") || error_str.contains("NotFound") {
+            Err(error) => {
+                let status = error.raw_response().map(|response| response.status().as_u16());
+                let code = error.as_service_error().and_then(ProvideErrorMetadata::code);
+                if status == Some(404) && matches!(code, Some("NotFound" | "NoSuchKey")) {
                    Ok(false)
                } else {
-                    // Also check the error code directly
-                    if let Some(service_err) = e.as_service_error()
-                        && service_err.is_not_found()
-                    {
-                        return Ok(false);
-                    }
-                    Err(e.into())
+                    Err(error.into())
                }
            }
        }
@@ -155,6 +150,22 @@ impl QuotaTestEnv {
        Ok(stats.get("current_usage").and_then(|v| v.as_u64()).unwrap_or(0))
    }

+    async fn wait_for_bucket_usage(&self, expected: u64) -> Result<u64, Box<dyn std::error::Error + Send + Sync>> {
+        let convergence = async {
+            loop {
+                let usage = self.get_bucket_usage().await?;
+                if usage == expected {
+                    return Ok::<u64, Box<dyn std::error::Error + Send + Sync>>(usage);
+                }
+                sleep(Duration::from_millis(100)).await;
+            }
+        };
+        match timeout(Duration::from_secs(30), convergence).await {
+            Ok(result) => result,
+            Err(_) => Err(format!("bucket usage did not converge to {expected} bytes within 30 seconds").into()),
+        }
+    }
+
    pub async fn set_bucket_quota_for(
        &self,
        bucket: &str,
@@ -262,7 +273,46 @@ impl QuotaTestEnv {
 #[cfg(test)]
 mod integration_tests {
    use super::*;
-    use aws_sdk_s3::error::ProvideErrorMetadata;
+
+    fn assert_error_response(status: StatusCode, body: &str, expected_status: StatusCode, expected_code: &str) {
+        assert_eq!(status, expected_status, "unexpected error status: {status} {body}");
+        assert!(
+            body.contains(&format!("<Code>{expected_code}</Code>")),
+            "expected {expected_code}, got: {body}"
+        );
+    }
+
+    fn assert_quota_rejection<E>(status: Option<u16>, service_error: Option<&E>, error: &impl std::fmt::Debug)
+    where
+        E: ProvideErrorMetadata + std::fmt::Debug,
+    {
+        assert_eq!(status, Some(400), "quota rejection must return HTTP 400: {error:?}");
+        let service_error = service_error.expect("quota rejection must be an S3 service error");
+        assert_eq!(service_error.code(), Some("InvalidRequest"), "unexpected quota error: {error:?}");
+        assert!(
+            service_error
+                .message()
+                .is_some_and(|message| message.starts_with("Bucket quota exceeded")),
+            "operation must fail specifically at quota admission: {error:?}"
+        );
+    }
+
+    async fn assert_put_rejected_by_quota(env: &QuotaTestEnv, key: &str, size_bytes: usize) {
+        let error = env
+            .client
+            .put_object()
+            .bucket(&env.bucket_name)
+            .key(key)
+            .body(aws_sdk_s3::primitives::ByteStream::from(vec![0u8; size_bytes]))
+            .send()
+            .await
+            .expect_err("PUT above quota must be rejected");
+        assert_quota_rejection(
+            error.raw_response().map(|response| response.status().as_u16()),
+            error.as_service_error(),
+            &error,
+        );
+    }

    #[tokio::test]
    async fn test_quota_basic_operations() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
@@ -288,8 +338,7 @@ mod integration_tests {
        assert!(env.object_exists("test2.txt").await?);

        // Try to upload 1KB more (should fail due to quota)
-        let upload_result = env.upload_object("test3.txt", 1024).await;
-        assert!(upload_result.is_err());
+        assert_put_rejected_by_quota(&env, "test3.txt", 1024).await;
        assert!(!env.object_exists("test3.txt").await?);

        // Clean up
@@ -340,10 +389,10 @@ mod integration_tests {
        let err = put_aws_chunked("over-quota.bin", 16 * 1024)
            .await
            .expect_err("declared aws-chunked PUT over quota must be rejected");
-        let err_debug = format!("{err:?}");
-        assert!(
-            !err_debug.contains("UnexpectedContent"),
-            "over-quota rejection must be the quota error, not UnexpectedContent: {err_debug}"
+        assert_quota_rejection(
+            err.raw_response().map(|response| response.status().as_u16()),
+            err.as_service_error(),
+            &err,
        );
        assert!(!env.object_exists("over-quota.bin").await?);

@@ -444,8 +493,8 @@ mod integration_tests {
            .send()
            .await?;

-        // Check updated usage
-        let updated_usage = env.get_bucket_usage().await?;
+        // A completed scanner generation releases the conservative quota floor after a delete.
+        let updated_usage = env.wait_for_bucket_usage(256 * 1024).await?;
        assert_eq!(updated_usage, 256 * 1024);

        env.cleanup_bucket().await?;
@@ -565,24 +614,35 @@ mod integration_tests {
        env.create_bucket().await?;

        // Test invalid quota type
-        let url = format!("{}/rustfs/admin/v3/quota/{}", env.env.url, env.bucket_name);
+        let quota_path = format!("/rustfs/admin/v3/quota/{}", env.bucket_name);

        let invalid_config = serde_json::json!({
            "quota": 1024,
            "quota_type": "SOFT"  // Invalid type
        });

-        let response = awscurl_put(&url, &invalid_config.to_string(), &env.env.access_key, &env.env.secret_key).await;
-        assert!(response.is_err());
-        let error_msg = response.unwrap_err().to_string();
-        assert!(error_msg.contains("InvalidArgument"));
+        let (status, body) = admin_request(
+            &env.env.url,
+            Method::PUT,
+            &quota_path,
+            Some(invalid_config.to_string()),
+            &env.env.access_key,
+            &env.env.secret_key,
+        )
+        .await?;
+        assert_error_response(status, &body, StatusCode::BAD_REQUEST, "InvalidArgument");

        // Test operations on non-existent bucket
-        let url = format!("{}/rustfs/admin/v3/quota/non-existent-bucket", env.env.url);
-        let response = awscurl_get(&url, &env.env.access_key, &env.env.secret_key).await;
-        assert!(response.is_err());
-        let error_msg = response.unwrap_err().to_string();
-        assert!(error_msg.contains("NoSuchBucket"));
+        let (status, body) = admin_request(
+            &env.env.url,
+            Method::GET,
+            "/rustfs/admin/v3/quota/non-existent-bucket",
+            None,
+            &env.env.access_key,
+            &env.env.secret_key,
+        )
+        .await?;
+        assert_error_response(status, &body, StatusCode::NOT_FOUND, "NoSuchBucket");

        env.cleanup_bucket().await?;

@@ -636,10 +696,16 @@ mod integration_tests {
            "quota": 1024,
            "quota_type": "SOFT"
        });
-        let response = awscurl_put(&url, &invalid_config.to_string(), &env.env.access_key, &env.env.secret_key).await;
-        assert!(response.is_err());
-        let error_msg = response.unwrap_err().to_string();
-        assert!(error_msg.contains("InvalidArgument"));
+        let (status, body) = admin_request(
+            &env.env.url,
+            Method::PUT,
+            &format!("/rustfs/admin/v3/quota/{}", env.bucket_name),
+            Some(invalid_config.to_string()),
+            &env.env.access_key,
+            &env.env.secret_key,
+        )
+        .await?;
+        assert_error_response(status, &body, StatusCode::BAD_REQUEST, "InvalidArgument");

        env.cleanup_bucket().await?;

@@ -682,18 +748,21 @@ mod integration_tests {
        assert!(resp.contains("quota_limit"));

        // Normal user sets quota — should be denied
-        let set_resp = awscurl_put(
-            &get_url,
-            &serde_json::json!({"quota": 2048, "quota_type": "HARD"}).to_string(),
+        let quota_path = format!("/rustfs/admin/v3/quota/{}", env.bucket_name);
+        let (status, body) = admin_request(
+            &env.env.url,
+            Method::PUT,
+            &quota_path,
+            Some(serde_json::json!({"quota": 2048, "quota_type": "HARD"}).to_string()),
            normal_ak,
            normal_sk,
        )
-        .await;
-        assert!(set_resp.is_err(), "normal user should not be able to set quota");
+        .await?;
+        assert_error_response(status, &body, StatusCode::FORBIDDEN, "AccessDenied");

        // Normal user clears quota — should be denied
-        let del_resp = awscurl_delete(&get_url, normal_ak, normal_sk).await;
-        assert!(del_resp.is_err(), "normal user should not be able to clear quota");
+        let (status, body) = admin_request(&env.env.url, Method::DELETE, &quota_path, None, normal_ak, normal_sk).await?;
+        assert_error_response(status, &body, StatusCode::FORBIDDEN, "AccessDenied");

        env.cleanup_bucket().await?;
        Ok(())
@@ -733,7 +802,12 @@ mod integration_tests {
            .send()
            .await;

-        assert!(copy_result.is_err());
+        let copy_error = copy_result.expect_err("copy above quota must be rejected");
+        assert_quota_rejection(
+            copy_error.raw_response().map(|response| response.status().as_u16()),
+            copy_error.as_service_error(),
+            &copy_error,
+        );
        assert!(!env.object_exists("copy2.txt").await?);

        env.cleanup_bucket().await?;
@@ -756,8 +830,7 @@ mod integration_tests {
        env.upload_object("file2.txt", 1024 * 1024).await?;

        // Verify quota is full
-        let upload_result = env.upload_object("file3.txt", 1024).await;
-        assert!(upload_result.is_err());
+        assert_put_rejected_by_quota(&env, "file3.txt", 1024).await;

        // Delete multiple objects using batch delete
        let objects = vec![
@@ -857,9 +930,7 @@ mod integration_tests {

        // Test 2: Multipart upload exceeds quota (should fail)
        // Upload 6MB filler (total now: 5MB + 6MB = 11MB > 10MB quota)
-        let upload_filler = env.upload_object("filler.txt", 6 * 1024 * 1024).await;
-        // This should fail due to quota
-        assert!(upload_filler.is_err());
+        assert_put_rejected_by_quota(&env, "filler.txt", 6 * 1024 * 1024).await;

        // Verify filler doesn't exist
        assert!(!env.object_exists("filler.txt").await?);
@@ -915,7 +986,11 @@ mod integration_tests {
            .await;

        let complete_error = complete_result.expect_err("multipart completion above quota must be rejected");
-        assert_eq!(complete_error.as_service_error().and_then(|error| error.code()), Some("InvalidRequest"));
+        assert_quota_rejection(
+            complete_error.raw_response().map(|response| response.status().as_u16()),
+            complete_error.as_service_error(),
+            &complete_error,
+        );
        assert!(!env.object_exists("over_quota.txt").await?);

        let staged_parts = env
@@ -23,8 +23,9 @@ use rustfs_protos::{
    proto_gen::node_service::{
        BatchGenerallyLockRequest, BatchGenerallyLockResponse, BatchReadVersionRequest, BatchReadVersionResponse,
        GenerallyLockRequest, GenerallyLockResponse, GenerallyLockResult, PingRequest, PingResponse,
-        SnapshotLeaseMutationResponse, SnapshotLeaseReleaseRequest, SnapshotLeaseRenewRequest, SnapshotLeaseRequest,
-        SnapshotLeaseResponse, node_service_server::NodeService,
+        ScannerPublicationLeaseReleaseRequest, ScannerPublicationLeaseReleaseResponse, ScannerPublicationLeaseRequest,
+        ScannerPublicationLeaseResponse, SnapshotLeaseMutationResponse, SnapshotLeaseReleaseRequest, SnapshotLeaseRenewRequest,
+        SnapshotLeaseRequest, SnapshotLeaseResponse, node_service_server::NodeService,
    },
 };
 use std::pin::Pin;
@@ -126,6 +127,20 @@ impl NodeService for MinimalLockNodeService {
        Err(Status::unimplemented("MinimalLockNodeService only supports lock RPCs"))
    }

+    async fn acquire_scanner_publication_lease(
+        &self,
+        _request: Request<ScannerPublicationLeaseRequest>,
+    ) -> Result<Response<ScannerPublicationLeaseResponse>, Status> {
+        Err(Status::unimplemented("MinimalLockNodeService only supports lock RPCs"))
+    }
+
+    async fn release_scanner_publication_lease(
+        &self,
+        _request: Request<ScannerPublicationLeaseReleaseRequest>,
+    ) -> Result<Response<ScannerPublicationLeaseReleaseResponse>, Status> {
+        Err(Status::unimplemented("MinimalLockNodeService only supports lock RPCs"))
+    }
+
    async fn lock(&self, request: Request<GenerallyLockRequest>) -> Result<Response<GenerallyLockResponse>, Status> {
        let request = request.into_inner();
        let args: LockRequest = match serde_json::from_str(&request.args) {
@@ -13,55 +13,37 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

-use aws_config::meta::region::RegionProviderChain;
+use crate::common::{RustFSTestEnvironment, init_logging};
 use aws_sdk_s3::Client;
-use aws_sdk_s3::config::{Credentials, Region};
+use aws_sdk_s3::error::ProvideErrorMetadata;
 use aws_sdk_s3::types::{
    CsvInput, CsvOutput, ExpressionType, FileHeaderInfo, InputSerialization, JsonInput, JsonOutput, JsonType, OutputSerialization,
 };
 use bytes::Bytes;
 use std::error::Error;
+use std::time::Duration;

-const ENDPOINT: &str = "http://localhost:9000";
-const ACCESS_KEY: &str = "rustfsadmin";
-const SECRET_KEY: &str = "rustfsadmin";
 const BUCKET: &str = "test-sql-bucket";
 const CSV_OBJECT: &str = "test-data.csv";
 const JSON_OBJECT: &str = "test-data.json";
+const SELECT_RESPONSE_TIMEOUT: Duration = Duration::from_secs(30);

-async fn create_aws_s3_client() -> Result<Client, Box<dyn Error>> {
-    let region_provider = RegionProviderChain::default_provider().or_else(Region::new("us-east-1"));
-    let shared_config = aws_config::defaults(aws_config::BehaviorVersion::latest())
-        .region(region_provider)
-        .credentials_provider(Credentials::new(ACCESS_KEY, SECRET_KEY, None, None, "static"))
-        .endpoint_url(ENDPOINT)
-        .load()
-        .await;
+type TestResult<T> = Result<T, Box<dyn Error + Send + Sync>>;

-    let client = Client::from_conf(
-        aws_sdk_s3::Config::from(&shared_config)
-            .to_builder()
-            .force_path_style(true) // Important for S3-compatible services
-            .build(),
-    );
-
-    Ok(client)
+async fn create_test_environment() -> TestResult<(RustFSTestEnvironment, Client)> {
+    init_logging();
+    let mut env = RustFSTestEnvironment::new().await?;
+    env.start_rustfs_server(vec![]).await?;
+    let client = env.create_s3_client();
+    Ok((env, client))
 }

-async fn setup_test_bucket(client: &Client) -> Result<(), Box<dyn Error>> {
-    match client.create_bucket().bucket(BUCKET).send().await {
-        Ok(_) => {}
-        Err(e) => {
-            let error_str = e.to_string();
-            if !error_str.contains("BucketAlreadyOwnedByYou") && !error_str.contains("BucketAlreadyExists") {
-                return Err(e.into());
-            }
-        }
-    }
+async fn setup_test_bucket(client: &Client) -> TestResult<()> {
+    client.create_bucket().bucket(BUCKET).send().await?;
    Ok(())
 }

-async fn upload_test_csv(client: &Client) -> Result<(), Box<dyn Error>> {
+async fn upload_test_csv(client: &Client) -> TestResult<()> {
    let csv_data = "name,age,city\nAlice,30,New York\nBob,25,Los Angeles\nCharlie,35,Chicago\nDiana,28,Boston";

    client
@@ -75,7 +57,7 @@ async fn upload_test_csv(client: &Client) -> Result<(), Box<dyn Error>> {
    Ok(())
 }

-async fn upload_test_json(client: &Client) -> Result<(), Box<dyn Error>> {
+async fn upload_test_json(client: &Client) -> TestResult<()> {
    let json_data = r#"{"name":"Alice","age":30,"city":"New York"}
 {"name":"Bob","age":25,"city":"Los Angeles"}
 {"name":"Charlie","age":35,"city":"Chicago"}
@@ -93,33 +75,38 @@ async fn upload_test_json(client: &Client) -> Result<(), Box<dyn Error>> {

 async fn process_select_response(
    mut event_stream: aws_sdk_s3::operation::select_object_content::SelectObjectContentOutput,
-) -> Result<String, Box<dyn Error>> {
-    let mut total_data = Vec::new();
+) -> TestResult<String> {
+    tokio::time::timeout(SELECT_RESPONSE_TIMEOUT, async move {
+        let mut total_data = Vec::new();
+        let mut saw_end = false;

-    while let Ok(Some(event)) = event_stream.payload.recv().await {
-        match event {
-            aws_sdk_s3::types::SelectObjectContentEventStream::Records(records_event) => {
-                if let Some(payload) = records_event.payload {
-                    let data = payload.into_inner();
-                    total_data.extend_from_slice(&data);
+        while let Some(event) = event_stream.payload.recv().await? {
+            match event {
+                aws_sdk_s3::types::SelectObjectContentEventStream::Records(records_event) => {
+                    if let Some(payload) = records_event.payload {
+                        total_data.extend_from_slice(payload.as_ref());
+                    }
                }
-            }
-            aws_sdk_s3::types::SelectObjectContentEventStream::End(_) => {
-                break;
-            }
-            _ => {
-                // Handle other event types (Stats, Progress, Cont, etc.)
+                aws_sdk_s3::types::SelectObjectContentEventStream::End(_) => {
+                    saw_end = true;
+                    break;
+                }
+                _ => {}
            }
        }
-    }

-    Ok(String::from_utf8(total_data)?)
+        if !saw_end {
+            return Err("Select response ended without an End event".into());
+        }
+        Ok(String::from_utf8(total_data)?)
+    })
+    .await
+    .map_err(|_| -> Box<dyn Error + Send + Sync> { "Select response timed out".into() })?
 }

 #[tokio::test(flavor = "multi_thread", worker_threads = 4)]
-#[ignore = "requires running RustFS server at localhost:9000"]
-async fn test_select_object_content_csv_basic() -> Result<(), Box<dyn Error>> {
-    let client = create_aws_s3_client().await?;
+async fn test_select_object_content_csv_basic() -> TestResult<()> {
+    let (_env, client) = create_test_environment().await?;
    setup_test_bucket(&client).await?;
    upload_test_csv(&client).await?;

@@ -158,9 +145,8 @@ async fn test_select_object_content_csv_basic() -> Result<(), Box<dyn Error>> {
 }

 #[tokio::test(flavor = "multi_thread", worker_threads = 4)]
-#[ignore = "requires running RustFS server at localhost:9000"]
-async fn test_select_object_content_csv_aggregation() -> Result<(), Box<dyn Error>> {
-    let client = create_aws_s3_client().await?;
+async fn test_select_object_content_csv_aggregation() -> TestResult<()> {
+    let (_env, client) = create_test_environment().await?;
    setup_test_bucket(&client).await?;
    upload_test_csv(&client).await?;

@@ -203,16 +189,15 @@ async fn test_select_object_content_csv_aggregation() -> Result<(), Box<dyn Erro
 }

 #[tokio::test(flavor = "multi_thread", worker_threads = 4)]
-#[ignore = "requires running RustFS server at localhost:9000"]
-async fn test_select_object_content_json_basic() -> Result<(), Box<dyn Error>> {
-    let client = create_aws_s3_client().await?;
+async fn test_select_object_content_json_basic() -> TestResult<()> {
+    let (_env, client) = create_test_environment().await?;
    setup_test_bucket(&client).await?;
    upload_test_json(&client).await?;

    // Construct JSON query
    let sql = "SELECT s.name, s.age FROM S3Object s WHERE s.age > 28";

-    let json_input = JsonInput::builder().set_type(Some(JsonType::Document)).build();
+    let json_input = JsonInput::builder().set_type(Some(JsonType::Lines)).build();

    let input_serialization = InputSerialization::builder().json(json_input).build();

@@ -244,9 +229,8 @@ async fn test_select_object_content_json_basic() -> Result<(), Box<dyn Error>> {
 }

 #[tokio::test(flavor = "multi_thread", worker_threads = 4)]
-#[ignore = "requires running RustFS server at localhost:9000"]
-async fn test_select_object_content_csv_limit() -> Result<(), Box<dyn Error>> {
-    let client = create_aws_s3_client().await?;
+async fn test_select_object_content_csv_limit() -> TestResult<()> {
+    let (_env, client) = create_test_environment().await?;
    setup_test_bucket(&client).await?;
    upload_test_csv(&client).await?;

@@ -286,9 +270,8 @@ async fn test_select_object_content_csv_limit() -> Result<(), Box<dyn Error>> {
 }

 #[tokio::test(flavor = "multi_thread", worker_threads = 4)]
-#[ignore = "requires running RustFS server at localhost:9000"]
-async fn test_select_object_content_csv_order_by() -> Result<(), Box<dyn Error>> {
-    let client = create_aws_s3_client().await?;
+async fn test_select_object_content_csv_order_by() -> TestResult<()> {
+    let (_env, client) = create_test_environment().await?;
    setup_test_bucket(&client).await?;
    upload_test_csv(&client).await?;

@@ -318,9 +301,10 @@ async fn test_select_object_content_csv_order_by() -> Result<(), Box<dyn Error>>
    println!("CSV Order By result: {result_str}");

    // Verify ordered by age descending
-    assert!(
-        result_str.lines().filter(|line| !line.trim().is_empty()).count() >= 2,
-        "Should return at least 2 records"
+    assert_eq!(
+        result_str.lines().filter(|line| !line.trim().is_empty()).count(),
+        2,
+        "Should return exactly 2 records"
    );

    // Check if contains highest age records
@@ -331,9 +315,8 @@ async fn test_select_object_content_csv_order_by() -> Result<(), Box<dyn Error>>
 }

 #[tokio::test(flavor = "multi_thread", worker_threads = 4)]
-#[ignore = "requires running RustFS server at localhost:9000"]
-async fn test_select_object_content_error_handling() -> Result<(), Box<dyn Error>> {
-    let client = create_aws_s3_client().await?;
+async fn test_select_object_content_error_handling() -> TestResult<()> {
+    let (_env, client) = create_test_environment().await?;
    setup_test_bucket(&client).await?;
    upload_test_csv(&client).await?;

@@ -348,7 +331,7 @@ async fn test_select_object_content_error_handling() -> Result<(), Box<dyn Error
    let output_serialization = OutputSerialization::builder().csv(csv_output).build();

    // This query should fail because invalid_column doesn't exist
-    let result = client
+    let error = client
        .select_object_content()
        .bucket(BUCKET)
        .key(CSV_OBJECT)
@@ -357,18 +340,20 @@ async fn test_select_object_content_error_handling() -> Result<(), Box<dyn Error
        .input_serialization(input_serialization)
        .output_serialization(output_serialization)
        .send()
-        .await;
+        .await
+        .expect_err("a query referencing an unknown column must fail");

-    // Verify query fails (expected behavior)
-    assert!(result.is_err(), "Query with invalid column should fail");
+    assert_eq!(
+        error.as_service_error().and_then(ProvideErrorMetadata::code),
+        Some("EvaluatorBindingDoesNotExist")
+    );

    Ok(())
 }

 #[tokio::test(flavor = "multi_thread", worker_threads = 4)]
-#[ignore = "requires running RustFS server at localhost:9000"]
-async fn test_select_object_content_nonexistent_object() -> Result<(), Box<dyn Error>> {
-    let client = create_aws_s3_client().await?;
+async fn test_select_object_content_nonexistent_object() -> TestResult<()> {
+    let (_env, client) = create_test_environment().await?;
    setup_test_bucket(&client).await?;

    // Test query on nonexistent object
@@ -381,7 +366,7 @@ async fn test_select_object_content_nonexistent_object() -> Result<(), Box<dyn E
    let csv_output = CsvOutput::builder().build();
    let output_serialization = OutputSerialization::builder().csv(csv_output).build();

-    let result = client
+    let error = client
        .select_object_content()
        .bucket(BUCKET)
        .key("nonexistent.csv")
@@ -390,10 +375,10 @@ async fn test_select_object_content_nonexistent_object() -> Result<(), Box<dyn E
        .input_serialization(input_serialization)
        .output_serialization(output_serialization)
        .send()
-        .await;
+        .await
+        .expect_err("selecting a missing object must fail");

-    // Verify query fails (expected behavior)
-    assert!(result.is_err(), "Query on nonexistent object should fail");
+    assert_eq!(error.as_service_error().and_then(ProvideErrorMetadata::code), Some("NoSuchKey"));

    Ok(())
 }
@@ -23,12 +23,12 @@
 //!
 //! There are no containers, no external S3 backend and no `awscurl`: the
 //! `AddTier` admin call is signed in-process with `rustfs_signer`, exactly like
-//! the other admin-API e2e suites in this crate. The RustFS warm backend has no
-//! loopback/SSRF restriction (that guard is replication-only), so `hot` can tier
-//! to `cold` over `http://127.0.0.1:<port>`.
+//! the other admin-API e2e suites in this crate. The source server uses the
+//! explicit test-only loopback opt-in to tier to `cold` over
+//! `http://127.0.0.1:<port>` while production keeps the SSRF guard enabled.
 //!
-//! A single test drives the full transition main path and pins the chain
-//! required by ilm-7:
+//! The hermetic tests drive the transition and restore paths and pin the
+//! chains required by ilm-7 and the restore follow-up:
 //!   1. `AddTier(RustFS)` on `hot` targeting `cold` — the real connectivity /
 //!      in-use probe runs (no `force`), so this also proves the tier is reachable.
 //!   2. A `Transition Days=0` rule installed before a multipart PUT transitions
@@ -42,20 +42,20 @@
 //!   6. The remote object is present in the cold-tier bucket after transition.
 //!   7. `DeleteObject` on `hot` drives free-version cleanup: the cold-tier copy
 //!      eventually disappears and the hot object is gone (no local residue).
+//!   8. `RestoreObject` copy-back failures clear the in-progress marker, a
+//!      retry serves the object locally until expiry, and expiry leaves the
+//!      remote object available for a second restore.

-use crate::common::{RustFSTestEnvironment, local_http_client};
+use crate::common::RustFSTestEnvironment;
 use aws_sdk_s3::Client;
 use aws_sdk_s3::error::ProvideErrorMetadata;
 use aws_sdk_s3::primitives::ByteStream;
 use aws_sdk_s3::types::{
    BucketLifecycleConfiguration, BucketVersioningStatus, CompletedMultipartUpload, CompletedPart, ExpirationStatus,
-    LifecycleRule, LifecycleRuleFilter, NoncurrentVersionTransition, Transition, TransitionStorageClass, VersioningConfiguration,
+    LifecycleRule, LifecycleRuleFilter, NoncurrentVersionTransition, RestoreRequest, Transition, TransitionStorageClass,
+    VersioningConfiguration,
 };
 use http::Method;
-use http::header::HOST;
-use rustfs_signer::constants::UNSIGNED_PAYLOAD;
-use rustfs_signer::sign_v4;
-use s3s::Body;
 use serde::Deserialize;
 use std::time::{Duration as StdDuration, Instant};
 use time::{OffsetDateTime, format_description::well_known::Rfc3339};
@@ -96,6 +96,7 @@ const MANUAL_ACTIVE_CANCEL_OBJECTS: usize = 512;
 const MANUAL_RESTART_CANCEL_OBJECTS: usize = 512;
 const MANUAL_ACTIVE_CANCEL_RUNNING_TIMEOUT: StdDuration = StdDuration::from_secs(15);
 const MANUAL_TRANSITION_CANCEL_BARRIER_ENV: &str = "RUSTFS_E2E_MANUAL_TRANSITION_CANCEL_BARRIER";
+const ALLOW_LOOPBACK_TIER_ENDPOINT_ENV: (&str, &str) = ("RUSTFS_TIER_RUSTFS_ALLOW_LOOPBACK_ENDPOINT", "true");
 const MANUAL_ASYNC_CONFLICT_TERMINAL_TIMEOUT: StdDuration = StdDuration::from_secs(90);
 const MANUAL_RESTART_RECOVERY_TIMEOUT: StdDuration = StdDuration::from_secs(80);
 const OBJECT_KEY: &str = "tier/鲁A12345/report.bin";
@@ -112,6 +113,20 @@ const HDR_SOURCE_REPLICATION_REQUEST: &str = "x-rustfs-source-replication-reques
 const HDR_SOURCE_MTIME: &str = "x-rustfs-source-mtime";
 const TIER_MUTATION_RECOVERY_CHANGED: &str = "Remote tier mutation recovery changed before publish";

+async fn start_tier_source(hot: &mut RustFSTestEnvironment, extra_env: &[(&str, &str)]) -> TestResult {
+    let mut env = Vec::with_capacity(extra_env.len() + 1);
+    env.push(ALLOW_LOOPBACK_TIER_ENDPOINT_ENV);
+    env.extend_from_slice(extra_env);
+    hot.start_rustfs_server_with_env(vec![], &env).await
+}
+
+async fn restart_tier_source(hot: &mut RustFSTestEnvironment, extra_env: &[(&str, &str)]) -> TestResult {
+    let mut env = Vec::with_capacity(extra_env.len() + 1);
+    env.push(ALLOW_LOOPBACK_TIER_ENDPOINT_ENV);
+    env.extend_from_slice(extra_env);
+    hot.restart_server_preserving_data(vec![], &env).await
+}
+
 /// 5 MiB — the S3 minimum size for a non-final multipart part; the object's only
 /// internal part boundary sits at this offset.
 const PART0_SIZE: usize = 5 * 1024 * 1024;
@@ -127,9 +142,8 @@ fn payload() -> Vec<u8> {

 /// Sign and send an admin request in-process (no `awscurl`).
 ///
-/// Mirrors the shared admin-API e2e pattern: the SigV4 signature is computed
-/// over `UNSIGNED_PAYLOAD`, so the JSON body rides on the wire without being
-/// pre-hashed. Returns the response status and body text.
+/// Thin wrapper over [`crate::common::admin_request`], kept local so the call
+/// sites below keep their `Option<&str>` body shape.
 async fn signed_admin_request(
    base_url: &str,
    method: Method,
@@ -138,30 +152,7 @@ async fn signed_admin_request(
    access_key: &str,
    secret_key: &str,
 ) -> Result<(reqwest::StatusCode, String), Box<dyn std::error::Error + Send + Sync>> {
-    let url = format!("{base_url}{path}");
-    let uri = url.parse::<http::Uri>()?;
-    let authority = uri.authority().ok_or("request URL missing authority")?.to_string();
-    let body_bytes = body.map(|b| b.as_bytes().to_vec()).unwrap_or_default();
-
-    let request = http::Request::builder()
-        .method(method.clone())
-        .uri(uri)
-        .header(HOST, authority)
-        .header("x-amz-content-sha256", UNSIGNED_PAYLOAD);
-    let signed = sign_v4(request.body(Body::empty())?, 0, access_key, secret_key, "", "us-east-1");
-
-    let client = local_http_client();
-    let mut request_builder = client.request(method, url.as_str());
-    for (name, value) in signed.headers() {
-        request_builder = request_builder.header(name, value);
-    }
-    if !body_bytes.is_empty() {
-        request_builder = request_builder.body(body_bytes);
-    }
-    let response = request_builder.send().await?;
-    let status = response.status();
-    let text = response.text().await?;
-    Ok((status, text))
+    crate::common::admin_request(base_url, method, path, body.map(str::to_string), access_key, secret_key).await
 }

 /// Wire `hot` -> `cold` as a `TierType::RustFS` remote tier via `AddTier`.
@@ -739,6 +730,77 @@ async fn wait_for_transition(client: &Client, bucket: &str, key: &str, deadline:
    }
 }

+/// Poll `HEAD` until the asynchronous copy-back reports a completed restore.
+async fn wait_for_restore_complete(client: &Client, bucket: &str, key: &str, deadline: StdDuration) -> TestResult {
+    let start = Instant::now();
+    loop {
+        let head = client.head_object().bucket(bucket).key(key).send().await?;
+        if head
+            .restore()
+            .is_some_and(|restore| restore.contains("ongoing-request=\"false\""))
+        {
+            return Ok(());
+        }
+        if start.elapsed() >= deadline {
+            return Err(format!(
+                "restore for {bucket}/{key} did not complete within {}s; restore={:?}",
+                deadline.as_secs(),
+                head.restore()
+            )
+            .into());
+        }
+        tokio::time::sleep(StdDuration::from_millis(250)).await;
+    }
+}
+
+/// Poll `HEAD` until the lifecycle restore-expiry action removes restore metadata.
+async fn wait_for_restore_clear(client: &Client, bucket: &str, key: &str, deadline: StdDuration) -> TestResult {
+    let start = Instant::now();
+    loop {
+        let head = client.head_object().bucket(bucket).key(key).send().await?;
+        if head.restore().is_none() {
+            return Ok(());
+        }
+        if start.elapsed() >= deadline {
+            return Err(format!(
+                "restore metadata for {bucket}/{key} was not cleared within {}s; restore={:?}",
+                deadline.as_secs(),
+                head.restore()
+            )
+            .into());
+        }
+        tokio::time::sleep(StdDuration::from_millis(250)).await;
+    }
+}
+
+/// Poll `HEAD` through the failed-copy transition, proving that the request
+/// first published an in-progress marker and that the failure then removed it.
+async fn wait_for_restore_failure(client: &Client, bucket: &str, key: &str, deadline: StdDuration) -> TestResult {
+    let start = Instant::now();
+    let mut saw_ongoing = false;
+    loop {
+        let head = client.head_object().bucket(bucket).key(key).send().await?;
+        if head
+            .restore()
+            .is_some_and(|restore| restore.contains("ongoing-request=\"true\""))
+        {
+            saw_ongoing = true;
+        } else if saw_ongoing && head.restore().is_none() {
+            return Ok(());
+        }
+        if start.elapsed() >= deadline {
+            return Err(format!(
+                "failed restore for {bucket}/{key} did not publish and clear its in-progress marker within {}s; \
+                 saw_ongoing={saw_ongoing}, restore={:?}",
+                deadline.as_secs(),
+                head.restore()
+            )
+            .into());
+        }
+        tokio::time::sleep(StdDuration::from_millis(100)).await;
+    }
+}
+
 /// Poll until the cold-tier bucket is empty (remote free-version cleanup done),
 /// or fail after `deadline`.
 async fn wait_for_cold_tier_empty(cold_client: &Client, deadline: StdDuration) -> TestResult {
@@ -813,8 +875,7 @@ async fn test_hermetic_transition_main_path() -> TestResult {
    // Hot/source server. A 1s scanner cycle is a backstop; transition is
    // primarily driven immediately by the multipart completion path.
    let mut hot = RustFSTestEnvironment::new().await?;
-    hot.start_rustfs_server_with_env(vec![], &[("RUSTFS_SCANNER_CYCLE", "1")])
-        .await?;
+    start_tier_source(&mut hot, &[("RUSTFS_SCANNER_CYCLE", "1")]).await?;
    let hot_client = hot.create_s3_client();

    // Wire the RustFS remote tier (real connectivity probe, no force).
@@ -894,6 +955,142 @@ async fn test_hermetic_transition_main_path() -> TestResult {
    Ok(())
 }

+/// Restore a transitioned object through a real RustFS remote tier.
+///
+/// The test covers the externally visible copy-back contract that a mock tier
+/// cannot prove: a failed remote read clears `x-amz-restore`, a retry creates a
+/// local copy that remains readable while the cold tier is unavailable, expiry
+/// removes only that local copy, and the same remote object can be restored a
+/// second time. The accelerated lifecycle clock keeps the expiry assertion
+/// bounded while the scanner remains enabled.
+#[tokio::test(flavor = "multi_thread", worker_threads = 4)]
+async fn test_hermetic_transition_restore_failure_expiry_and_retry() -> TestResult {
+    let mut cold = RustFSTestEnvironment::new().await?;
+    cold.access_key = "restorecoldtieradmin".to_string();
+    cold.secret_key = "restorecoldtiersecret".to_string();
+    cold.start_rustfs_server_without_cleanup(vec![]).await?;
+    let cold_client = cold.create_s3_client();
+    cold_client.create_bucket().bucket(TIER_BUCKET).send().await?;
+
+    let mut hot = RustFSTestEnvironment::new().await?;
+    start_tier_source(&mut hot, &[("RUSTFS_SCANNER_CYCLE", "1"), ("RUSTFS_ILM_DEBUG_DAY_SECS", "5")]).await?;
+    let hot_client = hot.create_s3_client();
+    add_rustfs_tier(&hot, &cold).await?;
+
+    hot_client.create_bucket().bucket(SOURCE_BUCKET).send().await?;
+    hot_client
+        .put_bucket_lifecycle_configuration()
+        .bucket(SOURCE_BUCKET)
+        .lifecycle_configuration(BucketLifecycleConfiguration::builder().rules(transition_rule()?).build()?)
+        .send()
+        .await?;
+
+    let data = payload();
+    put_multipart_object(&hot_client, SOURCE_BUCKET, OBJECT_KEY, &data).await?;
+    wait_for_transition(&hot_client, SOURCE_BUCKET, OBJECT_KEY, StdDuration::from_secs(90)).await?;
+    let transitioned = hot_client.head_object().bucket(SOURCE_BUCKET).key(OBJECT_KEY).send().await?;
+    assert_eq!(
+        transitioned.storage_class().map(|storage_class| storage_class.as_str()),
+        Some(TIER_NAME),
+        "restore fixture must be transitioned before the copy-back request"
+    );
+    assert!(transitioned.restore().is_none(), "transitioned object must not already be restored");
+    assert_eq!(
+        cold_tier_object_count(&cold_client).await?,
+        1,
+        "cold tier should contain one restore candidate"
+    );
+
+    // A failed remote read is accepted asynchronously, but it must not leave
+    // an object permanently advertising an in-progress restore.
+    cold.stop_server();
+    hot_client
+        .restore_object()
+        .bucket(SOURCE_BUCKET)
+        .key(OBJECT_KEY)
+        .restore_request(RestoreRequest::builder().days(1).build())
+        .send()
+        .await?;
+    wait_for_restore_failure(&hot_client, SOURCE_BUCKET, OBJECT_KEY, StdDuration::from_secs(30)).await?;
+
+    // Once the tier is available again, the same object can be restored.
+    cold.restart_server_preserving_data(vec![], &[]).await?;
+    hot_client
+        .restore_object()
+        .bucket(SOURCE_BUCKET)
+        .key(OBJECT_KEY)
+        .restore_request(RestoreRequest::builder().days(1).build())
+        .send()
+        .await?;
+    wait_for_restore_complete(&hot_client, SOURCE_BUCKET, OBJECT_KEY, StdDuration::from_secs(30)).await?;
+
+    let restored = hot_client.head_object().bucket(SOURCE_BUCKET).key(OBJECT_KEY).send().await?;
+    assert_eq!(
+        restored.storage_class().map(|storage_class| storage_class.as_str()),
+        Some(TIER_NAME),
+        "restore must retain the transitioned storage class"
+    );
+    assert!(
+        restored
+            .restore()
+            .is_some_and(|restore| restore.contains("ongoing-request=\"false\"")),
+        "completed restore must advertise a finished temporary copy"
+    );
+
+    // A completed restore must be served locally even if the remote tier is
+    // temporarily unavailable.
+    cold.stop_server();
+    let local_get = hot_client.get_object().bucket(SOURCE_BUCKET).key(OBJECT_KEY).send().await?;
+    let local_body = local_get.body.collect().await?.into_bytes();
+    assert_eq!(local_body.as_ref(), data.as_slice(), "restored local copy must be byte-identical");
+
+    cold.restart_server_preserving_data(vec![], &[]).await?;
+
+    // The test clock makes the one-day restore expire in roughly five seconds.
+    wait_for_restore_clear(&hot_client, SOURCE_BUCKET, OBJECT_KEY, StdDuration::from_secs(30)).await?;
+    let expired = hot_client.head_object().bucket(SOURCE_BUCKET).key(OBJECT_KEY).send().await?;
+    assert_eq!(
+        expired.storage_class().map(|storage_class| storage_class.as_str()),
+        Some(TIER_NAME),
+        "restore expiry must not clear the transitioned storage class"
+    );
+    assert_eq!(
+        cold_tier_object_count(&cold_client).await?,
+        1,
+        "restore expiry must retain the remote object"
+    );
+
+    // After expiry the local copy is gone, so an unavailable tier must make the
+    // read fail; with the tier back, the original bytes remain readable.
+    cold.stop_server();
+    let expired_get = hot_client.get_object().bucket(SOURCE_BUCKET).key(OBJECT_KEY).send().await;
+    assert!(expired_get.is_err(), "expired restore must not leave a local copy behind");
+    cold.restart_server_preserving_data(vec![], &[]).await?;
+    let remote_get = hot_client.get_object().bucket(SOURCE_BUCKET).key(OBJECT_KEY).send().await?;
+    let remote_body = remote_get.body.collect().await?.into_bytes();
+    assert_eq!(remote_body.as_ref(), data.as_slice(), "post-expiry GET must read the remote copy");
+
+    // The remote candidate survives expiry and can be restored again.
+    hot_client
+        .restore_object()
+        .bucket(SOURCE_BUCKET)
+        .key(OBJECT_KEY)
+        .restore_request(RestoreRequest::builder().days(1).build())
+        .send()
+        .await?;
+    wait_for_restore_complete(&hot_client, SOURCE_BUCKET, OBJECT_KEY, StdDuration::from_secs(30)).await?;
+
+    hot_client
+        .delete_object()
+        .bucket(SOURCE_BUCKET)
+        .key(OBJECT_KEY)
+        .send()
+        .await?;
+    wait_for_cold_tier_empty(&cold_client, StdDuration::from_secs(90)).await?;
+
+    Ok(())
+}
+
 #[tokio::test(flavor = "multi_thread", worker_threads = 4)]
 async fn test_manual_transition_run_black_box_semantics() -> TestResult {
    let mut cold = RustFSTestEnvironment::new().await?;
@@ -904,8 +1101,7 @@ async fn test_manual_transition_run_black_box_semantics() -> TestResult {
    cold_client.create_bucket().bucket(TIER_BUCKET).send().await?;

    let mut hot = RustFSTestEnvironment::new().await?;
-    hot.start_rustfs_server_with_env(vec![], &[("RUSTFS_SCANNER_ENABLED", "false"), ("RUSTFS_SCANNER_CYCLE", "3600")])
-        .await?;
+    start_tier_source(&mut hot, &[("RUSTFS_SCANNER_ENABLED", "false"), ("RUSTFS_SCANNER_CYCLE", "3600")]).await?;
    let hot_client = hot.create_s3_client();
    add_rustfs_tier(&hot, &cold).await?;
    let due_mtime = OffsetDateTime::now_utc() - time::Duration::hours(25);
@@ -1010,8 +1206,7 @@ async fn test_manual_transition_async_job_status_polling() -> TestResult {
    cold_client.create_bucket().bucket(TIER_BUCKET).send().await?;

    let mut hot = RustFSTestEnvironment::new().await?;
-    hot.start_rustfs_server_with_env(vec![], &[("RUSTFS_SCANNER_ENABLED", "false"), ("RUSTFS_SCANNER_CYCLE", "3600")])
-        .await?;
+    start_tier_source(&mut hot, &[("RUSTFS_SCANNER_ENABLED", "false"), ("RUSTFS_SCANNER_CYCLE", "3600")]).await?;
    let hot_client = hot.create_s3_client();
    add_rustfs_tier(&hot, &cold).await?;

@@ -1109,8 +1304,7 @@ async fn test_manual_transition_async_limit_reports_terminal_partial() -> TestRe
    cold_client.create_bucket().bucket(TIER_BUCKET).send().await?;

    let mut hot = RustFSTestEnvironment::new().await?;
-    hot.start_rustfs_server_with_env(vec![], &[("RUSTFS_SCANNER_ENABLED", "false"), ("RUSTFS_SCANNER_CYCLE", "3600")])
-        .await?;
+    start_tier_source(&mut hot, &[("RUSTFS_SCANNER_ENABLED", "false"), ("RUSTFS_SCANNER_CYCLE", "3600")]).await?;
    let hot_client = hot.create_s3_client();
    add_rustfs_tier(&hot, &cold).await?;

@@ -1271,8 +1465,8 @@ async fn test_manual_transition_async_scope_conflicts_report_active_job() -> Tes
    cold_client.create_bucket().bucket(TIER_BUCKET).send().await?;

    let mut hot = RustFSTestEnvironment::new().await?;
-    hot.start_rustfs_server_with_env(
-        vec![],
+    start_tier_source(
+        &mut hot,
        &[
            ("RUSTFS_SCANNER_ENABLED", "false"),
            ("RUSTFS_SCANNER_CYCLE", "3600"),
@@ -1381,8 +1575,7 @@ async fn test_manual_transition_async_different_buckets_admit_concurrently() ->
    cold_client.create_bucket().bucket(TIER_BUCKET).send().await?;

    let mut hot = RustFSTestEnvironment::new().await?;
-    hot.start_rustfs_server_with_env(vec![], &[("RUSTFS_SCANNER_ENABLED", "false"), ("RUSTFS_SCANNER_CYCLE", "3600")])
-        .await?;
+    start_tier_source(&mut hot, &[("RUSTFS_SCANNER_ENABLED", "false"), ("RUSTFS_SCANNER_CYCLE", "3600")]).await?;
    let hot_client = hot.create_s3_client();
    add_rustfs_tier(&hot, &cold).await?;

@@ -1502,8 +1695,7 @@ async fn test_manual_transition_async_tier_failure_reports_terminal_partial() ->
    cold_client.create_bucket().bucket(TIER_BUCKET).send().await?;

    let mut hot = RustFSTestEnvironment::new().await?;
-    hot.start_rustfs_server_with_env(vec![], &[("RUSTFS_SCANNER_ENABLED", "false"), ("RUSTFS_SCANNER_CYCLE", "3600")])
-        .await?;
+    start_tier_source(&mut hot, &[("RUSTFS_SCANNER_ENABLED", "false"), ("RUSTFS_SCANNER_CYCLE", "3600")]).await?;
    let hot_client = hot.create_s3_client();
    add_rustfs_tier(&hot, &cold).await?;

@@ -1595,8 +1787,7 @@ async fn test_manual_transition_async_worker_failure_reports_terminal_partial()
    cold_client.create_bucket().bucket(TIER_BUCKET).send().await?;

    let mut hot = RustFSTestEnvironment::new().await?;
-    hot.start_rustfs_server_with_env(vec![], &[("RUSTFS_SCANNER_ENABLED", "false"), ("RUSTFS_SCANNER_CYCLE", "3600")])
-        .await?;
+    start_tier_source(&mut hot, &[("RUSTFS_SCANNER_ENABLED", "false"), ("RUSTFS_SCANNER_CYCLE", "3600")]).await?;
    let hot_client = hot.create_s3_client();
    add_rustfs_tier(&hot, &cold).await?;
    cold.stop_server();
@@ -1688,8 +1879,8 @@ async fn test_manual_transition_async_active_cancel_reports_terminal_cancelled()
    cold_client.create_bucket().bucket(TIER_BUCKET).send().await?;

    let mut hot = RustFSTestEnvironment::new().await?;
-    hot.start_rustfs_server_with_env(
-        vec![],
+    start_tier_source(
+        &mut hot,
        &[
            ("RUSTFS_SCANNER_ENABLED", "false"),
            ("RUSTFS_SCANNER_CYCLE", "3600"),
@@ -1794,7 +1985,7 @@ async fn test_manual_transition_async_cancel_after_process_restart_recovers_term
        ("RUSTFS_TRANSITION_QUEUE_CAPACITY", "512"),
    ];
    let mut hot = RustFSTestEnvironment::new().await?;
-    hot.start_rustfs_server_with_env(vec![], &restart_env).await?;
+    start_tier_source(&mut hot, &restart_env).await?;
    let hot_client = hot.create_s3_client();
    add_rustfs_tier(&hot, &cold).await?;

@@ -1822,7 +2013,7 @@ async fn test_manual_transition_async_cancel_after_process_restart_recovers_term
        .ok_or("async response must include status_endpoint")?;
    assert_eq!(accepted.cancel_endpoint.as_deref(), Some(status_endpoint));

-    hot.restart_server_preserving_data(vec![], &restart_env).await?;
+    restart_tier_source(&mut hot, &restart_env).await?;

    let restarted = manual_transition_job_status(&hot, status_endpoint).await?;
    assert_eq!(restarted.job_id, job_id);
@@ -1946,8 +2137,7 @@ async fn test_manual_transition_run_contract_no_status_cancel_fields() -> TestRe
    cold_client.create_bucket().bucket(TIER_BUCKET).send().await?;

    let mut hot = RustFSTestEnvironment::new().await?;
-    hot.start_rustfs_server_with_env(vec![], &[("RUSTFS_SCANNER_ENABLED", "false"), ("RUSTFS_SCANNER_CYCLE", "3600")])
-        .await?;
+    start_tier_source(&mut hot, &[("RUSTFS_SCANNER_ENABLED", "false"), ("RUSTFS_SCANNER_CYCLE", "3600")]).await?;
    let hot_client = hot.create_s3_client();
    add_rustfs_tier(&hot, &cold).await?;

@@ -1985,8 +2175,7 @@ async fn test_manual_transition_run_continuation_token_resumes_without_raw_marke
    cold_client.create_bucket().bucket(TIER_BUCKET).send().await?;

    let mut hot = RustFSTestEnvironment::new().await?;
-    hot.start_rustfs_server_with_env(vec![], &[("RUSTFS_SCANNER_ENABLED", "false"), ("RUSTFS_SCANNER_CYCLE", "3600")])
-        .await?;
+    start_tier_source(&mut hot, &[("RUSTFS_SCANNER_ENABLED", "false"), ("RUSTFS_SCANNER_CYCLE", "3600")]).await?;
    let hot_client = hot.create_s3_client();
    add_rustfs_tier(&hot, &cold).await?;

@@ -2026,8 +2215,7 @@ async fn test_manual_transition_run_continuation_token_resumes_without_raw_marke
        "continuation token must not expose the raw object prefix: {continuation}"
    );

-    hot.restart_server_preserving_data(vec![], &[("RUSTFS_SCANNER_ENABLED", "false"), ("RUSTFS_SCANNER_CYCLE", "3600")])
-        .await?;
+    restart_tier_source(&mut hot, &[("RUSTFS_SCANNER_ENABLED", "false"), ("RUSTFS_SCANNER_CYCLE", "3600")]).await?;

    let second = manual_transition_run_with_max_and_continuation(
        &hot,
@@ -2060,8 +2248,8 @@ async fn test_manual_transition_run_queue_pressure_partial() -> TestResult {
    cold_client.create_bucket().bucket(TIER_BUCKET).send().await?;

    let mut hot = RustFSTestEnvironment::new().await?;
-    hot.start_rustfs_server_with_env(
-        vec![],
+    start_tier_source(
+        &mut hot,
        &[
            ("RUSTFS_SCANNER_ENABLED", "false"),
            ("RUSTFS_SCANNER_CYCLE", "3600"),
@@ -13,16 +13,20 @@
 // limitations under the License.

 use crate::common::{
-    RustFSTestEnvironment, admin_create_user, awscurl_post_sts_form_urlencoded, init_logging, local_http_client,
-    replication_fast_env, rustfs_binary_path, signed_request, signed_request_with_client, signed_request_with_session_token,
+    AdminTransport, RustFSTestEnvironment, admin_add_canned_policy_via, admin_attach_user_policy_via, admin_create_user,
+    awscurl_post_sts_form_urlencoded, init_logging, local_http_client, replication_fast_env, rustfs_binary_path, signed_request,
+    signed_request_with_client, signed_request_with_session_token,
 };
 use crate::fake_s3_target::{
    FAKE_ACCESS_KEY, FAKE_SECRET_KEY, FakeS3Target, FaultAction as FakeTargetFault, Operation as FakeTargetOperation,
    RequestRecord,
 };
-use crate::kms::common::{create_key_with_specific_id, sse_customer_key_md5_base64};
+use crate::kms::common::{
+    SSE_C_KEY_MISMATCH_MESSAGE, SSE_C_MISSING_PARAMETERS_MESSAGE, assert_s3_error, create_key_with_specific_id,
+    sse_customer_key_md5_base64,
+};
 use crate::storage_api::replication_extension::BucketTargetSys;
-use aws_sdk_s3::config::{Credentials, Region};
+use aws_sdk_s3::Client;
 use aws_sdk_s3::error::ProvideErrorMetadata;
 use aws_sdk_s3::operation::list_object_versions::ListObjectVersionsOutput;
 use aws_sdk_s3::primitives::ByteStream;
@@ -30,8 +34,7 @@ use aws_sdk_s3::types::{
    BucketVersioningStatus, CompletedMultipartUpload, CompletedPart, DeleteMarkerEntry, ObjectVersion, ServerSideEncryption,
    VersioningConfiguration,
 };
-use aws_sdk_s3::{Client, Config};
-use base64::{Engine, engine::general_purpose::STANDARD as BASE64_STANDARD};
+use base64_simd::STANDARD as BASE64_STANDARD;
 use bytes::Bytes;
 use flate2::read::GzDecoder;
 use futures::{Stream, StreamExt};
@@ -892,15 +895,7 @@ async fn wait_for_replicated_object_over_https(
 }

 fn create_user_s3_client(env: &RustFSTestEnvironment, access_key: &str, secret_key: &str) -> Client {
-    let credentials = Credentials::new(access_key, secret_key, None, None, "e2e-site-replication");
-    let config = Config::builder()
-        .credentials_provider(credentials)
-        .region(Region::new("us-east-1"))
-        .endpoint_url(&env.url)
-        .force_path_style(true)
-        .behavior_version_latest()
-        .build();
-    Client::from_conf(config)
+    env.create_s3_client_with_credentials(access_key, secret_key)
 }

 async fn admin_add_canned_policy(
@@ -908,24 +903,15 @@ async fn admin_add_canned_policy(
    policy_name: &str,
    policy: &serde_json::Value,
 ) -> Result<(), Box<dyn Error + Send + Sync>> {
-    let url = format!("{}/rustfs/admin/v3/add-canned-policy?name={}", env.url, policy_name);
-    let response = signed_request(
-        http::Method::PUT,
-        &url,
+    admin_add_canned_policy_via(
+        AdminTransport::Signed,
+        &env.url,
        &env.access_key,
        &env.secret_key,
-        Some(policy.to_string().into_bytes()),
-        Some("application/json"),
+        policy_name,
+        &policy.to_string(),
    )
-    .await?;
-
-    if response.status() != StatusCode::OK {
-        let status = response.status();
-        let body = response.text().await.unwrap_or_default();
-        return Err(format!("add canned policy failed: {status} {body}").into());
-    }
-
-    Ok(())
+    .await
 }

 async fn admin_attach_policy_to_user(
@@ -933,19 +919,7 @@ async fn admin_attach_policy_to_user(
    policy_name: &str,
    username: &str,
 ) -> Result<(), Box<dyn Error + Send + Sync>> {
-    let url = format!(
-        "{}/rustfs/admin/v3/set-user-or-group-policy?policyName={}&userOrGroup={}&isGroup=false",
-        env.url, policy_name, username
-    );
-    let response = signed_request(http::Method::PUT, &url, &env.access_key, &env.secret_key, Some(Vec::new()), None).await?;
-
-    if response.status() != StatusCode::OK {
-        let status = response.status();
-        let body = response.text().await.unwrap_or_default();
-        return Err(format!("attach policy to user failed: {status} {body}").into());
-    }
-
-    Ok(())
+    admin_attach_user_policy_via(AdminTransport::Signed, &env.url, &env.access_key, &env.secret_key, policy_name, username).await
 }

 async fn admin_update_group_members(
@@ -1241,7 +1215,7 @@ async fn wait_for_source_replication_pending_or_failed(
 }

 async fn wait_for_source_replication_status(client: &Client, bucket: &str, key: &str, expected: &str, ssec: bool) -> TestResult {
-    let customer_key = BASE64_STANDARD.encode(REPL17_SSEC_KEY);
+    let customer_key = BASE64_STANDARD.encode_to_string(REPL17_SSEC_KEY);
    let customer_key_md5 = sse_customer_key_md5_base64(REPL17_SSEC_KEY);
    let wait = async {
        loop {
@@ -1336,7 +1310,7 @@ async fn assert_failed_replication_stays_absent_for(
    ssec: bool,
    duration: Duration,
 ) -> TestResult {
-    let customer_key = BASE64_STANDARD.encode(REPL17_SSEC_KEY);
+    let customer_key = BASE64_STANDARD.encode_to_string(REPL17_SSEC_KEY);
    let customer_key_md5 = sse_customer_key_md5_base64(REPL17_SSEC_KEY);
    let wait = async {
        let deadline = tokio::time::Instant::now() + duration;
@@ -1938,6 +1912,21 @@ async fn site_replication_info(env: &RustFSTestEnvironment) -> Result<SiteReplic
    Ok(serde_json::from_slice(&response.bytes().await?)?)
 }

+async fn site_replication_rotate_svc_acct(
+    env: &RustFSTestEnvironment,
+) -> Result<ReplicateEditStatus, Box<dyn Error + Send + Sync>> {
+    let url = format!("{}/rustfs/admin/v3/site-replication/rotate-svc-acct", env.url);
+    let response = signed_request(http::Method::POST, &url, &env.access_key, &env.secret_key, None, None).await?;
+
+    if response.status() != StatusCode::OK {
+        let status = response.status();
+        let body = response.text().await.unwrap_or_default();
+        return Err(format!("site replication rotate-svc-acct failed: {status} {body}").into());
+    }
+
+    Ok(serde_json::from_slice(&response.bytes().await?)?)
+}
+
 async fn site_replication_resync_op(
    env: &RustFSTestEnvironment,
    operation: &str,
@@ -4329,7 +4318,7 @@ async fn test_bucket_replication_sse_c_contract() -> TestResult {
    let target_client = target_env.create_s3_client();
    let key = "ssec-contract.txt";
    let body = b"repl-17 SSE-C payload";
-    let customer_key = BASE64_STANDARD.encode(REPL17_SSEC_KEY);
+    let customer_key = BASE64_STANDARD.encode_to_string(REPL17_SSEC_KEY);
    let customer_key_md5 = sse_customer_key_md5_base64(REPL17_SSEC_KEY);

    source_client
@@ -4375,10 +4364,16 @@ async fn test_bucket_replication_sse_c_contract() -> TestResult {
    // Without the customer key the replica must not be readable — the direct
    // detection point for a silent-plaintext replica (backlog#1291).
    let plain_read = target_client.get_object().bucket(&target_bucket).key(key).send().await;
-    assert!(plain_read.is_err(), "SSE-C replica must not be readable without the customer key");
+    assert_s3_error(
+        plain_read,
+        400,
+        "InvalidRequest",
+        SSE_C_MISSING_PARAMETERS_MESSAGE,
+        "SSE-C replica must not be readable without the customer key",
+    );

    // A wrong customer key must fail too.
-    let wrong_key = BASE64_STANDARD.encode("99999999999999999999999999999999");
+    let wrong_key = BASE64_STANDARD.encode_to_string("99999999999999999999999999999999");
    let wrong_key_md5 = sse_customer_key_md5_base64("99999999999999999999999999999999");
    let wrong_read = target_client
        .get_object()
@@ -4389,7 +4384,13 @@ async fn test_bucket_replication_sse_c_contract() -> TestResult {
        .sse_customer_key_md5(&wrong_key_md5)
        .send()
        .await;
-    assert!(wrong_read.is_err(), "SSE-C replica must reject a wrong customer key");
+    assert_s3_error(
+        wrong_read,
+        400,
+        "InvalidRequest",
+        SSE_C_KEY_MISMATCH_MESSAGE,
+        "SSE-C replica must reject a wrong customer key",
+    );

    Ok(())
 }
@@ -4408,7 +4409,7 @@ async fn test_bucket_replication_sse_c_multipart_passthrough() -> TestResult {
    let source_client = source_env.create_s3_client();
    let target_client = target_env.create_s3_client();
    let key = "ssec-mp-contract.bin";
-    let customer_key = BASE64_STANDARD.encode(REPL17_SSEC_KEY);
+    let customer_key = BASE64_STANDARD.encode_to_string(REPL17_SSEC_KEY);
    let customer_key_md5 = sse_customer_key_md5_base64(REPL17_SSEC_KEY);

    let created = source_client
@@ -4486,9 +4487,12 @@ async fn test_bucket_replication_sse_c_multipart_passthrough() -> TestResult {
    assert_eq!(replica.body.collect().await?.into_bytes().as_ref(), payload.as_slice());

    let plain_read = target_client.get_object().bucket(&target_bucket).key(key).send().await;
-    assert!(
-        plain_read.is_err(),
-        "SSE-C multipart replica must not be readable without the customer key"
+    assert_s3_error(
+        plain_read,
+        400,
+        "InvalidRequest",
+        SSE_C_MISSING_PARAMETERS_MESSAGE,
+        "SSE-C multipart replica must not be readable without the customer key",
    );

    // Stability across scanner cycles: convergence must hold for passthrough.
@@ -4550,7 +4554,7 @@ async fn test_ssec_replication_fails_closed_when_target_drops_passthrough_header
    .await?;
    put_bucket_replication(&source_env, source_bucket, &target_arn).await?;

-    let customer_key = BASE64_STANDARD.encode(REPL17_SSEC_KEY);
+    let customer_key = BASE64_STANDARD.encode_to_string(REPL17_SSEC_KEY);
    let customer_key_md5 = sse_customer_key_md5_base64(REPL17_SSEC_KEY);
    let put_ssec = |key: &'static str| {
        source_client
@@ -4723,7 +4727,7 @@ async fn test_bucket_replication_sse_c_heals_after_target_outage() -> TestResult
    let source_client = source_env.create_s3_client();
    let key = "ssec-heal-contract.txt";
    let body = b"repl-22 ssec heal payload".to_vec();
-    let customer_key = BASE64_STANDARD.encode(REPL17_SSEC_KEY);
+    let customer_key = BASE64_STANDARD.encode_to_string(REPL17_SSEC_KEY);
    let customer_key_md5 = sse_customer_key_md5_base64(REPL17_SSEC_KEY);

    // Target outage: the SSE-C write cannot replicate.
@@ -4838,7 +4842,7 @@ async fn test_bucket_replication_sse_c_existing_object_resync() -> TestResult {
    // The SSE-C object exists before any replication wiring.
    let key = "ssec-existing-contract.txt";
    let body = b"repl-22 ssec existing-object payload".to_vec();
-    let customer_key = BASE64_STANDARD.encode(REPL17_SSEC_KEY);
+    let customer_key = BASE64_STANDARD.encode_to_string(REPL17_SSEC_KEY);
    let customer_key_md5 = sse_customer_key_md5_base64(REPL17_SSEC_KEY);
    source_client
        .put_object()
@@ -4892,15 +4896,12 @@ async fn test_bucket_replication_sse_c_existing_object_resync() -> TestResult {
    assert_eq!(replica.body.collect().await?.into_bytes().as_ref(), body.as_slice());

    // No plaintext leak: the replica stays unreadable without the key.
-    assert!(
-        target_client
-            .get_object()
-            .bucket(target_bucket)
-            .key(key)
-            .send()
-            .await
-            .is_err(),
-        "SSE-C replica must not be readable without the customer key"
+    assert_s3_error(
+        target_client.get_object().bucket(target_bucket).key(key).send().await,
+        400,
+        "InvalidRequest",
+        SSE_C_MISSING_PARAMETERS_MESSAGE,
+        "SSE-C resynced replica must not be readable without the customer key",
    );

    Ok(())
@@ -6324,6 +6325,112 @@ async fn test_site_replication_remove_all_real_dual_node() -> Result<(), Box<dyn
    Ok(())
 }

+#[tokio::test]
+async fn test_site_replication_rotate_svc_acct_completes_and_replication_survives_real_dual_node()
+-> Result<(), Box<dyn Error + Send + Sync>> {
+    init_logging();
+
+    let mut source_env = RustFSTestEnvironment::new().await?;
+    source_env
+        .start_rustfs_server_with_env(vec![], LOOPBACK_REPLICATION_TARGET_ENV)
+        .await?;
+
+    let mut target_env = RustFSTestEnvironment::new().await?;
+    target_env
+        .start_rustfs_server_without_cleanup_with_env(LOOPBACK_REPLICATION_TARGET_ENV)
+        .await?;
+
+    let source_client = source_env.create_s3_client();
+    let target_client = target_env.create_s3_client();
+    let bucket = "site-repl-rotate-svc-acct";
+
+    let add_status = site_replication_add(
+        &source_env,
+        &[
+            PeerSite {
+                name: "source-site".to_string(),
+                endpoint: source_env.url.clone(),
+                access_key: source_env.access_key.clone(),
+                secret_key: source_env.secret_key.clone(),
+                ..Default::default()
+            },
+            PeerSite {
+                name: "target-site".to_string(),
+                endpoint: target_env.url.clone(),
+                access_key: target_env.access_key.clone(),
+                secret_key: target_env.secret_key.clone(),
+                ..Default::default()
+            },
+        ],
+    )
+    .await?;
+    assert!(add_status.success, "unexpected site add result: {add_status:?}");
+
+    let _source_info = wait_for_site_replication_enabled(&source_env, 2).await?;
+    let _target_info = wait_for_site_replication_enabled(&target_env, 2).await?;
+
+    source_client.create_bucket().bucket(bucket).send().await?;
+    enable_bucket_versioning(&source_env, bucket).await?;
+    wait_for_bucket_on_target(&target_client, bucket).await?;
+    let baseline_payload = b"before rotation".to_vec();
+    source_client
+        .put_object()
+        .bucket(bucket)
+        .key("before-rotate.txt")
+        .body(ByteStream::from(baseline_payload.clone()))
+        .send()
+        .await?;
+    let replicated_baseline = wait_for_object_on_target(&target_client, bucket, "before-rotate.txt").await?;
+    assert_eq!(replicated_baseline, baseline_payload);
+
+    // A single rotation call must finish the whole hand-over. Before the fix
+    // the join push could only sign with the freshly installed secret, every
+    // peer rejected it, the rotation stayed pending forever, and both
+    // replication directions were dead until an operator retried.
+    let rotate_status = site_replication_rotate_svc_acct(&source_env).await?;
+    assert!(rotate_status.success, "rotation did not complete in one call: {rotate_status:?}");
+
+    for env in [&source_env, &target_env] {
+        let deadline = std::time::Instant::now() + Duration::from_secs(30);
+        loop {
+            let info = site_replication_info(env).await?;
+            if info.enabled && info.pending_operation.is_none() {
+                break;
+            }
+            if std::time::Instant::now() > deadline {
+                return Err(format!("rotation left {} with a pending operation: {:?}", env.url, info.pending_operation).into());
+            }
+            tokio::time::sleep(Duration::from_millis(500)).await;
+        }
+    }
+
+    // Replication must actually flow again in both directions with the
+    // rotated service-account secret.
+    let forward_payload = b"after rotation from source".to_vec();
+    source_client
+        .put_object()
+        .bucket(bucket)
+        .key("after-rotate-forward.txt")
+        .body(ByteStream::from(forward_payload.clone()))
+        .send()
+        .await?;
+    let replicated_forward = wait_for_object_on_target(&target_client, bucket, "after-rotate-forward.txt").await?;
+    assert_eq!(replicated_forward, forward_payload);
+
+    let reverse_payload = b"after rotation from target".to_vec();
+    target_client
+        .put_object()
+        .bucket(bucket)
+        .key("after-rotate-reverse.txt")
+        .body(ByteStream::from(reverse_payload.clone()))
+        .send()
+        .await?;
+    let replicated_reverse = wait_for_object_on_target(&source_client, bucket, "after-rotate-reverse.txt").await?;
+    assert_eq!(replicated_reverse, reverse_payload);
+
+    Ok(())
+}
+
 #[tokio::test]
 async fn test_site_replication_state_edit_fresh_and_stale_real_dual_node() -> Result<(), Box<dyn Error + Send + Sync>> {
    init_logging();
@@ -9137,7 +9244,7 @@ async fn test_get_and_head_proxy_unreplicated_object_to_replication_target() ->
    // the real SSE-C decryption; the plaintext fake simply ignores them).
    target.take_requests();
    let ssec_key = "01234567890123456789012345678901";
-    let ssec_key_b64 = BASE64_STANDARD.encode(ssec_key);
+    let ssec_key_b64 = BASE64_STANDARD.encode_to_string(ssec_key);
    let ssec_key_md5 = sse_customer_key_md5_base64(ssec_key);
    let _ = source_client
        .get_object()
@@ -21,11 +21,13 @@
 //! - SSRF prevention (internal/private endpoints rejected for tiering)
 //! - Race condition handling (concurrent writes converge without corruption)

-use crate::common::{RustFSTestEnvironment, awscurl_put, init_logging, require_awscurl};
+use crate::common::{RustFSTestEnvironment, init_logging, signed_s3_request};
 use aws_sdk_s3::error::ProvideErrorMetadata;
 use aws_sdk_s3::primitives::ByteStream;
 use aws_sdk_s3::types::{CompletedMultipartUpload, CompletedPart, Tag, Tagging};
 use std::error::Error;
+use std::time::Duration;
+use tokio::net::TcpListener;

 /// Oversized tagging payloads must be rejected by the per-object tag limit.
 ///
@@ -74,8 +76,12 @@ async fn test_large_xml_body_rejection() -> Result<(), Box<dyn Error + Send + Sy
    let _ = client.delete_object().bucket(&bucket_name).key(object_key).send().await;
    let _ = client.delete_bucket().bucket(&bucket_name).send().await;

-    assert!(result.is_err(), "Server must reject an oversized tagging payload, but it was accepted");
-    let err = result.expect_err("checked is_err above");
+    let err = result.expect_err("Server must reject an oversized tagging payload");
+    assert_eq!(
+        err.raw_response().map(|response| response.status().as_u16()),
+        Some(400),
+        "Oversized tagging should return HTTP 400, got: {err:?}"
+    );
    let code = err.as_service_error().and_then(|e| e.code());
    assert_eq!(
        code,
@@ -87,7 +93,7 @@ async fn test_large_xml_body_rejection() -> Result<(), Box<dyn Error + Send + Sy
    Ok(())
 }

-/// Excessive multipart parts must be rejected.
+/// Multipart completion must reject part numbers above the 10,000-part limit.
 #[tokio::test]
 async fn test_excessive_multipart_parts() -> Result<(), Box<dyn Error + Send + Sync>> {
    init_logging();
@@ -107,11 +113,24 @@ async fn test_excessive_multipart_parts() -> Result<(), Box<dyn Error + Send + S

    let upload_id = create_result.upload_id().expect("upload_id should be present").to_string();

-    // Try to complete with too many parts (should be rejected).
-    let mut parts = Vec::new();
-    for i in 1..=10001 {
-        parts.push(CompletedPart::builder().part_number(i).e_tag(format!("etag-{i}")).build());
-    }
+    // Upload one real control part so a generic missing-part rejection cannot
+    // masquerade as enforcement of the 10,000-part boundary.
+    let uploaded = client
+        .upload_part()
+        .bucket(&bucket_name)
+        .key("test-large")
+        .upload_id(&upload_id)
+        .part_number(1)
+        .body(ByteStream::from_static(b"valid-control-part"))
+        .send()
+        .await?;
+    let parts = vec![
+        CompletedPart::builder()
+            .part_number(1)
+            .e_tag(uploaded.e_tag().expect("uploaded part should return an ETag"))
+            .build(),
+        CompletedPart::builder().part_number(10001).e_tag("out-of-range").build(),
+    ];

    let result = client
        .complete_multipart_upload()
@@ -132,7 +151,21 @@ async fn test_excessive_multipart_parts() -> Result<(), Box<dyn Error + Send + S
        .await;
    let _ = client.delete_bucket().bucket(&bucket_name).send().await;

-    assert!(result.is_err(), "Server should reject excessive multipart parts");
+    let error = result.expect_err("server should reject a completion part number above 10000");
+    assert_eq!(
+        error.raw_response().map(|response| response.status().as_u16()),
+        Some(400),
+        "part-limit rejection must return HTTP 400, got: {error:?}"
+    );
+    let service_error = error
+        .as_service_error()
+        .expect("part-limit rejection must be an S3 service error");
+    assert_eq!(service_error.code(), Some("InvalidPart"), "unexpected part-limit error: {error:?}");
+    assert_eq!(
+        service_error.message(),
+        Some("Part number 10001 must be between 1 and 10000"),
+        "completion must fail at the part-number boundary, not a later missing-part check"
+    );

    env.stop_server();
    Ok(())
@@ -197,15 +230,14 @@ async fn test_concurrent_object_operations() -> Result<(), Box<dyn Error + Send
    // After deletion, the object must be absent.
    client.delete_object().bucket(&bucket_name).key(key).send().await?;
    let get_after_delete = client.get_object().bucket(&bucket_name).key(key).send().await;
-    assert!(get_after_delete.is_err(), "Object must be absent after delete");
-    let code = get_after_delete
-        .err()
-        .and_then(|e| e.as_service_error().and_then(|se| se.code()).map(str::to_string));
+    let error = get_after_delete.expect_err("Object must be absent after delete");
    assert_eq!(
-        code.as_deref(),
-        Some("NoSuchKey"),
-        "GET after delete should return NoSuchKey, got {code:?}"
+        error.raw_response().map(|response| response.status().as_u16()),
+        Some(404),
+        "GET after delete should return HTTP 404, got: {error:?}"
    );
+    let code = error.as_service_error().and_then(|service_error| service_error.code());
+    assert_eq!(code, Some("NoSuchKey"), "GET after delete should return NoSuchKey, got: {error:?}");

    let _ = client.delete_bucket().bucket(&bucket_name).send().await;

@@ -213,35 +245,31 @@ async fn test_concurrent_object_operations() -> Result<(), Box<dyn Error + Send
    Ok(())
 }

-/// Internal/private endpoints must be rejected as remote tier backends (SSRF).
+/// Internal/private endpoints must be rejected as remote tier backends (SSRF)
+/// before RustFS opens a connection to them.
 ///
 /// This issues a real admin AddTier call (`PUT /rustfs/admin/v3/tier`) for each
-/// internal/private endpoint and asserts the server rejects it (non-2xx, so the
-/// signed request helper returns an error). An internal endpoint must never be
-/// accepted as a tier backend. The rejection may originate from explicit
-/// SSRF/internal-address filtering or from the backend connectivity/credential
-/// validation performed during AddTier; either way the security-relevant
-/// outcome — the internal endpoint is not accepted — is asserted here.
-///
-/// The admin API is exercised via signed `awscurl` requests, matching the
-/// pattern used by the other admin-API E2E tests in this crate. The full E2E
-/// lane installs and verifies the pinned `awscurl` prerequisite.
+/// internal/private endpoint and requires the exact validation response. A
+/// loopback listener additionally proves the literal loopback case is rejected
+/// without an outbound connection; a generic connectivity failure is not
+/// sufficient evidence of SSRF protection.
 #[tokio::test]
 async fn test_tiering_url_validation() -> Result<(), Box<dyn Error + Send + Sync>> {
    init_logging();
-    require_awscurl()?;
    let mut env = RustFSTestEnvironment::new().await?;
    env.start_rustfs_server(vec![]).await?;

    let tier_url = format!("{}/rustfs/admin/v3/tier", env.url);
+    let sentinel = TcpListener::bind("127.0.0.1:0").await?;
+    let sentinel_endpoint = format!("http://{}", sentinel.local_addr()?);
    let internal_endpoints = [
-        "http://127.0.0.1:8080",
-        "http://localhost:8080",
-        "http://169.254.169.254", // cloud instance metadata endpoint
-        "http://[::1]:8080",
+        (sentinel_endpoint.as_str(), true),
+        ("http://localhost:8080", false),
+        ("http://169.254.169.254", false), // cloud instance metadata endpoint
+        ("http://[::1]:8080", false),
    ];

-    for endpoint in internal_endpoints {
+    for (endpoint, checks_connection_attempt) in internal_endpoints {
        // AddTier expects an uppercase tier name and a backend configuration.
        let body = serde_json::json!({
            "type": "s3",
@@ -258,11 +286,30 @@ async fn test_tiering_url_validation() -> Result<(), Box<dyn Error + Send + Sync
        })
        .to_string();

-        let result = awscurl_put(&tier_url, &body, &env.access_key, &env.secret_key).await;
+        let response = signed_s3_request(
+            http::Method::PUT,
+            &tier_url,
+            Some(body),
+            Some("application/json"),
+            &env.access_key,
+            &env.secret_key,
+        )
+        .await?;
+        let status = response.status();
+        let response_body = response.text().await?;
+        assert_eq!(status, 400, "AddTier must reject internal endpoint {endpoint} with HTTP 400");
        assert!(
-            result.is_err(),
-            "AddTier must reject internal endpoint {endpoint}, but it was accepted: {result:?}"
+            response_body.contains("<Code>InvalidArgument</Code>") && response_body.contains("tier endpoint is not allowed"),
+            "AddTier must reject internal endpoint {endpoint} during URL validation, got: {response_body}"
        );
+
+        if checks_connection_attempt {
+            match tokio::time::timeout(Duration::from_secs(1), sentinel.accept()).await {
+                Err(_) => {}
+                Ok(Ok((_, peer))) => panic!("SSRF guard connected to loopback endpoint {endpoint} from {peer}"),
+                Ok(Err(error)) => return Err(error.into()),
+            }
+        }
    }

    env.stop_server();
--- a/Show More
+++ b/Show More