CubeFS

v3.6.0Storage & Data

2026년 8월 12일

메타데이터와 저장소 동작 범위가 넓어졌으며, 마이그레이션과 복구 관련 기능도 보강되었습니다. 업그레이드는 명시된 순서와 사전 조건을 전제로 하며, 보안 권고나 명시적인 보안 결함은 공개되지 않았습니다.

그 외 기록된 변경 24건 전체additions 11 · value changes 8 · fixes 4 · constraints 1

- Added tag-aware migration and decommissioning for MetaPartitions and DataPartitions.
- Added per-tag concurrency limits and improved target-node selection, scheduling, and recovery control.
- Added Region management for metadata nodes and volumes, including default Regions, allowed Region sets, and MetaPartition Region policies.
- Added Region-aware placement validation, online policy updates, CLI operations, persistence, and monitoring metrics.
- Added pool-aware writable DataPartition metrics
- Supported lifecycle-based data rotation across storage pools.
- Added Raft learner support to MetaPartition recovery and decommissioning workflows.
- Added metadata alignment checks and automatic MetaPartition repair capabilities.
- Added Meta NearRead and follower-read support to route metadata reads to nearby, low-latency replicas.
- Added follower-read leases and dirty-inode tracking to preserve read-your-writes consistency after metadata updates.
- Added experimental RocksDB storage mode for MetaPartitions, with volume-level defaultStoreMode and per-replica memory / rocksdb selection.

- Improved DataPartition placement and accounting across multiple storage pools.
- Improved learner selection, apply-progress validation, membership checks, and recovery scheduling to reduce service impact during replica replacement.
- Improved recovery diagnostics, apply-progress gating, snapshot handling, and cross-version compatibility.
- Improved RocksDB disk selection, unavailable reporting on disk failure, and learner or follower placement that mixes memory and RocksDB store modes.
- Optimized client read-ahead, retry backoff, batched directory reads, and inode/cache maintenance.
- Improved extent migration with lease validation, inode generation updates, retry budgets, and clearer error classification.
- Improved FlashNode cache-key generation, capacity accounting, prewarm status handling, heartbeat processing, and management messages.
- Improved DataPartition decommission token scheduling, cancellation cleanup, progress handling, and automatic bad-disk decommissioning.

- Fixed stale client read-ahead data, cache-block double release, ticker leaks, and incomplete node-cache cleanup.
- Fixed DataNode available-space underflow
delayed repair of new replicas with insufficient tiny extents.
- Fixed DataPartition decommission token scheduling, cancellation cleanup, incorrect disk marking, and stalled bad-disk decommissioning.

CubeFS 스택에 추가

조치가 필요한 변경이 나오면 주간 메일로 받아볼 수 있습니다. 이번 릴리스처럼 조치가 없는 주에는 메일을 보내지 않습니다.