mirror of
https://github.com/rustfs/rustfs.git
synced 2026-08-16 18:08:21 +00:00
Zhengchao An
69719c257e
* chore(ecstore): remove the pool-level ListObjects pagination copy The ListObjects pagination pipeline existed in three near-copies in one file; production listing never reaches the Sets copy, which ECStore bypasses by expanding straight to per-set disks. This removes it: impl ListOperations for Sets (61 lines of pure forwarding in core/sets.rs) and the impl Sets pagination block (826 lines of inner_list_objects_v2 / list_objects_generic / inner_list_object_versions / list_path / list_merged / walk_internal in store/list_objects.rs). Two preconditions verified before deleting rather than taken on faith: the architecture guard pins only set_disks_implements_storage_list_operations_contract, so nothing requires the Sets trait impl; and the four Sets pagination methods had no cross-file caller besides that trait impl. The single test consumer moves to the surviving pipeline instead of being deleted: writes still go through the pool, and the listing assertion now targets the set-level implementation. It is renamed accordingly so the name still describes what it covers. The logging guardrail's TRACE-only requirement for Sets::list_objects_v2 retires in the same diff — the wrapper it pinned no longer exists. The ECStore and SetDisks entries are untouched. The SetDisks copy stays for now: its trait impl is guard-pinned, so replacing the duplicate pipeline behind it needs the generic helper the issue schedules for post-1.0. Verification: cargo nextest run -p rustfs-ecstore 4020 passed; check_architecture_migration_rules.sh and check_logging_guardrails.sh pass; clippy --lib --tests -D warnings clean; make pre-commit green. Ref rustfs/backlog#1821 (PR1). * chore(ecstore): fold the ListObjects forwarders into the ECStore impl store/list.rs held two thin forwarders, handle_list_objects_v2 and handle_list_object_versions, that only re-entered the inner_* implementations. The ListOperations impl now calls those directly and the file goes away. The logging guardrail's trace_hot_spans list pinned handle_list_objects_v2 as TRACE-only; that entry is retired in the same diff, adjacent to the sets.rs entry retired by the preceding commit. Ref rustfs/backlog#1821. * chore(ecstore): drop the type aliases orphaned by the pagination removal core/sets.rs declared four local type aliases — ListObjectsV2Info, ListObjectVersionsInfo, ObjectInfoOrErr and WalkOptions — used only by the pool-level pagination pipeline removed earlier in this branch. store/list_objects.rs keeps its own live copies of the same aliases. They only surface now that #6087 removed the core module's dead_code blanket: on that older base each PR was warning-free on its own, and the combination is what exposes them. Their storage_api_contracts imports go with them. Ref rustfs/backlog#1823, rustfs/backlog#1821. * fix(ecstore): preserve Sets listing compatibility
ecstore Integration Tests
MinIO-generated encrypted fixtures
rustfs/src/storage/minio_generated_read_test.rs validates the bitrot -> GetObjectReader path against raw MinIO backend data captured by
.\rustfs\scripts\minio_fixture_lab\lab.py.
It currently covers multipart fixtures for:
sse-s3-multipart-8msse-kms-multipart-8m
Required environment variables:
RUSTFS_MINIO_FIXTURE_ROOTRUSTFS_MINIO_STATIC_KMS_KEY_B64
Example:
$env:RUSTFS_MINIO_FIXTURE_ROOT = '.\rustfs\tmp\minio-fixture-lab-local-key'
$env:RUSTFS_MINIO_STATIC_KMS_KEY_B64 = '<base64-32-byte-local-minio-kms-key>'
cargo +1.97.1 test -p rustfs --features rio-v2 storage::minio_generated_read_test --lib -- --ignored