seaweedfs

13264 Commits

128 Branches

331 Tags

368 MiB

Tree: 782ab84f95

Commit Graph

Author	SHA1	Message	Date
Chris Lu	9c3bc138a0	lifecycle worker: scan-time rule evaluation for object expiration (#8809 ) * s3api: extend lifecycle XML types with NoncurrentVersionExpiration, AbortIncompleteMultipartUpload Add missing S3 lifecycle rule types to the XML data model: - NoncurrentVersionExpiration with NoncurrentDays and NewerNoncurrentVersions - NoncurrentVersionTransition with NoncurrentDays and StorageClass - AbortIncompleteMultipartUpload with DaysAfterInitiation - Filter.ObjectSizeGreaterThan and ObjectSizeLessThan - And.ObjectSizeGreaterThan and ObjectSizeLessThan - Filter.UnmarshalXML to properly parse Tag, And, and size filter elements Each new type follows the existing set-field pattern for conditional XML marshaling. No behavior changes - these types are not yet wired into handlers or the lifecycle worker. * s3lifecycle: add lifecycle rule evaluator package New package weed/s3api/s3lifecycle/ provides a pure-function lifecycle rule evaluation engine. The evaluator accepts flattened Rule structs and ObjectInfo metadata, and returns the appropriate Action. Components: - evaluator.go: Evaluate() for per-object actions with S3 priority ordering (delete marker > noncurrent version > current expiration), ShouldExpireNoncurrentVersion() with NewerNoncurrentVersions support, EvaluateMPUAbort() for multipart upload rules - filter.go: prefix, tag, and size-based filter matching - tags.go: ExtractTags() extracts S3 tags from filer Extended metadata, HasTagRules() for scan-time optimization - version_time.go: GetVersionTimestamp() extracts timestamps from SeaweedFS version IDs (both old and new format) Comprehensive test coverage: 54 tests covering all action types, filter combinations, edge cases, and version ID formats. * s3api: add UnmarshalXML for Expiration, Transition, ExpireDeleteMarker Add UnmarshalXML methods that set the internal 'set' flag during XML parsing. Previously these flags were only set programmatically, causing XML round-trip to drop elements. This ensures lifecycle configurations stored as XML survive unmarshal/marshal cycles correctly. Add comprehensive XML round-trip tests for all lifecycle rule types including NoncurrentVersionExpiration, AbortIncompleteMultipartUpload, Filter with Tag/And/size constraints, and a complete Terraform-style lifecycle configuration. * s3lifecycle: address review feedback - Fix version_time.go overflow: guard timestampPart > MaxInt64 before the inversion subtraction to prevent uint64 wrap - Make all expiry checks inclusive (!now.Before instead of now.After) so actions trigger at the exact scheduled instant - Add NoncurrentIndex to ObjectInfo so Evaluate() can properly handle NewerNoncurrentVersions via ShouldExpireNoncurrentVersion() - Add test for high-bit overflow version ID * s3lifecycle: guard ShouldExpireNoncurrentVersion against zero SuccessorModTime Add early return when obj.IsLatest or obj.SuccessorModTime.IsZero() to prevent premature expiration of versions with uninitialized successor timestamps (zero value would compute to epoch, always expired). * lifecycle worker: detect buckets with lifecycle XML, not just filer.conf TTLs Update the detection phase to check for stored lifecycle XML in bucket metadata (key: s3-bucket-lifecycle-configuration-xml) in addition to filer.conf TTL entries. A bucket is proposed for lifecycle processing if it has lifecycle XML OR filer.conf TTLs (backward compatible). New proposal parameters: - has_lifecycle_xml: whether the bucket has stored lifecycle XML - versioning_status: the bucket's versioning state (Enabled/Suspended/"") These parameters will be used by the execution phase (subsequent PR) to determine which evaluation path to use. * lifecycle worker: update detection function comment to reflect XML support * lifecycle worker: add lifecycle XML parsing and rule conversion Add rules.go with: - parseLifecycleXML() converts stored lifecycle XML to evaluator-friendly s3lifecycle.Rule structs, handling Filter.Prefix, Filter.Tag, Filter.And, size constraints, NoncurrentVersionExpiration, AbortIncompleteMultipartUpload, Expiration.Date, and ExpiredObjectDeleteMarker - loadLifecycleRulesFromBucket() reads lifecycle XML from bucket metadata - parseExpirationDate() supports RFC3339 and ISO 8601 date-only formats Comprehensive tests for all XML variants, filter types, and date formats. * lifecycle worker: add scan-time rule evaluation for object expiration Update executeLifecycleForBucket to try lifecycle XML evaluation first, falling back to TTL-only evaluation when no lifecycle XML exists. New listExpiredObjectsByRules() function: - Walks the bucket directory tree - Builds s3lifecycle.ObjectInfo from each filer entry - Calls s3lifecycle.Evaluate() to check lifecycle rules - Skips objects already handled by TTL fast path (TtlSec set) - Extracts tags only when rules use tag-based filters (optimization) - Skips .uploads and .versions directories (handled by other phases) Supports Expiration.Days, Expiration.Date, Filter.Prefix, Filter.Tag, Filter.And, and Filter.ObjectSize* in the scan-time evaluation path. Existing TTL-based path remains for backward compatibility. * lifecycle worker: address review feedback - Use sentinel error (errLimitReached) instead of string matching for scan limit detection - Fix loadLifecycleRulesFromBucket path: use bucketsPath directly as directory for LookupEntry instead of path.Dir which produced the wrong parent * lifecycle worker: fix And filter detection for size-only constraints The And branch condition only triggered when Prefix or Tags were present, missing the case where And contains only ObjectSizeGreaterThan or ObjectSizeLessThan without a prefix or tags. * lifecycle worker: address review feedback round 3 - rules.go: pass through Filter-level size constraints when Tag is present without And (Tag+size combination was dropping sizes) - execution.go: add doc comment to listExpiredObjectsByRules noting that it handles non-versioned objects only; versioned objects are handled by processVersionsDirectory - rules_test.go: add bounds checks before indexing rules[0] --------- Co-authored-by: Copilot <copilot@github.com>	5 days ago

Author

SHA1

Message

Date

Chris Lu

9c3bc138a0

lifecycle worker: scan-time rule evaluation for object expiration (#8809 )

* s3api: extend lifecycle XML types with NoncurrentVersionExpiration, AbortIncompleteMultipartUpload

Add missing S3 lifecycle rule types to the XML data model:
- NoncurrentVersionExpiration with NoncurrentDays and NewerNoncurrentVersions
- NoncurrentVersionTransition with NoncurrentDays and StorageClass
- AbortIncompleteMultipartUpload with DaysAfterInitiation
- Filter.ObjectSizeGreaterThan and ObjectSizeLessThan
- And.ObjectSizeGreaterThan and ObjectSizeLessThan
- Filter.UnmarshalXML to properly parse Tag, And, and size filter elements

Each new type follows the existing set-field pattern for conditional
XML marshaling. No behavior changes - these types are not yet wired
into handlers or the lifecycle worker.

* s3lifecycle: add lifecycle rule evaluator package

New package weed/s3api/s3lifecycle/ provides a pure-function lifecycle
rule evaluation engine. The evaluator accepts flattened Rule structs and
ObjectInfo metadata, and returns the appropriate Action.

Components:
- evaluator.go: Evaluate() for per-object actions with S3 priority
  ordering (delete marker > noncurrent version > current expiration),
  ShouldExpireNoncurrentVersion() with NewerNoncurrentVersions support,
  EvaluateMPUAbort() for multipart upload rules
- filter.go: prefix, tag, and size-based filter matching
- tags.go: ExtractTags() extracts S3 tags from filer Extended metadata,
  HasTagRules() for scan-time optimization
- version_time.go: GetVersionTimestamp() extracts timestamps from
  SeaweedFS version IDs (both old and new format)

Comprehensive test coverage: 54 tests covering all action types,
filter combinations, edge cases, and version ID formats.

* s3api: add UnmarshalXML for Expiration, Transition, ExpireDeleteMarker

Add UnmarshalXML methods that set the internal 'set' flag during XML
parsing. Previously these flags were only set programmatically, causing
XML round-trip to drop elements. This ensures lifecycle configurations
stored as XML survive unmarshal/marshal cycles correctly.

Add comprehensive XML round-trip tests for all lifecycle rule types
including NoncurrentVersionExpiration, AbortIncompleteMultipartUpload,
Filter with Tag/And/size constraints, and a complete Terraform-style
lifecycle configuration.

* s3lifecycle: address review feedback

- Fix version_time.go overflow: guard timestampPart > MaxInt64 before
  the inversion subtraction to prevent uint64 wrap
- Make all expiry checks inclusive (!now.Before instead of now.After)
  so actions trigger at the exact scheduled instant
- Add NoncurrentIndex to ObjectInfo so Evaluate() can properly handle
  NewerNoncurrentVersions via ShouldExpireNoncurrentVersion()
- Add test for high-bit overflow version ID

* s3lifecycle: guard ShouldExpireNoncurrentVersion against zero SuccessorModTime

Add early return when obj.IsLatest or obj.SuccessorModTime.IsZero()
to prevent premature expiration of versions with uninitialized
successor timestamps (zero value would compute to epoch, always expired).

* lifecycle worker: detect buckets with lifecycle XML, not just filer.conf TTLs

Update the detection phase to check for stored lifecycle XML in bucket
metadata (key: s3-bucket-lifecycle-configuration-xml) in addition to
filer.conf TTL entries. A bucket is proposed for lifecycle processing if
it has lifecycle XML OR filer.conf TTLs (backward compatible).

New proposal parameters:
- has_lifecycle_xml: whether the bucket has stored lifecycle XML
- versioning_status: the bucket's versioning state (Enabled/Suspended/"")

These parameters will be used by the execution phase (subsequent PR)
to determine which evaluation path to use.

* lifecycle worker: update detection function comment to reflect XML support

* lifecycle worker: add lifecycle XML parsing and rule conversion

Add rules.go with:
- parseLifecycleXML() converts stored lifecycle XML to evaluator-friendly
  s3lifecycle.Rule structs, handling Filter.Prefix, Filter.Tag, Filter.And,
  size constraints, NoncurrentVersionExpiration, AbortIncompleteMultipartUpload,
  Expiration.Date, and ExpiredObjectDeleteMarker
- loadLifecycleRulesFromBucket() reads lifecycle XML from bucket metadata
- parseExpirationDate() supports RFC3339 and ISO 8601 date-only formats

Comprehensive tests for all XML variants, filter types, and date formats.

* lifecycle worker: add scan-time rule evaluation for object expiration

Update executeLifecycleForBucket to try lifecycle XML evaluation first,
falling back to TTL-only evaluation when no lifecycle XML exists.

New listExpiredObjectsByRules() function:
- Walks the bucket directory tree
- Builds s3lifecycle.ObjectInfo from each filer entry
- Calls s3lifecycle.Evaluate() to check lifecycle rules
- Skips objects already handled by TTL fast path (TtlSec set)
- Extracts tags only when rules use tag-based filters (optimization)
- Skips .uploads and .versions directories (handled by other phases)

Supports Expiration.Days, Expiration.Date, Filter.Prefix, Filter.Tag,
Filter.And, and Filter.ObjectSize* in the scan-time evaluation path.
Existing TTL-based path remains for backward compatibility.

* lifecycle worker: address review feedback

- Use sentinel error (errLimitReached) instead of string matching
  for scan limit detection
- Fix loadLifecycleRulesFromBucket path: use bucketsPath directly
  as directory for LookupEntry instead of path.Dir which produced
  the wrong parent

* lifecycle worker: fix And filter detection for size-only constraints

The And branch condition only triggered when Prefix or Tags were present,
missing the case where And contains only ObjectSizeGreaterThan or
ObjectSizeLessThan without a prefix or tags.

* lifecycle worker: address review feedback round 3

- rules.go: pass through Filter-level size constraints when Tag is
  present without And (Tag+size combination was dropping sizes)
- execution.go: add doc comment to listExpiredObjectsByRules noting
  that it handles non-versioned objects only; versioned objects are
  handled by processVersionsDirectory
- rules_test.go: add bounds checks before indexing rules[0]

---------

Co-authored-by: Copilot <copilot@github.com>

5 days ago

1 Commits (782ab84f9511e900795f3d77b5e8e9fc42e96e20)