[HBASE-24984] WAL corruption due to early DBBs re-use when Durability.ASYNC_WAL is used with multi operation

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 2.0.0
• Fix Version/s: 2.5.0, 2.3.6, 3.0.0-alpha-2, 2.4.5
• Component/s: rpc, wal
• Hadoop Flags: Reviewed
• Assignee: Gaurav Kanade
• Reporter: Liu Junhong
• Due: 01/Sep/20
• Created: 06/Sep/20 10:22
• Updated: 22/Jul/21 13:52
• Resolved: 22/Jul/21 03:10

Description

After bugfix HBASE-22539, When client use BufferedMutator or multiple mutation , there will be one RpcCall and mutliple FSWALEntry .  At the time RpcCall finish and one FSWALEntry call release() , the remain FSWALEntries may trigger RuntimeException or segmentation fault .

We should use RefCnt  instead of AtomicInteger for org.apache.hadoop.hbase.ipc.ServerCall.reference?

Issue Links

• is related to: HBASE-22539 WAL corruption due to early DBBs re-use when Durability.ASYNC_WAL is used
• is related to: HBASE-26092 JVM core dump in the replication path
• relates to: HBASE-25701 RegionServer JVM crash when append wal entry
• links to: GitHub Pull Request #3495
• links to: GitHub Pull Request #3512
• links to: PR for branch-2