* Preserve multipart replication recovery state
Multipart uploads previously only scheduled replication after completion, leaving no persisted pending state for scanner recovery if the initial async work was lost. Persist the same pending replication metadata during multipart initialization and let completion evaluate the object metadata that was actually stored.
The scanner heal path also treated ordinary pending objects as delete-replication candidates. Restrict that path to delete markers and version purge state so pending objects remain eligible for object replication heal.
Constraint: Bucket replication recovery depends on persisted object metadata after the async queue is unavailable.
Rejected: Rely only on immediate completion-time scheduling | it cannot recover after process restart or worker loss.
Confidence: high
Scope-risk: moderate
Directive: Keep multipart upload initialization aligned with single PUT replication metadata semantics.
Tested: cargo fmt --all --check
Tested: cargo clippy --workspace --all-features --all-targets -- -D warnings
Tested: LC_ALL=en_US.UTF-8 LANG=en_US.UTF-8 make pre-commit
Tested: Runtime replication outage check confirmed multipart xl.meta stores PENDING status and timestamp.
* fix(replication): preserve version purge scanner state
Role-derived replication configs need target-scoped status strings before scanner heal can build per-target purge status. The duplicated replication-status assignment left version purge status unset, so scanner recovery could lose the target-level purge state.
Constraint: Scanner heal derives per-target purge decisions from version_purge_status_internal.
Rejected: Leave the duplicate as a harmless cleanup | it changes recovery behavior for role-only configs with version purge state.
Confidence: high
Scope-risk: narrow
Directive: Keep role-derived replication and version purge internal status mapping symmetric.
Tested: cargo fmt --all --check
Tested: cargo test -p rustfs-ecstore heal -- --nocapture
---------
Co-authored-by: wly <wlywly0735@126.com>
Co-authored-by: houseme <housemecn@gmail.com>
* Initial plan
* Implement priority-based heal queue with deduplication
Co-authored-by: houseme <4829346+houseme@users.noreply.github.com>
* Apply cargo fmt formatting fixes
Co-authored-by: houseme <4829346+houseme@users.noreply.github.com>
* Add queue monitoring, better error handling, and adaptive processing
- Add priority-based queue statistics tracking
- Implement queue capacity warnings (>80% full)
- Process multiple tasks per cycle when capacity allows
- Add proper error logging for failed heal request submissions
- Add Hash trait to HealPriority for HashMap support
- Improve observability with detailed queue status logs
Co-authored-by: houseme <4829346+houseme@users.noreply.github.com>
* Add comprehensive diagnostic logging to heal flow
- Add detailed logging for shard availability checks during reads
- Log successful heal request submissions
- Add detailed logging in heal_object for disk checking
- Log each disk's heal decision with reason and endpoint
- Add logging for heal writing and renaming phases
- Improve variable naming (nil_count -> available_shards)
- Add success confirmation messages for renamed healed data
These logs will help diagnose where the heal flow is failing.
Co-authored-by: houseme <4829346+houseme@users.noreply.github.com>
* fix
* feat(heal): enhance robustness, safety, and observability of heal process
- **Logging & Observability**:
- Upgrade critical heal logs from [info](/crates/e2e_test/src/reliant/node_interact_test.rs:196:0-213:1) to `warn` for better visibility.
- Implement structured logging with `tracing` fields for machine readability.
- Add `#[tracing::instrument]` to [HealTask](c/crates/ahm/src/heal/task.rs:182:0-205:1) and [SetDisks](/crates/ecstore/src/set_disk.rs:120:0-131:1) methods for automatic context propagation.
- **Robustness**:
- Add exponential backoff retry (3 attempts) for acquiring write locks in [heal_object](/crates/ahm/src/heal/storage.rs:438:4-460:5) to handle contention.
- Handle [rename_data](/crates/ecstore/src/set_disk.rs:392:4-516:5) failures gracefully by preserving temporary files instead of forcing deletion, preventing potential data loss.
- **Data Safety**:
- Fix [object_exists](/crates/ahm/src/heal/storage.rs:395:4-412:5) to propagate IO errors instead of treating them as "object not found".
- Update [ErasureSetHealer](/crates/ahm/src/heal/erasure_healer.rs:28:0-33:1) to mark objects as failed rather than skipped when existence checks error, ensuring they are tracked for retry.
* fix
* fmt
* improve code for heal_object
* fix
* fix
* fix
---------
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: houseme <4829346+houseme@users.noreply.github.com>
Co-authored-by: houseme <housemecn@gmail.com>
- Normalize ETags by removing quotes before comparison in complete_multipart_upload
- Fix ETag comparison in replication logic to handle quoted ETags from API responses
- Fix ETag comparison in transition object logic
- Add unit tests for trim_etag function
This fixes the ETag mismatch error when uploading large files (5GB+) via multipart upload,
which was caused by PR #592 adding quotes to ETag responses while internal storage remains unquoted.
Fixes#625
This commit introduces a significant reorganization of the project structure to improve maintainability and clarity.
Key changes include:
- Adjusted the directory layout for a more logical module organization.
- Removed unused crate dependencies, reducing the overall project size and potentially speeding up build times.
- Updated import paths and configuration files to reflect the structural changes.
安正超