seaweedfs

Chris Lu e8a6fcaafb s3api: skip TTL fast-path for versioned buckets (#8823 ) * s3api: skip TTL fast-path for versioned buckets (#8757) PutBucketLifecycleConfiguration was translating Expiration.Days into filer.conf TTL entries for all buckets. For versioned buckets this is wrong: 1. TTL volumes expire as a unit, destroying all data — including noncurrent versions that should be preserved. 2. Filer-backend TTL (RocksDB compaction filter, Redis key expiry) removes entries without triggering chunk deletion, leaving orphaned volume data with 0 deleted bytes. 3. On AWS S3, Expiration.Days on a versioned bucket creates a delete marker — it does not hard-delete data. TTL has no such nuance. Fix: skip the TTL fast-path when the bucket has versioning enabled or suspended. All lifecycle rules are evaluated at scan time by the lifecycle worker instead. Also fix the lifecycle worker to evaluate Expiration rules against the latest version in .versions/ directories, which was previously skipped entirely — only NoncurrentVersionExpiration was handled. * lifecycle worker: handle SeaweedList error in versions dir cleanup Do not assume the directory is empty when the list call fails — log the error and skip the directory to avoid incorrect deletion. * address review feedback - Fetch version file for tag-based rules instead of reading tags from the .versions directory entry where they are not cached. - Handle getBucketVersioningStatus error by failing closed (treat as versioned) to avoid creating TTL entries on transient failures. - Capture and assert deleteExpiredObjects return values in test. - Improve test documentation.	2 days ago
..
worker	fix(filer): apply default disk type after location-prefix resolution in gRPC AssignVolume (#8836)	2 days ago

s3api: skip TTL fast-path for versioned buckets (#8823 )

* s3api: skip TTL fast-path for versioned buckets (#8757)

PutBucketLifecycleConfiguration was translating Expiration.Days into
filer.conf TTL entries for all buckets. For versioned buckets this is
wrong:

 1. TTL volumes expire as a unit, destroying all data — including
    noncurrent versions that should be preserved.
 2. Filer-backend TTL (RocksDB compaction filter, Redis key expiry)
    removes entries without triggering chunk deletion, leaving orphaned
    volume data with 0 deleted bytes.
 3. On AWS S3, Expiration.Days on a versioned bucket creates a delete
    marker — it does not hard-delete data. TTL has no such nuance.

Fix: skip the TTL fast-path when the bucket has versioning enabled or
suspended. All lifecycle rules are evaluated at scan time by the
lifecycle worker instead.

Also fix the lifecycle worker to evaluate Expiration rules against the
latest version in .versions/ directories, which was previously skipped
entirely — only NoncurrentVersionExpiration was handled.

* lifecycle worker: handle SeaweedList error in versions dir cleanup

Do not assume the directory is empty when the list call fails — log
the error and skip the directory to avoid incorrect deletion.

* address review feedback

- Fetch version file for tag-based rules instead of reading tags from
  the .versions directory entry where they are not cached.
- Handle getBucketVersioningStatus error by failing closed (treat as
  versioned) to avoid creating TTL entries on transient failures.
- Capture and assert deleteExpiredObjects return values in test.
- Improve test documentation.

2 days ago

worker

fix(filer): apply default disk type after location-prefix resolution in gRPC AssignVolume (#8836)

2 days ago