Include the erasure set drive count in the distributed EC8+4 drive restart oracle so Scanner/Heal release evidence matches the registry contract.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
Handle cargo-nextest writing JUnit reports under the workspace target directory even when the Rust build uses CARGO_TARGET_DIR. This keeps measured Scanner/Heal evidence cases from passing the real test but failing final receipt packaging.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Keep the multi-pool evidence runner within the registry object budget while retaining deferred outage-write diagnostics for release-gate validation.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Include optional outage-write diagnostics in G14 evidence oracles so deferred multi-pool outage writes bind their down-window refusals and post-rejoin acceptance.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
test(e2e): keep multi-set restart graceful
Include the EC8+4 multi-set restart scenario in the graceful interruption lane so the unclean-shutdown marker assertion matches the scenario semantics.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Use the pool-wide namespace lock client domain for every set in a pool so degraded EC8+4 multi-set writes are gated by node-level lock quorum instead of the narrower per-set endpoint host slice.
Keep namespace-lock domain deduplication tied to both the pool namespace and shared clients, and add regression coverage for three-locker degraded writes plus cross-pool domain separation.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
fix(ecstore): admit metadata snapshots at read quorum
Allow guarded bucket metadata snapshot existence checks to use read quorum so degraded erasure sets can continue Object Lock snapshot reads without weakening bucket mutation or object write quorum.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Validate the G14 multi-pool oracle field that records down-window outage PUT refusal and requires the deferred post-rejoin outage object to be accepted and checked through S3.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
test(scanner): align G14 multi-pool outage evidence
Treat the localhost multi-pool topology as whole-pool loss when the target node is down. If that topology cannot admit the outage object while the pool is offline, defer that object write until the pool rejoins and record the oracle marker instead of failing before the real crash/restart evidence runs.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Add a status mode for the Scanner/Heal Linux evidence planner so operators can identify missing or incomplete artifacts before final bundle assembly.
The status output stays plan-only and never reports release approval.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Choose the replacement disk from the target node by the presence of complete pool metadata instead of assuming the first configured drive is the scanner metadata holder. This keeps the G14 multi-set harness aligned with multi-drive EC layouts and adds clearer diagnostics when multi-pool outage writes fail closed.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Add a Scanner/Heal Linux release-evidence planner that emits a machine-readable execution manifest for the remaining measured validation lanes.
The planner can run only lightweight preflight checks and keeps plan-only output distinct from measured release evidence.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Require lost scoped dirty-usage ACK reconciliation to observe a clean peer activity generation that covers the requested ACK generation. Apply the same guard in the scanner aggregation path so a stale clean activity snapshot cannot discharge pending maintenance after an uncertain ACK response.
Refs rustfs/backlog#2427
Refs rustfs/backlog#2281
Refs rustfs/backlog#2240
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Normalize live Scanner/Heal status/outcome observations into the measured raw artifacts consumed by the G05/G06/R-D release descriptor producer.
Reject synthetic observations, incomplete required cases, and reused run/window identities before raw artifacts can enter the release bundle flow.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Harden the Scanner/Heal checkpoint restart and G14 EC evidence producers so release descriptors cannot be assembled from marked fixture, dry-run, or synthetic case inputs.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
* fix(scanner): bound native backlog fixture resources
Keep native scanner backlog restart fixtures within low file descriptor limits by reducing the native-only disk layout, making fixture shutdown drain background work, and avoiding long-lived ECStore retention from background loops.
Convert ECStore-backed background refresh/recovery/monitor tasks to upgrade weak owners only while doing work so completed test stores release their disk graph before the next native fixture starts.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
* test(tier): refresh release-merged failure fixtures
Reuse the raw legacy-transition fixture helper so delete-all expiry actually exercises Unknown transitioned history metadata after the release merge.
Rewrite restore-failure disk fixtures through per-disk xl.meta snapshots so stale destination identities can be persisted without tripping ordinary metadata update guards.
Update the SSE KMS mismatch expectation to the InvalidRequest/context_mismatch behavior now returned by the rio-v2 path.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
---------
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
* feat(nightly): publish packages as assets of the rolling 'nightly' release (sync from release) (#7593)
feat(nightly): publish packages as assets of the rolling 'nightly' release (#7592)
Replace the assets-branch scheme with a proper GitHub Release on
rustfs/auto-testing: a single 'nightly' release whose deb/rpm assets
are replaced in place on every build. This is the standard channel —
visible on the repo's Releases page, stable download URLs, no git
history growth (release assets live outside the repository).
- New scripts/release/publish_nightly_assets.sh: resolves-or-creates
the 'nightly' release via the REST API, deletes same-name assets,
uploads rustfs-nightly-latest.{deb,rpm}, then PATCHes the release
body with the build provenance (ref@sha, run link, sizes, SHA256).
Plain curl + python3, no gh CLI (the build fleet has none — #7586).
- The workflow step shrinks to invoking the script; full flow
exercised end-to-end against the real release with probe files
(create / upload / overwrite / download round-trip / body update).
* test(scanner): stabilize W13 release gate evidence
Treat only real raw-entry windows as replayed scanner enumeration in the restart diagnostic, so final completion rounds without raw entries are not fail-closed as raw replays.
Allow the EC8:4 multi-pool heal evidence case to select an outage object key that routes to an online pool while an entire target pool is down, preserving strict behavior for single-pool cases.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
---------
Co-authored-by: hector <42570491+majinghe@users.noreply.github.com>
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Require Scanner/Heal G14 release bundle JSON wrappers to mirror their outer evidence and carry self-contained case artifact provenance.
Copy proof-json case artifacts into the generated G14 descriptor bundle so assembled release bundles can validate case file hashes after relocation.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Classify a final remote scanner publication lease validation failure from a fresh activity snapshot instead of treating every validation error as data movement. If the peer session and movement generation still match the granted leases and publication is otherwise allowed, keep the publication rejected as an activity-baseline miss without creating pause backlog movement debt. Preserve DataMovement for expired leases, peer restarts, movement generation changes, and active publication blocks.
Tests cover namespace-only validation invalidation and the remote fence-loss cases that must still defer as DataMovement.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Require measured G05/G06/R-D raw status, compatibility, and disposition artifacts to share run identity before producing release bundle descriptors.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Require Scanner/Heal G03 scoped ACK evidence fields to carry their own measured provenance and concrete ACK, capability, and mixed-peer observations before release-bundle gate verification.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Require Scanner/Heal release profile wrappers to carry bundled raw profile artifacts plus the matching profile cost metrics.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
* fix(scanner): seed current backlog members before committing
* test(app): run recursive force delete test on large stack
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
---------
Co-authored-by: houseme <housemecn@gmail.com>
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Drop an unnecessary clone in the legacy transition state digest regression so release CI clippy lanes can run with warnings denied.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Replace the assets-branch scheme with a proper GitHub Release on
rustfs/auto-testing: a single 'nightly' release whose deb/rpm assets
are replaced in place on every build. This is the standard channel —
visible on the repo's Releases page, stable download URLs, no git
history growth (release assets live outside the repository).
- New scripts/release/publish_nightly_assets.sh: resolves-or-creates
the 'nightly' release via the REST API, deletes same-name assets,
uploads rustfs-nightly-latest.{deb,rpm}, then PATCHes the release
body with the build provenance (ref@sha, run link, sizes, SHA256).
Plain curl + python3, no gh CLI (the build fleet has none — #7586).
- The workflow step shrinks to invoking the script; full flow
exercised end-to-end against the real release with probe files
(create / upload / overwrite / download round-trip / body update).
Keep heal admission-only tests off durable root recovery so mock storage does not require local metadata disks. Align forceStart retry assertions with replacement semantics and fix the e2e metric clippy collapse.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Treat an authoritative scanner pause-backlog commit as stable when every member in that commit stores the matching stable and committed record. Retired source replicas can remain missing in the observed topology while survivor membership is already stable.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
fix(nightly): push assets with plain git — the build fleet has no gh CLI
The sm-standard-4 runners used by the nightly build have no gh binary
(functional-chain workflows run elsewhere, on the jumpbox). The assets
publish step died on 'gh: command not found' at the credential-helper
setup, and the preceding clone failure had been masked by 2>/dev/null,
misleading the step into the orphan path. Swap clone and remote setup
to plain git with the token embedded in the URL; push semantics are
unchanged.
Add EC8+4 multi-set and multi-pool scanner/heal release evidence coverage, including e2e registry cases, oracle checks, and a G14 descriptor assembler for measured case artifacts.
Signed-off-by: houseme <housemecn@gmail.com>
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Persist scanner dirty producer replay records from committed replication and tier mutations, hydrate journal state on restart, and clear durable records only after acknowledged dirty generations.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Bound durable root-heal terminal receipt retention to the completed-status TTL while keeping expired receipts authoritative enough to retire stale pending intents before deleting the receipt itself.
Add a budgeted background GC pass and fail closed on corrupt or non-terminal receipt records.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
* fix(scanner): mark committed replication and tier mutations
Bridge ECStore committed replication status writebacks and lifecycle tier expiration cleanup into scanner dirty-usage producer identity tracking. Only successful replication metadata updates mark Replication; validate-only, superseded, and retry paths remain unmarked.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
* feat(scanner): replay durable dirty producer records
Add a strict scanner dirty-usage producer replay schema that can restore typed producer coverage after restart without treating ordinary process-local mutations as durable evidence. Replay records validate schema, cache key format, generations, scopes, producer identities, duplicate buckets, byte limits, and entry limits before modifying dirty state.
Keep segment activation fail-closed unless the restored bucket state came from durable replay; subsequent local mutations clear the durable authority bit for that bucket. Also fix the release scanner clippy lint in scoped cold-reuse proof creation.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
---------
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Record an explicit cold bucket reuse proof for scoped set scans and treat fully cold sets as valid cold-zero-walk reuse when the baseline and bucket incarnations are bound.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Fail closed when a signed scoped ACK response claims more cleared entries than its request contained, while preserving partial-clear semantics and zero-mutation capability probes.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Track producer identity completeness per pending bucket and retain process coverage across durable acknowledgements. Admit a prior same-process segment proof only for an exact or strictly earlier fully typed generation window.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Configure the s3-tests harness with a local KMS key so SSE-KMS cases run in CI without relying on an external KMS service.
Also move the anonymous POST default SSE-KMS regression onto the shared local KMS test environment.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Add a W13 durable MRF evidence runner that emits measured G07, G08, and P4 JSON artifacts and release descriptors through the existing scanner/heal bundle gate.
The runner now executes the ignored MRF replay evidence test with an exact full test path, validates raw artifact kinds and gate decisions, prepares Linux tmpfs-backed ENOSPC roots for G08, and documents the Linux/long-soak boundaries.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Refresh the committed e2e-distributed selection digest after the EC8+4 Scanner/Heal evidence case joined the profile, and document the registry-aware runner entry for the distributed case.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Linux libtest defaults can abort these deep EC-store futures before the assertions run. Wrap the async scoped fallback cases in the repository's dedicated-stack test pattern so the default cargo test path exercises the oracle directly.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
* feat(scanner): admit durable segment proof evidence
Validate complete set snapshot segment invalidation proof metadata against the current dirty usage generation window and scanner process epoch before clearing durable producer and restart-gap activation blockers.
Production segment reuse remains gated by the explicit activation flag.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
* feat(scanner): replay segment proof from root snapshots
Carry segment invalidation proof metadata into root snapshot set states so a complete published baseline can replay the durable producer evidence recorded by each set cache.
Keep the field additive for older readers and leave incomplete or LKG set states unproven.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
---------
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Add a W16 Scanner/Heal evidence runner that executes the recovery-intent crash-boundary and quota-authority lanes, emits measured G04/G12 JSON artifacts, and validates the resulting single-gate release descriptors.
Wire its shell self-test into script-tests and document the release evidence entry point.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Align the Scanner/Heal release requirement registry with the release bundle gate for P4 so the closure checklist advertises MRF scale, replay cost, retained responsibility, and cleanup/GC soak evidence together.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Record scanner segment producer identities with the dirty usage generation they invalidated, then expose a cycle-local producer evidence snapshot for segment reuse activation preflight.
Production activation remains fail-closed until durable producer identity and restart-gap proof are available.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Require Scanner/Heal release bundle JSON artifacts to carry hard-domain evidence fields for mixed-version, crash, capacity, disk-full, replica-loss, and MRF cleanup gates. This prevents a descriptor from approving a hard gate while pointing at a generic measured artifact that lacks the boundary-specific oracle fields.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Exercise the committed MRF replay checkpoint path with a real ECStore-backed bucket incarnation, retain the replay anchor until an exact verified proof arrives, then delete the proof-discharged checkpoint and prove a subsequent restart does not resurrect the replayed work.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Keep checksum verification output out of the command substitution that resolves the previous-release binary for the G09 runner.
Also tolerate non-GNU sha256sum in local self-tests by falling back to shasum when GNU --check support is unavailable.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Validate the selected G09 evidence lane before process-substitution case expansion so --plan-only cannot turn an unknown --test value into an empty successful plan.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
* fix(replication): close the GA blocker set from backlog#2366 (#7503)
* fix(replication): close GA blockers from backlog#2366
Implements the P1 set from the pre-GA replication audit:
- Replication rule tag filters now require every And.Tag to match, replacing
the s3s OR semantics with a local AND matcher that fails closed on a
malformed tag.
- A replicated group membership change no longer writes the group status, so
a membership update carrying the default Enabled status cannot silently
re-enable a disabled group on the peer.
- A successful IAM import schedules one collapsed full-IAM snapshot per remote
peer instead of leaving the imported entities local-only.
- A pending endpoint refresh is redriven by the heavyweight reconcile tick,
carries its own ilm-expiry override, and no longer blocks a remove that
drops every unacknowledged peer.
- Site metrics expose local replication failure totals and rolling windows;
node-level counters no longer report a constructed zero.
- set/remove-remote-target notify peer metadata caches before returning, so a
follow-up put-bucket-replication on another node sees the target.
- Adds the site-replication operations runbook, a docs index, a replication
support boundary section, and the Replication changelog section.
* fix(site-replication): resume only a locally driven endpoint refresh
The peer-side edit handler journals a pending endpoint refresh with an empty
`remote_peers` map and commits it inside the same request through
`apply_internal_peer_edit`. The reconcile tick could not tell that journal
from the coordinator's own: with no required peers it reads as complete on
sight, so the tick committed it with `edit_state` - losing the local-name
sync - and cleared it under the request that owned it, whose commit then
reported the refresh as changed and denied the coordinator the peer
acknowledgement it was waiting for.
Resume now runs only for a journal that carries the fan-out topology. A
receiver's journal stays for the coordinator to redrive with the same
refresh id, which is the path that already recovers it.
* fix(site-replication): keep an explicit disabled group status on a snapshot
Skipping the group-status write whenever an item carries members stopped a
membership change from re-enabling a disabled group, but it also silenced the
full-IAM snapshot, which always sends members together with the sender's real
status. A peer that did not have the group yet created it through
`GroupInfo::new` - enabled - so a bootstrap, a repair, or the snapshot an IAM
import now schedules handed every member of a frozen group live access there.
The madmin wire maps an unset `groupStatus` to Enabled, so only Enabled can be
a default. Disabled is always explicit and is applied again.
* fix(site-replication): schedule the import snapshot without recording a failure
`import-iam` reused the failure-recording path to queue its full-IAM
snapshot. That raises `retry_count` on every call, so three imports - the
normal shape of a bulk migration done one archive at a time - escalated a
healthy peer to `retryStats.failed` with the scheduling note shown as
`lastError`, which is exactly the signal the runbook tells operators to
repair. A full retry queue also turned a completed import into a 503.
Scheduling now only ensures the collapsed entry exists, and a failure to
schedule is logged instead of failing the request: the entities are already
imported and the reconcile pass still closes the gap.
* fix(admin): stop reporting replication failures as retries
`retries` is the minio-go counter for redeliveries, and mc prints it as such.
Filling it with the failure count claimed a redelivery that never happens: a
failed object is not retried by an event today, it waits for the scanner heal
pass. `errors` keeps the failure counters; `retries` stays zero until there is
a real redelivery to count, and the runbook now says so.
* perf(site-replication): aggregate failure windows without cloning bucket stats
`site_metrics_snapshot` went through `get_all`, which clones every bucket's
stats, and then scanned each target's sample deque twice. That deque is
bounded only by the one-hour window, so an unreachable target under load -
the case an operator polls this endpoint for - made every
`mc admin replicate status` copy the whole backlog and hold the read lock
against the failure path while doing it.
It now folds under the read lock and takes both windows in one walk. The
`max` against the serialized `last_minute` / `last_hour` snapshots is dropped:
those are stamped onto per-bucket clones elsewhere and are always zero in this
node-local cache.
* fix(site-replication): reject a conflicting ilm-expiry override on a re-run
The commit now reads the ilm-expiry override back out of the pending refresh
journal, so a second edit that asks for a different value had it dropped while
the request still reported success. Re-running without the flag keeps pinning
the recorded value - that is the documented way to redrive a stuck refresh -
but an explicit different value is now rejected instead of ignored.
* fix(admin): do not fail a remote-target write on a peer reload error
set/remove-remote-target propagated the peer metadata reload error, so a
target that was already persisted and live on this node reported a 5xx to the
client whenever one peer could not be reached. Every S3 bucket-config write
path treats that reload as best effort and only warns; these two admin
handlers now do the same, and the reason is logged with the bucket and action.
* fix(site-replication): undo every bucket a cut-short refresh rewrote
When a remove accepted on another node clears the refresh journal mid-pass,
only the bucket holding the lock at that moment had its restored target
undone. The buckets rewritten earlier in the same pass kept a target pointing
at the removed peer whenever the remove's own cleanup had already walked past
them. The undo now covers every bucket this pass rewrote, attempting all of
them so one failure does not strand the rest.
* fix(site-replication): keep replay running while an endpoint refresh is pending
A pending endpoint refresh took the whole heavyweight pass with it, so a peer
that never came back froze IAM and bucket replay to every healthy peer too -
the stall this journal's resume path was meant to end. The refresh arm now
drains the retry queue before returning; it replays per-peer deliveries
against the endpoints currently committed in state, so it is unaffected by the
edit in flight. Bucket wiring reconciliation still waits, because it rewrites
the very targets the refresh is changing, and the runbook now says so.
* test(e2e): cover the AND semantics of a two-tag replication filter
The acceptance matrix only had a single-tag rule, which matches under both AND
and OR semantics and therefore proved nothing about the filter this fix
changed. It now also carries a two-tag `And` rule - the shape
`mc replicate add --tags "k1=v1&k2=v2"` writes - and asserts that an object
with one of the two tags is not admitted while an object with both is.
No new test function, so the nightly selection digest is unchanged.
* refactor(site-replication): fold the refresh state-change error into one constructor
The endpoint-refresh work added three `s3_error!` invocation lines, which the
s3s footprint ratchet is meant to prevent. Five copies of the same
concurrent-change error now share one constructor, so the surface nets one
line smaller than main; the baseline is retightened to match.
* fix(site-replication): report a peer whose IAM snapshot waits for a repair
An escalated snapshot entry records a deletion a snapshot cannot replay, so
only a repair settles it and the marker must survive. Scheduling an import
snapshot therefore leaves that peer's entry alone - and now says so, instead
of returning success while nothing was scheduled for it.
* docs(operations): state the group-status and escalation convergence limits
Two boundaries the fixes in this branch make load-bearing: a membership change
never carries an enable, so a group disabled on one site only has to be
re-enabled there explicitly; and a peer holding an escalated IAM entry does
not receive a scheduled snapshot, including the one a bulk import schedules,
until a repair settles it.
* fix(ci): bind performance runs to selected inputs (#7512)
* fix(targets): reject trailing batch items (#7508)
* test(scanner): emit G09 release bundle gate evidence
Write a bundle-ready G09 gate descriptor from the Linux upgrade evidence runner and validate the single G09 gate with the shared release-bundle rules without approving the full Scanner/Heal release.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
---------
Co-authored-by: 唐小鸭 <tangtang1251@qq.com>
Co-authored-by: Zhengchao An <anzhengchao@gmail.com>
Co-authored-by: cui fliter <imcusg@gmail.com>
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
* fix(replication): close the GA blocker set from backlog#2366 (#7503)
* fix(replication): close GA blockers from backlog#2366
Implements the P1 set from the pre-GA replication audit:
- Replication rule tag filters now require every And.Tag to match, replacing
the s3s OR semantics with a local AND matcher that fails closed on a
malformed tag.
- A replicated group membership change no longer writes the group status, so
a membership update carrying the default Enabled status cannot silently
re-enable a disabled group on the peer.
- A successful IAM import schedules one collapsed full-IAM snapshot per remote
peer instead of leaving the imported entities local-only.
- A pending endpoint refresh is redriven by the heavyweight reconcile tick,
carries its own ilm-expiry override, and no longer blocks a remove that
drops every unacknowledged peer.
- Site metrics expose local replication failure totals and rolling windows;
node-level counters no longer report a constructed zero.
- set/remove-remote-target notify peer metadata caches before returning, so a
follow-up put-bucket-replication on another node sees the target.
- Adds the site-replication operations runbook, a docs index, a replication
support boundary section, and the Replication changelog section.
* fix(site-replication): resume only a locally driven endpoint refresh
The peer-side edit handler journals a pending endpoint refresh with an empty
`remote_peers` map and commits it inside the same request through
`apply_internal_peer_edit`. The reconcile tick could not tell that journal
from the coordinator's own: with no required peers it reads as complete on
sight, so the tick committed it with `edit_state` - losing the local-name
sync - and cleared it under the request that owned it, whose commit then
reported the refresh as changed and denied the coordinator the peer
acknowledgement it was waiting for.
Resume now runs only for a journal that carries the fan-out topology. A
receiver's journal stays for the coordinator to redrive with the same
refresh id, which is the path that already recovers it.
* fix(site-replication): keep an explicit disabled group status on a snapshot
Skipping the group-status write whenever an item carries members stopped a
membership change from re-enabling a disabled group, but it also silenced the
full-IAM snapshot, which always sends members together with the sender's real
status. A peer that did not have the group yet created it through
`GroupInfo::new` - enabled - so a bootstrap, a repair, or the snapshot an IAM
import now schedules handed every member of a frozen group live access there.
The madmin wire maps an unset `groupStatus` to Enabled, so only Enabled can be
a default. Disabled is always explicit and is applied again.
* fix(site-replication): schedule the import snapshot without recording a failure
`import-iam` reused the failure-recording path to queue its full-IAM
snapshot. That raises `retry_count` on every call, so three imports - the
normal shape of a bulk migration done one archive at a time - escalated a
healthy peer to `retryStats.failed` with the scheduling note shown as
`lastError`, which is exactly the signal the runbook tells operators to
repair. A full retry queue also turned a completed import into a 503.
Scheduling now only ensures the collapsed entry exists, and a failure to
schedule is logged instead of failing the request: the entities are already
imported and the reconcile pass still closes the gap.
* fix(admin): stop reporting replication failures as retries
`retries` is the minio-go counter for redeliveries, and mc prints it as such.
Filling it with the failure count claimed a redelivery that never happens: a
failed object is not retried by an event today, it waits for the scanner heal
pass. `errors` keeps the failure counters; `retries` stays zero until there is
a real redelivery to count, and the runbook now says so.
* perf(site-replication): aggregate failure windows without cloning bucket stats
`site_metrics_snapshot` went through `get_all`, which clones every bucket's
stats, and then scanned each target's sample deque twice. That deque is
bounded only by the one-hour window, so an unreachable target under load -
the case an operator polls this endpoint for - made every
`mc admin replicate status` copy the whole backlog and hold the read lock
against the failure path while doing it.
It now folds under the read lock and takes both windows in one walk. The
`max` against the serialized `last_minute` / `last_hour` snapshots is dropped:
those are stamped onto per-bucket clones elsewhere and are always zero in this
node-local cache.
* fix(site-replication): reject a conflicting ilm-expiry override on a re-run
The commit now reads the ilm-expiry override back out of the pending refresh
journal, so a second edit that asks for a different value had it dropped while
the request still reported success. Re-running without the flag keeps pinning
the recorded value - that is the documented way to redrive a stuck refresh -
but an explicit different value is now rejected instead of ignored.
* fix(admin): do not fail a remote-target write on a peer reload error
set/remove-remote-target propagated the peer metadata reload error, so a
target that was already persisted and live on this node reported a 5xx to the
client whenever one peer could not be reached. Every S3 bucket-config write
path treats that reload as best effort and only warns; these two admin
handlers now do the same, and the reason is logged with the bucket and action.
* fix(site-replication): undo every bucket a cut-short refresh rewrote
When a remove accepted on another node clears the refresh journal mid-pass,
only the bucket holding the lock at that moment had its restored target
undone. The buckets rewritten earlier in the same pass kept a target pointing
at the removed peer whenever the remove's own cleanup had already walked past
them. The undo now covers every bucket this pass rewrote, attempting all of
them so one failure does not strand the rest.
* fix(site-replication): keep replay running while an endpoint refresh is pending
A pending endpoint refresh took the whole heavyweight pass with it, so a peer
that never came back froze IAM and bucket replay to every healthy peer too -
the stall this journal's resume path was meant to end. The refresh arm now
drains the retry queue before returning; it replays per-peer deliveries
against the endpoints currently committed in state, so it is unaffected by the
edit in flight. Bucket wiring reconciliation still waits, because it rewrites
the very targets the refresh is changing, and the runbook now says so.
* test(e2e): cover the AND semantics of a two-tag replication filter
The acceptance matrix only had a single-tag rule, which matches under both AND
and OR semantics and therefore proved nothing about the filter this fix
changed. It now also carries a two-tag `And` rule - the shape
`mc replicate add --tags "k1=v1&k2=v2"` writes - and asserts that an object
with one of the two tags is not admitted while an object with both is.
No new test function, so the nightly selection digest is unchanged.
* refactor(site-replication): fold the refresh state-change error into one constructor
The endpoint-refresh work added three `s3_error!` invocation lines, which the
s3s footprint ratchet is meant to prevent. Five copies of the same
concurrent-change error now share one constructor, so the surface nets one
line smaller than main; the baseline is retightened to match.
* fix(site-replication): report a peer whose IAM snapshot waits for a repair
An escalated snapshot entry records a deletion a snapshot cannot replay, so
only a repair settles it and the marker must survive. Scheduling an import
snapshot therefore leaves that peer's entry alone - and now says so, instead
of returning success while nothing was scheduled for it.
* docs(operations): state the group-status and escalation convergence limits
Two boundaries the fixes in this branch make load-bearing: a membership change
never carries an enable, so a group disabled on one site only has to be
re-enabled there explicitly; and a peer holding an escalated IAM entry does
not receive a scheduled snapshot, including the one a bulk import schedules,
until a repair settles it.
* fix(ci): bind performance runs to selected inputs (#7512)
* test(e2e): add G09 upgrade evidence runner
Add a Linux x86_64 runner that downloads the pinned previous release, builds the current RustFS binary, runs the mixed-version and rollback upgrade compatibility lanes, and verifies the required Scanner/Heal G09 raw evidence artifacts.
Document the runner and add a shell self-test for help, dry-run, SHA validation, and non-empty artifact directory guards.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
---------
Co-authored-by: 唐小鸭 <tangtang1251@qq.com>
Co-authored-by: Zhengchao An <anzhengchao@gmail.com>
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
* fix(replication): close the GA blocker set from backlog#2366 (#7503)
* fix(replication): close GA blockers from backlog#2366
Implements the P1 set from the pre-GA replication audit:
- Replication rule tag filters now require every And.Tag to match, replacing
the s3s OR semantics with a local AND matcher that fails closed on a
malformed tag.
- A replicated group membership change no longer writes the group status, so
a membership update carrying the default Enabled status cannot silently
re-enable a disabled group on the peer.
- A successful IAM import schedules one collapsed full-IAM snapshot per remote
peer instead of leaving the imported entities local-only.
- A pending endpoint refresh is redriven by the heavyweight reconcile tick,
carries its own ilm-expiry override, and no longer blocks a remove that
drops every unacknowledged peer.
- Site metrics expose local replication failure totals and rolling windows;
node-level counters no longer report a constructed zero.
- set/remove-remote-target notify peer metadata caches before returning, so a
follow-up put-bucket-replication on another node sees the target.
- Adds the site-replication operations runbook, a docs index, a replication
support boundary section, and the Replication changelog section.
* fix(site-replication): resume only a locally driven endpoint refresh
The peer-side edit handler journals a pending endpoint refresh with an empty
`remote_peers` map and commits it inside the same request through
`apply_internal_peer_edit`. The reconcile tick could not tell that journal
from the coordinator's own: with no required peers it reads as complete on
sight, so the tick committed it with `edit_state` - losing the local-name
sync - and cleared it under the request that owned it, whose commit then
reported the refresh as changed and denied the coordinator the peer
acknowledgement it was waiting for.
Resume now runs only for a journal that carries the fan-out topology. A
receiver's journal stays for the coordinator to redrive with the same
refresh id, which is the path that already recovers it.
* fix(site-replication): keep an explicit disabled group status on a snapshot
Skipping the group-status write whenever an item carries members stopped a
membership change from re-enabling a disabled group, but it also silenced the
full-IAM snapshot, which always sends members together with the sender's real
status. A peer that did not have the group yet created it through
`GroupInfo::new` - enabled - so a bootstrap, a repair, or the snapshot an IAM
import now schedules handed every member of a frozen group live access there.
The madmin wire maps an unset `groupStatus` to Enabled, so only Enabled can be
a default. Disabled is always explicit and is applied again.
* fix(site-replication): schedule the import snapshot without recording a failure
`import-iam` reused the failure-recording path to queue its full-IAM
snapshot. That raises `retry_count` on every call, so three imports - the
normal shape of a bulk migration done one archive at a time - escalated a
healthy peer to `retryStats.failed` with the scheduling note shown as
`lastError`, which is exactly the signal the runbook tells operators to
repair. A full retry queue also turned a completed import into a 503.
Scheduling now only ensures the collapsed entry exists, and a failure to
schedule is logged instead of failing the request: the entities are already
imported and the reconcile pass still closes the gap.
* fix(admin): stop reporting replication failures as retries
`retries` is the minio-go counter for redeliveries, and mc prints it as such.
Filling it with the failure count claimed a redelivery that never happens: a
failed object is not retried by an event today, it waits for the scanner heal
pass. `errors` keeps the failure counters; `retries` stays zero until there is
a real redelivery to count, and the runbook now says so.
* perf(site-replication): aggregate failure windows without cloning bucket stats
`site_metrics_snapshot` went through `get_all`, which clones every bucket's
stats, and then scanned each target's sample deque twice. That deque is
bounded only by the one-hour window, so an unreachable target under load -
the case an operator polls this endpoint for - made every
`mc admin replicate status` copy the whole backlog and hold the read lock
against the failure path while doing it.
It now folds under the read lock and takes both windows in one walk. The
`max` against the serialized `last_minute` / `last_hour` snapshots is dropped:
those are stamped onto per-bucket clones elsewhere and are always zero in this
node-local cache.
* fix(site-replication): reject a conflicting ilm-expiry override on a re-run
The commit now reads the ilm-expiry override back out of the pending refresh
journal, so a second edit that asks for a different value had it dropped while
the request still reported success. Re-running without the flag keeps pinning
the recorded value - that is the documented way to redrive a stuck refresh -
but an explicit different value is now rejected instead of ignored.
* fix(admin): do not fail a remote-target write on a peer reload error
set/remove-remote-target propagated the peer metadata reload error, so a
target that was already persisted and live on this node reported a 5xx to the
client whenever one peer could not be reached. Every S3 bucket-config write
path treats that reload as best effort and only warns; these two admin
handlers now do the same, and the reason is logged with the bucket and action.
* fix(site-replication): undo every bucket a cut-short refresh rewrote
When a remove accepted on another node clears the refresh journal mid-pass,
only the bucket holding the lock at that moment had its restored target
undone. The buckets rewritten earlier in the same pass kept a target pointing
at the removed peer whenever the remove's own cleanup had already walked past
them. The undo now covers every bucket this pass rewrote, attempting all of
them so one failure does not strand the rest.
* fix(site-replication): keep replay running while an endpoint refresh is pending
A pending endpoint refresh took the whole heavyweight pass with it, so a peer
that never came back froze IAM and bucket replay to every healthy peer too -
the stall this journal's resume path was meant to end. The refresh arm now
drains the retry queue before returning; it replays per-peer deliveries
against the endpoints currently committed in state, so it is unaffected by the
edit in flight. Bucket wiring reconciliation still waits, because it rewrites
the very targets the refresh is changing, and the runbook now says so.
* test(e2e): cover the AND semantics of a two-tag replication filter
The acceptance matrix only had a single-tag rule, which matches under both AND
and OR semantics and therefore proved nothing about the filter this fix
changed. It now also carries a two-tag `And` rule - the shape
`mc replicate add --tags "k1=v1&k2=v2"` writes - and asserts that an object
with one of the two tags is not admitted while an object with both is.
No new test function, so the nightly selection digest is unchanged.
* refactor(site-replication): fold the refresh state-change error into one constructor
The endpoint-refresh work added three `s3_error!` invocation lines, which the
s3s footprint ratchet is meant to prevent. Five copies of the same
concurrent-change error now share one constructor, so the surface nets one
line smaller than main; the baseline is retightened to match.
* fix(site-replication): report a peer whose IAM snapshot waits for a repair
An escalated snapshot entry records a deletion a snapshot cannot replay, so
only a repair settles it and the marker must survive. Scheduling an import
snapshot therefore leaves that peer's entry alone - and now says so, instead
of returning success while nothing was scheduled for it.
* docs(operations): state the group-status and escalation convergence limits
Two boundaries the fixes in this branch make load-bearing: a membership change
never carries an enable, so a group disabled on one site only has to be
re-enabled there explicitly; and a peer holding an escalated IAM entry does
not receive a scheduled snapshot, including the one a bulk import schedules,
until a repair settles it.
* fix(ci): bind performance runs to selected inputs (#7512)
* fix(targets): reject trailing batch items (#7508)
* fix(scanner): respect cargo target dir in G09 runner
Write the RustFS feature stamp under the resolved Cargo target directory so remote validation hosts with CARGO_TARGET_DIR set can reuse the built PR-head binary.
Extend the runner plan/self-test path to cover relative target-dir resolution.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
---------
Co-authored-by: 唐小鸭 <tangtang1251@qq.com>
Co-authored-by: Zhengchao An <anzhengchao@gmail.com>
Co-authored-by: cui fliter <imcusg@gmail.com>
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
* fix(replication): close the GA blocker set from backlog#2366 (#7503)
* fix(replication): close GA blockers from backlog#2366
Implements the P1 set from the pre-GA replication audit:
- Replication rule tag filters now require every And.Tag to match, replacing
the s3s OR semantics with a local AND matcher that fails closed on a
malformed tag.
- A replicated group membership change no longer writes the group status, so
a membership update carrying the default Enabled status cannot silently
re-enable a disabled group on the peer.
- A successful IAM import schedules one collapsed full-IAM snapshot per remote
peer instead of leaving the imported entities local-only.
- A pending endpoint refresh is redriven by the heavyweight reconcile tick,
carries its own ilm-expiry override, and no longer blocks a remove that
drops every unacknowledged peer.
- Site metrics expose local replication failure totals and rolling windows;
node-level counters no longer report a constructed zero.
- set/remove-remote-target notify peer metadata caches before returning, so a
follow-up put-bucket-replication on another node sees the target.
- Adds the site-replication operations runbook, a docs index, a replication
support boundary section, and the Replication changelog section.
* fix(site-replication): resume only a locally driven endpoint refresh
The peer-side edit handler journals a pending endpoint refresh with an empty
`remote_peers` map and commits it inside the same request through
`apply_internal_peer_edit`. The reconcile tick could not tell that journal
from the coordinator's own: with no required peers it reads as complete on
sight, so the tick committed it with `edit_state` - losing the local-name
sync - and cleared it under the request that owned it, whose commit then
reported the refresh as changed and denied the coordinator the peer
acknowledgement it was waiting for.
Resume now runs only for a journal that carries the fan-out topology. A
receiver's journal stays for the coordinator to redrive with the same
refresh id, which is the path that already recovers it.
* fix(site-replication): keep an explicit disabled group status on a snapshot
Skipping the group-status write whenever an item carries members stopped a
membership change from re-enabling a disabled group, but it also silenced the
full-IAM snapshot, which always sends members together with the sender's real
status. A peer that did not have the group yet created it through
`GroupInfo::new` - enabled - so a bootstrap, a repair, or the snapshot an IAM
import now schedules handed every member of a frozen group live access there.
The madmin wire maps an unset `groupStatus` to Enabled, so only Enabled can be
a default. Disabled is always explicit and is applied again.
* fix(site-replication): schedule the import snapshot without recording a failure
`import-iam` reused the failure-recording path to queue its full-IAM
snapshot. That raises `retry_count` on every call, so three imports - the
normal shape of a bulk migration done one archive at a time - escalated a
healthy peer to `retryStats.failed` with the scheduling note shown as
`lastError`, which is exactly the signal the runbook tells operators to
repair. A full retry queue also turned a completed import into a 503.
Scheduling now only ensures the collapsed entry exists, and a failure to
schedule is logged instead of failing the request: the entities are already
imported and the reconcile pass still closes the gap.
* fix(admin): stop reporting replication failures as retries
`retries` is the minio-go counter for redeliveries, and mc prints it as such.
Filling it with the failure count claimed a redelivery that never happens: a
failed object is not retried by an event today, it waits for the scanner heal
pass. `errors` keeps the failure counters; `retries` stays zero until there is
a real redelivery to count, and the runbook now says so.
* perf(site-replication): aggregate failure windows without cloning bucket stats
`site_metrics_snapshot` went through `get_all`, which clones every bucket's
stats, and then scanned each target's sample deque twice. That deque is
bounded only by the one-hour window, so an unreachable target under load -
the case an operator polls this endpoint for - made every
`mc admin replicate status` copy the whole backlog and hold the read lock
against the failure path while doing it.
It now folds under the read lock and takes both windows in one walk. The
`max` against the serialized `last_minute` / `last_hour` snapshots is dropped:
those are stamped onto per-bucket clones elsewhere and are always zero in this
node-local cache.
* fix(site-replication): reject a conflicting ilm-expiry override on a re-run
The commit now reads the ilm-expiry override back out of the pending refresh
journal, so a second edit that asks for a different value had it dropped while
the request still reported success. Re-running without the flag keeps pinning
the recorded value - that is the documented way to redrive a stuck refresh -
but an explicit different value is now rejected instead of ignored.
* fix(admin): do not fail a remote-target write on a peer reload error
set/remove-remote-target propagated the peer metadata reload error, so a
target that was already persisted and live on this node reported a 5xx to the
client whenever one peer could not be reached. Every S3 bucket-config write
path treats that reload as best effort and only warns; these two admin
handlers now do the same, and the reason is logged with the bucket and action.
* fix(site-replication): undo every bucket a cut-short refresh rewrote
When a remove accepted on another node clears the refresh journal mid-pass,
only the bucket holding the lock at that moment had its restored target
undone. The buckets rewritten earlier in the same pass kept a target pointing
at the removed peer whenever the remove's own cleanup had already walked past
them. The undo now covers every bucket this pass rewrote, attempting all of
them so one failure does not strand the rest.
* fix(site-replication): keep replay running while an endpoint refresh is pending
A pending endpoint refresh took the whole heavyweight pass with it, so a peer
that never came back froze IAM and bucket replay to every healthy peer too -
the stall this journal's resume path was meant to end. The refresh arm now
drains the retry queue before returning; it replays per-peer deliveries
against the endpoints currently committed in state, so it is unaffected by the
edit in flight. Bucket wiring reconciliation still waits, because it rewrites
the very targets the refresh is changing, and the runbook now says so.
* test(e2e): cover the AND semantics of a two-tag replication filter
The acceptance matrix only had a single-tag rule, which matches under both AND
and OR semantics and therefore proved nothing about the filter this fix
changed. It now also carries a two-tag `And` rule - the shape
`mc replicate add --tags "k1=v1&k2=v2"` writes - and asserts that an object
with one of the two tags is not admitted while an object with both is.
No new test function, so the nightly selection digest is unchanged.
* refactor(site-replication): fold the refresh state-change error into one constructor
The endpoint-refresh work added three `s3_error!` invocation lines, which the
s3s footprint ratchet is meant to prevent. Five copies of the same
concurrent-change error now share one constructor, so the surface nets one
line smaller than main; the baseline is retightened to match.
* fix(site-replication): report a peer whose IAM snapshot waits for a repair
An escalated snapshot entry records a deletion a snapshot cannot replay, so
only a repair settles it and the marker must survive. Scheduling an import
snapshot therefore leaves that peer's entry alone - and now says so, instead
of returning success while nothing was scheduled for it.
* docs(operations): state the group-status and escalation convergence limits
Two boundaries the fixes in this branch make load-bearing: a membership change
never carries an enable, so a group disabled on one site only has to be
re-enabled there explicitly; and a peer holding an escalated IAM entry does
not receive a scheduled snapshot, including the one a bulk import schedules,
until a repair settles it.
* fix(ci): bind performance runs to selected inputs (#7512)
* fix(targets): reject trailing batch items (#7508)
* test(scanner): preflight G09 evidence disk space
Fail the Scanner/Heal G09 upgrade evidence runner before downloading or building when the validation host does not have enough free space for a full raw evidence pass.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
---------
Co-authored-by: 唐小鸭 <tangtang1251@qq.com>
Co-authored-by: Zhengchao An <anzhengchao@gmail.com>
Co-authored-by: cui fliter <imcusg@gmail.com>
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
* fix(replication): close the GA blocker set from backlog#2366 (#7503)
* fix(replication): close GA blockers from backlog#2366
Implements the P1 set from the pre-GA replication audit:
- Replication rule tag filters now require every And.Tag to match, replacing
the s3s OR semantics with a local AND matcher that fails closed on a
malformed tag.
- A replicated group membership change no longer writes the group status, so
a membership update carrying the default Enabled status cannot silently
re-enable a disabled group on the peer.
- A successful IAM import schedules one collapsed full-IAM snapshot per remote
peer instead of leaving the imported entities local-only.
- A pending endpoint refresh is redriven by the heavyweight reconcile tick,
carries its own ilm-expiry override, and no longer blocks a remove that
drops every unacknowledged peer.
- Site metrics expose local replication failure totals and rolling windows;
node-level counters no longer report a constructed zero.
- set/remove-remote-target notify peer metadata caches before returning, so a
follow-up put-bucket-replication on another node sees the target.
- Adds the site-replication operations runbook, a docs index, a replication
support boundary section, and the Replication changelog section.
* fix(site-replication): resume only a locally driven endpoint refresh
The peer-side edit handler journals a pending endpoint refresh with an empty
`remote_peers` map and commits it inside the same request through
`apply_internal_peer_edit`. The reconcile tick could not tell that journal
from the coordinator's own: with no required peers it reads as complete on
sight, so the tick committed it with `edit_state` - losing the local-name
sync - and cleared it under the request that owned it, whose commit then
reported the refresh as changed and denied the coordinator the peer
acknowledgement it was waiting for.
Resume now runs only for a journal that carries the fan-out topology. A
receiver's journal stays for the coordinator to redrive with the same
refresh id, which is the path that already recovers it.
* fix(site-replication): keep an explicit disabled group status on a snapshot
Skipping the group-status write whenever an item carries members stopped a
membership change from re-enabling a disabled group, but it also silenced the
full-IAM snapshot, which always sends members together with the sender's real
status. A peer that did not have the group yet created it through
`GroupInfo::new` - enabled - so a bootstrap, a repair, or the snapshot an IAM
import now schedules handed every member of a frozen group live access there.
The madmin wire maps an unset `groupStatus` to Enabled, so only Enabled can be
a default. Disabled is always explicit and is applied again.
* fix(site-replication): schedule the import snapshot without recording a failure
`import-iam` reused the failure-recording path to queue its full-IAM
snapshot. That raises `retry_count` on every call, so three imports - the
normal shape of a bulk migration done one archive at a time - escalated a
healthy peer to `retryStats.failed` with the scheduling note shown as
`lastError`, which is exactly the signal the runbook tells operators to
repair. A full retry queue also turned a completed import into a 503.
Scheduling now only ensures the collapsed entry exists, and a failure to
schedule is logged instead of failing the request: the entities are already
imported and the reconcile pass still closes the gap.
* fix(admin): stop reporting replication failures as retries
`retries` is the minio-go counter for redeliveries, and mc prints it as such.
Filling it with the failure count claimed a redelivery that never happens: a
failed object is not retried by an event today, it waits for the scanner heal
pass. `errors` keeps the failure counters; `retries` stays zero until there is
a real redelivery to count, and the runbook now says so.
* perf(site-replication): aggregate failure windows without cloning bucket stats
`site_metrics_snapshot` went through `get_all`, which clones every bucket's
stats, and then scanned each target's sample deque twice. That deque is
bounded only by the one-hour window, so an unreachable target under load -
the case an operator polls this endpoint for - made every
`mc admin replicate status` copy the whole backlog and hold the read lock
against the failure path while doing it.
It now folds under the read lock and takes both windows in one walk. The
`max` against the serialized `last_minute` / `last_hour` snapshots is dropped:
those are stamped onto per-bucket clones elsewhere and are always zero in this
node-local cache.
* fix(site-replication): reject a conflicting ilm-expiry override on a re-run
The commit now reads the ilm-expiry override back out of the pending refresh
journal, so a second edit that asks for a different value had it dropped while
the request still reported success. Re-running without the flag keeps pinning
the recorded value - that is the documented way to redrive a stuck refresh -
but an explicit different value is now rejected instead of ignored.
* fix(admin): do not fail a remote-target write on a peer reload error
set/remove-remote-target propagated the peer metadata reload error, so a
target that was already persisted and live on this node reported a 5xx to the
client whenever one peer could not be reached. Every S3 bucket-config write
path treats that reload as best effort and only warns; these two admin
handlers now do the same, and the reason is logged with the bucket and action.
* fix(site-replication): undo every bucket a cut-short refresh rewrote
When a remove accepted on another node clears the refresh journal mid-pass,
only the bucket holding the lock at that moment had its restored target
undone. The buckets rewritten earlier in the same pass kept a target pointing
at the removed peer whenever the remove's own cleanup had already walked past
them. The undo now covers every bucket this pass rewrote, attempting all of
them so one failure does not strand the rest.
* fix(site-replication): keep replay running while an endpoint refresh is pending
A pending endpoint refresh took the whole heavyweight pass with it, so a peer
that never came back froze IAM and bucket replay to every healthy peer too -
the stall this journal's resume path was meant to end. The refresh arm now
drains the retry queue before returning; it replays per-peer deliveries
against the endpoints currently committed in state, so it is unaffected by the
edit in flight. Bucket wiring reconciliation still waits, because it rewrites
the very targets the refresh is changing, and the runbook now says so.
* test(e2e): cover the AND semantics of a two-tag replication filter
The acceptance matrix only had a single-tag rule, which matches under both AND
and OR semantics and therefore proved nothing about the filter this fix
changed. It now also carries a two-tag `And` rule - the shape
`mc replicate add --tags "k1=v1&k2=v2"` writes - and asserts that an object
with one of the two tags is not admitted while an object with both is.
No new test function, so the nightly selection digest is unchanged.
* refactor(site-replication): fold the refresh state-change error into one constructor
The endpoint-refresh work added three `s3_error!` invocation lines, which the
s3s footprint ratchet is meant to prevent. Five copies of the same
concurrent-change error now share one constructor, so the surface nets one
line smaller than main; the baseline is retightened to match.
* fix(site-replication): report a peer whose IAM snapshot waits for a repair
An escalated snapshot entry records a deletion a snapshot cannot replay, so
only a repair settles it and the marker must survive. Scheduling an import
snapshot therefore leaves that peer's entry alone - and now says so, instead
of returning success while nothing was scheduled for it.
* docs(operations): state the group-status and escalation convergence limits
Two boundaries the fixes in this branch make load-bearing: a membership change
never carries an enable, so a group disabled on one site only has to be
re-enabled there explicitly; and a peer holding an escalated IAM entry does
not receive a scheduled snapshot, including the one a bulk import schedules,
until a repair settles it.
* fix(ci): bind performance runs to selected inputs (#7512)
* test(scanner): add G09 upgrade evidence runner
Add a reusable Scanner/Heal G09 runner for Linux PR-head validation. The script downloads the pinned previous release, builds the current checkout, runs the mixed-version and rollback upgrade E2E lanes, and validates the measured raw artifacts before they can be consumed by the release bundle gate.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
---------
Co-authored-by: 唐小鸭 <tangtang1251@qq.com>
Co-authored-by: Zhengchao An <anzhengchao@gmail.com>
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
* fix(replication): close the GA blocker set from backlog#2366 (#7503)
* fix(replication): close GA blockers from backlog#2366
Implements the P1 set from the pre-GA replication audit:
- Replication rule tag filters now require every And.Tag to match, replacing
the s3s OR semantics with a local AND matcher that fails closed on a
malformed tag.
- A replicated group membership change no longer writes the group status, so
a membership update carrying the default Enabled status cannot silently
re-enable a disabled group on the peer.
- A successful IAM import schedules one collapsed full-IAM snapshot per remote
peer instead of leaving the imported entities local-only.
- A pending endpoint refresh is redriven by the heavyweight reconcile tick,
carries its own ilm-expiry override, and no longer blocks a remove that
drops every unacknowledged peer.
- Site metrics expose local replication failure totals and rolling windows;
node-level counters no longer report a constructed zero.
- set/remove-remote-target notify peer metadata caches before returning, so a
follow-up put-bucket-replication on another node sees the target.
- Adds the site-replication operations runbook, a docs index, a replication
support boundary section, and the Replication changelog section.
* fix(site-replication): resume only a locally driven endpoint refresh
The peer-side edit handler journals a pending endpoint refresh with an empty
`remote_peers` map and commits it inside the same request through
`apply_internal_peer_edit`. The reconcile tick could not tell that journal
from the coordinator's own: with no required peers it reads as complete on
sight, so the tick committed it with `edit_state` - losing the local-name
sync - and cleared it under the request that owned it, whose commit then
reported the refresh as changed and denied the coordinator the peer
acknowledgement it was waiting for.
Resume now runs only for a journal that carries the fan-out topology. A
receiver's journal stays for the coordinator to redrive with the same
refresh id, which is the path that already recovers it.
* fix(site-replication): keep an explicit disabled group status on a snapshot
Skipping the group-status write whenever an item carries members stopped a
membership change from re-enabling a disabled group, but it also silenced the
full-IAM snapshot, which always sends members together with the sender's real
status. A peer that did not have the group yet created it through
`GroupInfo::new` - enabled - so a bootstrap, a repair, or the snapshot an IAM
import now schedules handed every member of a frozen group live access there.
The madmin wire maps an unset `groupStatus` to Enabled, so only Enabled can be
a default. Disabled is always explicit and is applied again.
* fix(site-replication): schedule the import snapshot without recording a failure
`import-iam` reused the failure-recording path to queue its full-IAM
snapshot. That raises `retry_count` on every call, so three imports - the
normal shape of a bulk migration done one archive at a time - escalated a
healthy peer to `retryStats.failed` with the scheduling note shown as
`lastError`, which is exactly the signal the runbook tells operators to
repair. A full retry queue also turned a completed import into a 503.
Scheduling now only ensures the collapsed entry exists, and a failure to
schedule is logged instead of failing the request: the entities are already
imported and the reconcile pass still closes the gap.
* fix(admin): stop reporting replication failures as retries
`retries` is the minio-go counter for redeliveries, and mc prints it as such.
Filling it with the failure count claimed a redelivery that never happens: a
failed object is not retried by an event today, it waits for the scanner heal
pass. `errors` keeps the failure counters; `retries` stays zero until there is
a real redelivery to count, and the runbook now says so.
* perf(site-replication): aggregate failure windows without cloning bucket stats
`site_metrics_snapshot` went through `get_all`, which clones every bucket's
stats, and then scanned each target's sample deque twice. That deque is
bounded only by the one-hour window, so an unreachable target under load -
the case an operator polls this endpoint for - made every
`mc admin replicate status` copy the whole backlog and hold the read lock
against the failure path while doing it.
It now folds under the read lock and takes both windows in one walk. The
`max` against the serialized `last_minute` / `last_hour` snapshots is dropped:
those are stamped onto per-bucket clones elsewhere and are always zero in this
node-local cache.
* fix(site-replication): reject a conflicting ilm-expiry override on a re-run
The commit now reads the ilm-expiry override back out of the pending refresh
journal, so a second edit that asks for a different value had it dropped while
the request still reported success. Re-running without the flag keeps pinning
the recorded value - that is the documented way to redrive a stuck refresh -
but an explicit different value is now rejected instead of ignored.
* fix(admin): do not fail a remote-target write on a peer reload error
set/remove-remote-target propagated the peer metadata reload error, so a
target that was already persisted and live on this node reported a 5xx to the
client whenever one peer could not be reached. Every S3 bucket-config write
path treats that reload as best effort and only warns; these two admin
handlers now do the same, and the reason is logged with the bucket and action.
* fix(site-replication): undo every bucket a cut-short refresh rewrote
When a remove accepted on another node clears the refresh journal mid-pass,
only the bucket holding the lock at that moment had its restored target
undone. The buckets rewritten earlier in the same pass kept a target pointing
at the removed peer whenever the remove's own cleanup had already walked past
them. The undo now covers every bucket this pass rewrote, attempting all of
them so one failure does not strand the rest.
* fix(site-replication): keep replay running while an endpoint refresh is pending
A pending endpoint refresh took the whole heavyweight pass with it, so a peer
that never came back froze IAM and bucket replay to every healthy peer too -
the stall this journal's resume path was meant to end. The refresh arm now
drains the retry queue before returning; it replays per-peer deliveries
against the endpoints currently committed in state, so it is unaffected by the
edit in flight. Bucket wiring reconciliation still waits, because it rewrites
the very targets the refresh is changing, and the runbook now says so.
* test(e2e): cover the AND semantics of a two-tag replication filter
The acceptance matrix only had a single-tag rule, which matches under both AND
and OR semantics and therefore proved nothing about the filter this fix
changed. It now also carries a two-tag `And` rule - the shape
`mc replicate add --tags "k1=v1&k2=v2"` writes - and asserts that an object
with one of the two tags is not admitted while an object with both is.
No new test function, so the nightly selection digest is unchanged.
* refactor(site-replication): fold the refresh state-change error into one constructor
The endpoint-refresh work added three `s3_error!` invocation lines, which the
s3s footprint ratchet is meant to prevent. Five copies of the same
concurrent-change error now share one constructor, so the surface nets one
line smaller than main; the baseline is retightened to match.
* fix(site-replication): report a peer whose IAM snapshot waits for a repair
An escalated snapshot entry records a deletion a snapshot cannot replay, so
only a repair settles it and the marker must survive. Scheduling an import
snapshot therefore leaves that peer's entry alone - and now says so, instead
of returning success while nothing was scheduled for it.
* docs(operations): state the group-status and escalation convergence limits
Two boundaries the fixes in this branch make load-bearing: a membership change
never carries an enable, so a group disabled on one site only has to be
re-enabled there explicitly; and a peer holding an escalated IAM entry does
not receive a scheduled snapshot, including the one a bulk import schedules,
until a repair settles it.
* fix(ci): bind performance runs to selected inputs (#7512)
* test(scanner): add G09 upgrade evidence runner
Add a reusable Linux x86_64 runner for the Scanner/Heal G09 mixed-version and rollback upgrade evidence lanes.
The helper reads the pinned previous-release asset metadata from the upgrade workflow, verifies the downloaded binary, builds the current head, runs both ignored E2E tests, and fails unless the expected G09 JSON artifacts exist.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
---------
Co-authored-by: 唐小鸭 <tangtang1251@qq.com>
Co-authored-by: Zhengchao An <anzhengchao@gmail.com>
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Record measured Scanner/Heal G09 evidence artifacts from the upgrade compatibility lanes when a fresh evidence directory is provided.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
* fix(sse): resolve bucket default encryption per request
PUT and the POST-object/extract path resolved a bucket's default
encryption with a hard-coded "no explicit SSE-C" flag, so the default was
layered onto a request that already carried an SSE-C header triple and
then tripped that request's own mutual-exclusion check. Every bucket with
default encryption refused SSE-C single PUTs with 400 InvalidArgument,
while CreateMultipartUpload on the same bucket succeeded because it
resolves SSE elsewhere. Both call sites now derive the flag from the
request headers, as COPY already did.
The bucket default's KMS key id was also inherited independently of the
effective algorithm, so an explicit AES256 request against an aws:kms
default bucket produced a self-contradictory algorithm/key-id pair and
was rejected. The key id is now inherited only when the effective
algorithm is aws:kms, matching the storage-layer resolver.
Refs backlog#2368 B1, B2.
* fix(sse): refuse SSE-KMS without a running KMS service
A write requesting aws:kms on a node with no KMS service fell back to the
node-local SSE-S3 provider: the data key was wrapped with
RUSTFS_SSE_S3_MASTER_KEY while the object metadata still recorded
aws:kms and the requested KMS key id. The stored object claimed a KMS
protection it never had, under a key that was never consulted, and no
signal distinguished it from a genuine SSE-KMS object.
The managed-encryption path now asks the resolved DEK provider whether it
wraps with a node-local master key and refuses SSE-KMS in that case:
InvalidRequest when KMS was never configured, ServiceUnavailable when a
configured service is not running. The check sits after the per-key
authorization gate so an unauthorized caller still receives AccessDenied
whatever the KMS runtime state is, and asks the provider rather than a
parallel availability signal because the provider is what actually wraps
the key. A missing master key no longer answers an SSE-KMS request with
an SSE-S3-worded configuration error.
The SSE-S3 local fallback is unchanged.
Refs backlog#2368 B4.
* fix(ecstore): restore and archive tiers in stored coordinates
Multipart restore addressed the remote tier in plaintext coordinates
while the copy-back reads the stored representation. Each part received a
misaligned slice of the remote object whose length still satisfied the
range, the hash reader and the completion size check, so the restore
reported success and silently replaced the object's bytes. Encrypted and
compressed multipart objects were both affected. Restore now accumulates
stored part sizes, passes the stored length to the hash reader alongside
the plaintext length, and validates against the stored size.
The copy-back digests stored bytes, so its computed MD5 is not the
object's public ETag. Restore now preserves the object ETag on both the
single-part and multipart paths, and gives each restored part its own
recorded part ETag rather than the object-level value.
Transition also handed the tier the object's SSE headers and its
RustFS-wrapped data key as request headers. Any S3 target rejected an
SSE-C archive outright, an SSE-KMS archive asked the target to encrypt a
second time under a key id it does not own, and the wrapped DEK left the
cluster. The archive request now strips every SSE header and encryption
marker with the predicate the replication path already uses; the local
xl.meta keeps all of it, so read-through and restore are unaffected.
Objects restored by an affected release are not detected or repaired
retroactively and must be re-restored from the tier.
Refs backlog#2368 B3, B5; backlog#2369 P7.1.
* fix(rio): lock the v1 nonce layout within a segment
Decrypting a v1 segment tried three historical nonce layouts per frame,
independently for every frame. The last of them exists for streams
written before 1.0.0-alpha.91, which reused a segment's part nonce for
every block in it; because block zero's derived nonce equals that base
nonce, a frame encrypted at index zero authenticated at any position. An
attacker able to rewrite the underlying shards could replay it and have
the forged plaintext returned with 200 and an unchanged length. Shard
integrity uses a keyed-hash-free checksum, which such an attacker can
recompute, so it is not a barrier.
A segment now locks onto whichever layout decoded its first non-zero-index
frame and rejects any later frame needing a different one. That leaves one
residual shape: a stream built purely from repeats of frame zero has no
later frame to disagree. New RUSTFS_ENCRYPTION_LEGACY_NONCE_FALLBACK
(default true, so pre-alpha.91 objects keep decrypting) drops the third
layout entirely when set to false, which closes it. Turning it off refuses
pre-alpha.91 objects, so migrate them first by rewriting in place.
Refs backlog#2369 P2.
* fix(kms): reload a service that failed to start
POST /rustfs/admin/v3/kms/reload short-circuited whenever the persisted
configuration matched the in-memory one byte for byte. A node whose KMS
failed to start keeps that configuration and sits in Error, so the
documented recovery call returned "reloaded successfully" while leaving
the node down. Peers reached the same path through the reload broadcast,
so a cluster that lost Vault during a rolling restart had no working
recovery route other than the node-local start endpoint. Reload now
short-circuits only for a service that is actually running, and otherwise
reconfigures, which starts a service that is not running.
The AWS backend also advertised key-version enumeration through
kms/status, which its own documentation says it cannot do; the capability
and its golden snapshot now say false.
Refs backlog#2369 P1, P7.3.
* docs: record the SSE and KMS changes for 1.0.0
The Unreleased changelog section carried no entry for any encryption work
merged since 1.0.0-rc.5, including three items with operational impact:
the config-secret variable whose absence persists secrets in cleartext
with only a warning, the v2 frame write switch and its rolling-upgrade
constraint, and per-key authorization making a public bucket incompatible
with SSE-KMS objects. Adds those plus this batch, including the SSE-KMS
refusal as a breaking change with both routes out.
Also corrects four places where documentation contradicted the code: the
cleanup register still called encrypted range seek opt-in after its
default flipped, the Helm README claimed vault_mount_path only applies to
Transit while the template also feeds the KV2 mount, the disaster-recovery
drill listed bundle contents for backends whose export is refused with
501, and the Chinese README capability table predated most of the feature
set. Documents the SSE-S3 local master key as a first-class operational
mode with its rotation dead end, and what the v1 frame layout does and
does not authenticate.
Refs backlog#2369 P5.
* fix(kms): classify data-path KMS failures by what the caller can do
Only "key not found" and a backend outage were classified; every other
KMS failure that reached the S3 data path fell through to
500 InternalError with a generic message. A disabled or pending-deletion
key, a denied KMS grant, an encryption-context mismatch, an unsupported
algorithm, a credential or timeout failure, and a capability the
configured backend does not have all looked identical to a server fault.
SDKs therefore applied exponential backoff to configuration errors no
retry can fix, and monitoring counted every one of them against the
server's own error rate.
Unusable-key and request-side failures now answer 400, a denied grant
403, transient backend failures 503, and a missing backend capability
501. Damaged, unreadable, or unknown-format key material keeps its 500:
it is a server-side integrity fault, and existing tests pin it.
The classifier is deliberately separate from the admin lifecycle
mapping, which answers 404 for a missing key because there a key id is
the resource being addressed; on the data path it arrives inside a
request header or a bucket default. Messages either name what the caller
asked for or stay generic, with deployment-side detail left on the error
source the way the storage-IO mapping already does.
Refs backlog#2368 B6.
* fix(kms): track and renew static Vault tokens
Token authentication hard-coded "this token carries no lease", so the
renewal task never started, the remaining-TTL gauge was never published,
and nothing looked wrong. `vault token create` grants a 768-hour TTL by
default, so a cluster that had been healthy for a month turned every KMS
call into a 403 and could not recover without a restart or a
reconfigure. Production configuration validation only rejects the
literal dev-token, so an ordinary expiring token reaches a whole cluster.
The source now reads `auth/token/lookup-self` at login and adopts what
Vault reports. A token with no expiry behaves exactly as before. An
expiring renewable one is picked up by the existing renewal loop and
renewed at half TTL like every other auth method. An expiring
non-renewable one warns with its remaining lifetime and publishes the
gauge, so the fail-closed window is visible before it arrives.
The probe never fails the login: a policy that omits lookup-self, or a
Vault that is briefly unreachable, warns and falls back to exactly the
previous behaviour rather than taking down a deployment that works
today. The scripted Vault test double answers the lookup out of band so
existing scripts keep describing only the protocol under test.
Refs backlog#2369 P3.
* feat(sse): report SSE-C requests that arrive without TLS
An SSE-C request carries the customer's AES key in a request header, so
AWS S3 and MinIO both refuse one that did not arrive over TLS. RustFS
accepted them on any transport: a plaintext hop hands the key to anyone
on the path, and since the object cannot be read without that same key,
the exposure lasts as long as the object does.
Refusing outright is the correct end state but not a safe default to
adopt inside a release window, because the project's own s3-tests and
e2e lanes and most staging deployments speak plain HTTP. This release
reports instead: each such request increments
rustfs_ssec_plaintext_requests_total and logs one warning per process, so
an operator can confirm nothing would break before the default flips.
RUSTFS_SSE_C_REQUIRE_TLS=true opts into the AWS 400 now.
The verdict is per connection rather than per deployment: the layer is
built with whether this listener terminated TLS, and additionally accepts
an https protocol forwarded by a proxy the trusted-proxy configuration
already vetted. It sits beside the rate limiter, after the layer that
makes a forwarded protocol trustworthy and after the request context, so
a rejection can echo the request id.
Refs backlog#2369 P7.2.
* fix(kms): say what a node-local backend means for a cluster
The Local backend keeps key material on each node's own disk and
generates its Argon2id salt per node, so two nodes derive different keys
from the same master_key and an object encrypted on one node cannot be
decrypted on another. Behind a load balancer that surfaces as
intermittent 500s on reads that succeeded moments earlier, with nothing
tying the symptom to the cause: the only signal was a generic
"development, testing and demos only" positioning warning that says
nothing about what actually breaks.
Configuring or reconfiguring Local while the deployment is distributed
now logs a dedicated event and appends the consequence to the configure
response, so the operator who made the change sees it. The product
decision to warn rather than refuse is unchanged.
Refs backlog#2369 P7.4.
* docs: record the remaining SSE and KMS changes for 1.0.0
Adds changelog entries for the KMS data-path status classification, the
Vault static-token lease probe, the SSE-C plaintext-transport report and
its switch, and the node-local backend warning.
Documents two things the backend security guide never stated: that SSE-C
belongs on a secure transport, with the counter and switch to plan the
change around, and that the Local backend cannot be shared by a
multi-node deployment because each node derives different keys from the
same master key.
Refs backlog#2368 B6; backlog#2369 P3, P5, P7.2, P7.4.
* fix(kms): report an unreadable key store as an outage on the S3 path
A backend now distinguishes a key store it could not read from a key
that is genuinely absent, but the S3 boundary collapsed the first one
back onto 500 InternalError through the fallthrough for integrity
faults. The distinction was therefore invisible to the client: a
temporary key-directory outage looked exactly like a permanently damaged
key record, and neither the status nor the metric said the request was
worth retrying.
An unreadable key store joins the retryable class and answers 503, next
to a backend error and a credential failure. Damaged, unreadable or
unknown-format key material keeps its 500.
Refs backlog#2368 B6; builds on rustfs/rustfs#7470.
Merge main at 73957d0faf into release. Resolve the multipart test conflict with the current main fixture and remove its superseded ReadPlan test export.
Surface the scanned port window, attempt count, and last bind error when the e2e port allocator cannot reserve a localhost port. This keeps Scanner/Heal evidence failures actionable when the environment blocks binds before business assertions run.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Require the P4 release bundle to carry measured cleanup/GC soak evidence for retained MRF replay responsibilities. The bundle now needs a two-hour cleanup window, exact cleanup case coverage, observed verified idle GC, and zero pending responsibilities or stale journals after GC.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Require P4 retained-responsibility release bundle evidence to list the retained replay anchor and idle cleanup cases, prove a two-hour retention window, and record both idle cleanup and verified-proof discharge observations.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Add a regression oracle that keeps scoped-only MRF responsibilities in the authoritative runtime snapshot while omitting them from the v1 legacy rollback mirror.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Require scanner/heal release evidence JSON artifacts to repeat their measured source revision, run identity, measurement window, gate, and field identity inside the artifact payload. This keeps a refreshed outer bundle hash from accepting stale summary or profile JSON from another run.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Require the Scanner/Heal release registry and bundle checker to carry explicit G08 MRF capacity, disk-full, and replica-loss evidence fields before release approval can pass.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Align e2e_test build provenance with Scanner/Heal evidence receipts and server binary provenance by treating only tracked source changes as dirty.
This prevents unrelated untracked worktrees or evidence directories from causing compiled test identity mismatches before real evidence cases can run.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Add a runtime cleanup regression test that publishes both retained replay and runtime committed checkpoints, writes scoped and legacy journals, and verifies idle cleanup removes every recovery anchor from the registered local disks.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Tighten the Scanner/Heal release bundle checker so G14 same-window evidence must name the EC8+4, multi-set, and multi-pool fields covered in that measurement window.
Keep the release gate blocked when same-window evidence omits one of the required G14 fields, without changing production runtime behavior.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Require Scanner/Heal release bundle artifact paths to resolve to non-empty files before hashing them.
Cover empty hard-gate artifacts in the existing release bundle checker self-test and keep profile artifact size checking on the shared artifact boundary.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Give Scanner/Heal evidence cases explicit runtime profiles so EC8+4 background restart and crash cases use their own object count, object size, and partial-progress timeout defaults instead of inheriting the legacy 4x1 case assumptions.
Expose the runtime profile in plan-only output and cover every registry case in the script self-test.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Group the EC8+4 Scanner/Heal evidence writer inputs into a typed payload so the distributed e2e crate stays within the clippy argument limit without weakening the lint.
The evidence writer still validates the same S3 bodies, physical shard census, process restart PIDs, and node listings.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Reject measured Scanner/Heal ABBA manifests whose mixed-version evidence uses the same baseline and candidate source revision or binary hash.
Require crash fault modes and profile artifact names to match the exact supported sets, rejecting missing, duplicate, and unknown values.
Update harness fixtures and regression coverage for same-build mixed-version claims and exact-set release evidence fields.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Tighten the Scanner/Heal release bundle checker so mixed-version, scoped-ACK, and rollback fields cannot reuse a generic versions list without proving the expected evidence role.
Require version lists to use source revision identities and include the tested source revision. Also require profile evidence fields to name the core profiling artifacts before release approval.
The release gate remains blocked until measured field evidence is present.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Cover the committed MRF manifest path through LocalDisk conditional CAS when the metadata directory fsync fails. The fixture proves the previous manifest anchor survives rollback, an unanchored first successor is removed, and the legacy MRF journal remains readable even while global durability is relaxed.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Reject journal records with unknown version-presence flags even when their CRC is valid, so rollback/future payloads cannot be accepted as known records.
Gate committed checkpoint cleanup by the writer owner captured from the replay source, preserving retained manifests from other owners inside the same sequence window.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Bind the distributed EC8+4 restart evidence lane to a scanner/heal oracle artifact so release validation can consume the real nextest run instead of accepting only a passing test.
Require the registry to assert 8+4 erasure geometry for the three-node, four-drive case.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Keep the EC8+4 background restart lane on the graceful-stop path and assert clean-restart marker absence only for restart scenarios. This prevents the hard evidence gate from silently exercising the crash path when it claims restart coverage.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Keep same-request-id replay receipts accepted for the receipt API, but preserve the legacy submit_heal_request duplicate admission result as Merged.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
Add focused oracles for transient quorum results that carry a matching receipt and for mixed grace plus repaired receipt batches. Only the repaired object may produce positive proof.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
Register and wire Scanner/Heal background target restart and crash evidence cases for the 3x4 EC8+4 topology.
Validate the observed data/parity geometry in scanner-heal evidence receipts so multi-drive runs cannot satisfy the gate without proving EC8+4 metadata.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Cover bucket and root heal sweeps recording authoritative object outcomes only when storage receipts match the latched bucket incarnation.
Verify unavailable or stale receipt ownership keeps object repair execution intact while leaving canonical outcome proof as Unknown.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Cover the receipt consumer path where a storage repair result carries both an error and a matching positive receipt. The failure may be recorded, but the receipt must not create repaired, healthy, or absent proof.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
Do not record positive storage repair receipts once an object heal task has been cancelled, even if the receipt still matches the requested owner and object identity.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
Do not record positive storage repair receipts for dry-run object heal tasks, even if a producer accidentally returns a matching receipt.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
Require Scanner/Heal release bundle fields to carry source, run, window, timestamp, command, and artifact format provenance before a measured gate can pass.
Keep EC8+4 and performance gate fields tied to a single measurement window so unrelated artifacts cannot be stitched into a release approval.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Capture the expected bucket incarnation before invoking object repair so a post-repair owner change cannot rewrite the responsibility that a storage receipt is allowed to prove.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
Require object heal receipts to match the expected bucket incarnation before they can be recorded as positive repair evidence. This prevents stale or cross-incarnation receipts from clearing the wrong heal responsibility.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
Add a committed snapshot predecessor cleanup primitive that deletes only an older slot after the successor is read back as the current committed snapshot.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
Keep startup replay journal until a durable successor snapshot exists after Accepted/Merged admission.
Co-Authored-By: heihutu <heihutu@gmail.com>
Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
Keep the synthetic ABBA test adapter aligned with the measured release evidence contract so result validation covers release_evidence drift.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Retry heal-control RPCs once after transport auth rejects a stale replay-scope epoch, and add a distributed EC8+4 restart heal evidence case that rebuilds a replaced drive with exact shard/body assertions.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Record production-facing segment invalidation producer identities when existing object-level dirty usage hooks observe PUT, CopyObject, DeleteObject/DeleteMarker, and CompleteMultipartUpload mutations. Keep the data non-authoritative and process-local so segment reuse activation still requires durable generation-window proof.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Require explicit scanner/heal release evidence field contracts for the scoped ACK and mixed-version rollback gates.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Keep a retried heal start with the same request id from bypassing admission deduplication when the transport replay cache is unavailable but the manager still owns the task.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Add a production-entry scanner cohort regression that combines a wide flat bucket with a small bucket under a fixed object budget. The test keeps partial budgeted rounds unpublished, verifies small buckets are only marked after real execution, and confirms a later unbudgeted round can publish the complete aggregate.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
Add a recorded verified-repair consumer that discharges only exact retained anchors for the requested bucket while leaving unmatched proofs in the event ring. This gives the future durable successor writer a fail-closed primitive before any tombstone or GC path is enabled.
Co-authored-by: zhi22915 <qiuzgang@gmail.com>
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))' \
--status-level all --final-status-level all \
2>&1 | tee artifacts/ilm-integration/nextest.log
status=${PIPESTATUS[0]}
@@ -997,7 +966,6 @@ jobs:
# debug binary; each test spawns its own rustfs server on a random port.
-`RUSTFS_CORS_ALLOWED_ORIGINS` defaults to empty, so the S3 endpoint emits no generic CORS headers unless configured. Set `*` for wildcard origins without credentials, or a comma-separated allow-list for credentialed explicit origins.
-`RUSTFS_CONSOLE_CORS_ALLOWED_ORIGINS` defaults to `*` for the console service.
## Console URL prefix
`RUSTFS_CONSOLE_PREFIX` changes the embedded console URL prefix. The default is
`/rustfs/console`. For example, `RUSTFS_CONSOLE_PREFIX=/console` serves the UI at
`http://localhost:9001/console/`. Nested prefixes such as `/management/console`
are supported; one trailing slash is removed. Restart the server after changing it.
The prefix must be a non-root absolute path of at most 256 bytes, with nonempty
segments containing only ASCII letters, digits, `-`, `_`, `.`, or `~`. Dot
segments, encoded characters, and overlaps with reserved admin, RPC, health,
profiling, browser entry, and icon routes are rejected at startup. `/` is not supported.
Choose a prefix that does not collide with S3 bucket paths.
The console routes, embedded frontend asset URLs, browser redirects, and OIDC
console redirects use this prefix. Admin API paths and the identity provider's
continues to control only the listening address and port. The server adapts bundled
console asset references from their build-time base path to the runtime prefix.
OEM builds can set `RUSTFS_CONSOLE_BASE_PATH` when compiling RustFS to embed a
different default, such as `/nuofans/console`. Build the bundled console with the
same `NEXT_PUBLIC_BASE_PATH`. An unset or empty build variable retains
`/rustfs/console`. The build path must satisfy the validation rules above and must
not have a trailing slash.
At startup, `RUSTFS_CONSOLE_PREFIX` takes precedence over the compiled default.
Changing `RUSTFS_CONSOLE_BASE_PATH` when starting an existing binary has no effect;
rebuild both components to change the embedded default. If a runtime prefix is
configured, asset adaptation uses the compiled base path as its source, including
when restoring `/rustfs/console` for a custom OEM build.
## Browser redirect environment variables
-`RUSTFS_BROWSER_REDIRECT_URL` sets the externally reachable browser origin used for OIDC callback, console success redirect, and logout fallback URLs. Configure it to the public scheme and authority without a path, for example `https://console.example.com`. In load-balancer deployments, keep OIDC authorize and callback requests on the same backend node because the in-flight OIDC `state` is local to the RustFS node.
@@ -38,13 +38,6 @@ All commands assume repo root. `cargo test` triggers an on-demand build of the
`rustfs` binary from [`src/common.rs`](src/common.rs) (`rustfs_binary_path`) on
first use — the first invocation is slow, later ones reuse the binary.
Root-heal interruption scenarios use a test-only commit barrier. Prebuild with `e2e-test-hooks` and pin that binary so concurrent cases do not replace it through on-demand builds:
#[ignore = "FAILING on main: excluded from the serial ILM lane pending a fix, see rustfs/backlog#1148 (ilm-1 partial)"]
#[ignore = "global-state ILM integration test: runs serialized in the CI ILM Integration (serial) lane, see ci.yml test-ilm-integration-serial and rustfs/backlog#1148 (ilm-1)"]
#[ignore = "FAILING on main: excluded from the serial ILM lane pending a fix, see rustfs/backlog#1148 (ilm-1 partial)"]
#[ignore = "global-state ILM integration test: runs serialized in the CI ILM Integration (serial) lane, see ci.yml test-ilm-integration-serial and rustfs/backlog#1148 (ilm-1)"]
@@ -30,11 +30,23 @@ These are approved-target invariants. A protocol's explicitly labeled current ex
| Remote PUT is in flight or its response is unknown | Transition transaction | Only cleanup of its own canonical candidate, subject to the transaction recovery predicate | Durable transaction identity plus a known remote-version state; the approved target also requires expiry and durable takeover of the creator fence |
| Local transition commit is complete | Exact transitioned version in `xl.meta` | No | Recovery finds the transaction's logical bucket/object/version and requires the complete recorded source identity (version ID, data directory, modification time, size, and ETag), `TRANSITION_COMPLETE`, and the same remote object, tier, and remote version before removing only the transaction record |
| An ordinary delete removes that transitioned version | Hidden `xl.meta` free-version | Yes | Metadata quorum atomically removes the visible version and preserves its exact tier tuple in the free-version |
| PUT or materialized self-copy replaces a transitioned null version | Hidden `xl.meta` free-version | Yes, after replacement commit and complete physical reference checks | The coordinator supplies one cleanup UUID to every disk; replacement metadata and the old remote tuple are written in the same `xl.meta` commit |
| A recursive prefix/delete-all operation cannot preserve per-object markers | v6 journal bound to an immutable single dispatch manifest or a chunk-parent-bound child manifest | Yes, but only after child/manifest completion and all-pool absence proof | `DispatchAuthorized`, exact local destructive mutation, every journal `Committed`, then child/manifest `Completed`; a chunk parent advances only after that child completion |
| Tier configuration mutation, manual job, or decommission receipt | Intent/admission/copy proof only | No | These records gate configuration, scheduling, or migration; they never become remote-object cleanup owners |
An old journal and a free-version can coexist during compatibility recovery. That coexistence is evidence of multiple possible owners, not permission to choose one: the journal path must retain its record until the version-specific recovery rule proves which owner is authoritative.
Null-version replacement uses the existing free-version format and recovery
worker. Failed metadata preparation preserves both the old version and its
inline bytes; rename rollback restores the complete previous metadata. Recovery
must retain cleanup while any physical replica still references the remote tuple,
including a minority version omitted by quorum merging, or any disk cannot be
checked. A successful replacement needs no in-memory queue receipt to survive
restart: the normal free-version sweep discovers its committed owner. Restores
that retain the same remote tuple and ordinary versioned writes retain their
existing ownership. Older binaries can read this format, but all writers and
cleanup workers need the overwrite fix before these guarantees cover the fleet.
## Persisted record inventory
All keys below are objects in the internal metadata bucket. The table gives the canonical target form. Transition-transaction runtime recovery extracts the final 32 hexadecimal characters and UUID while ignoring shard directories and accepting uppercase hex. Manual-job runtime recovery requires exactly two shards matching the filename prefix, but accepts an uppercase UUID when the shards use the same uppercase text; it then loads the lowercase canonical job by UUID. The decommission validator recomputes and rejects a noncanonical manual-job path, but currently inherits the weaker transition parser. Exact runtime canonical-path validation for both protocols is an approved target.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
houseme