CubeFS Files

UPGRAGDE NOTICE If you are using a CubeFS version earlier than v3.5.0, please refer to the UPGRADE NOTICE in version v3.5.0 for detailed upgrade steps and upgrade to v3.5.0 first.

Upgrade nodes in this order: master → metanode → datanode → flashnode → objectnode → lcnode → cli → client.

Upgrade lcnode and flashnode when needed.

Before enabling MetaPartition learner-based decommissioning (enableMpDecommissionByLearner), upgrade all MetaNodes first.

Clients should use versions later than 3.2.0. Older versions need to be upgraded promptly; otherwise, there will be a risk of compromising stability.

Main Feature

Tag-based MP and DP migration ( [#4072] @RyanWYR @aaronwu2010 )

• Added tag-aware migration and decommissioning for MetaPartitions and DataPartitions.
• Added per-tag concurrency limits and improved target-node selection, scheduling, and recovery control.

Multi-Region metadata support (#4071 @Victor1319 )

• 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.

Multiple data storage pools ( [#4076] @Victor1319 )

• Improved DataPartition placement and accounting across multiple storage pools.
• Added pool-aware writable DataPartition metrics and strengthened storage-mode, storage-type, and topology validation.
• Supported lifecycle-based data rotation across storage pools.

Raft learners and learner-based MP decommissioning ( [#4073] @Victor1319 )

• Added Raft learner support to MetaPartition recovery and decommissioning workflows.
• Improved learner selection, apply-progress validation, membership checks, and recovery scheduling to reduce service impact during replica replacement.

MP metadata alignment and self-healing ( [#4074] @zhumingze1108 )

• Added metadata alignment checks and automatic MetaPartition repair capabilities.
• Improved recovery diagnostics, apply-progress gating, snapshot handling, and cross-version compatibility.

MetaNode proximity reads ( [#4075] @Victor1319 )

• 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.

RocksDB experimental store mode ( [#4083] @aaronwu2010 )

• Added experimental RocksDB storage mode for MetaPartitions, with volume-level defaultStoreMode and per-replica memory / rocksdb selection.
• Improved RocksDB disk selection, unavailable reporting on disk failure, and learner or follower placement that mixes memory and RocksDB store modes.

Enhancements

• Optimized client read-ahead, retry backoff, batched directory reads, and inode/cache maintenance. ( [#4077] @clinx @bboyCH4 @leonrayang ). (
• Improved extent migration with lease validation, inode generation updates, retry budgets, and clearer error classification. (#4081 @zhumingze1108 )
• Improved FlashNode cache-key generation, capacity accounting, prewarm status handling, heartbeat processing, and management messages. ( [#4082] @clinx @bboyCH4 )
• Improved DataPartition decommission token scheduling, cancellation cleanup, progress handling, and automatic bad-disk decommissioning. ( [#4079] @shuqiang-zheng )

Bug Fixes

• Fixed stale client read-ahead data, cache-block double release, ticker leaks, and incomplete node-cache cleanup. (#4077 @bboyCH4 @clinx )
• Fixed DataNode available-space underflow and delayed repair of new replicas with insufficient tiny extents. ( [#4078] @shuqiang-zheng @RyanWYR )
• Fixed DataPartition decommission token scheduling, cancellation cleanup, incorrect disk marking, and stalled bad-disk decommissioning. (#4078 @shuqiang-zheng )