[HBASE-27464] In memory compaction 'COMPACT' may cause data corruption when adding cells large than maxAlloc(default 256k) size

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Fix Version/s: 2.6.0, 3.0.0-alpha-4, 2.5.2, 2.4.16
• Component/s: in-memory-compaction
• Hadoop Flags: Reviewed
• Assignee: chenglei
• Reporter: zhuobin zheng
• Created: 04/Nov/22 07:59
• Updated: 18/Nov/22 19:29
• Resolved: 17/Nov/22 15:26

Description

When init 'CellChunkImmutableSegment' for 'COMPACT' action, we not force copy to current MSLab. 

When cell size bigger than maxAlloc, cell will stay in previous chunk which will recycle after segment replace, and we may read wrong data when these chunk reused by others.

 

Timeline:

1. add a cell 'A' bigger than 256K
2. cell 'A' will copy to a chunk 'A' when first compact
3. cell 'A' will retain in chunk 'A' when second compact
4. chunk 'A' recycled after segment swap and close

Issue Links

• links to: GitHub Pull Request #4881