[HBASE-21031] Memory leak if replay edits failed during region opening

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.1.0, 2.0.1
• Fix Version/s: 3.0.0-alpha-1, 2.2.0, 2.1.1, 2.0.2
• Assignee: Allan Yang
• Reporter: Allan Yang
• Created: 09/Aug/18 12:53
• Updated: 04/Mar/19 09:06
• Resolved: 22/Aug/18 14:43

Description

Due to HBASE-21029, when replaying edits with a lot of same cells, the memstore won't flush, a exception will throw when all heap space was used:

2018-08-06 15:52:27,590 ERROR [RS_OPEN_REGION-regionserver/hb-bp10cw4ejoy0a2f3f-009:16020-2] handler.OpenRegionHandler(302): Failed open of region=hbase_test,dffa78,1531227033378.cbf9a2daf3aaa0c7e931e9c9a7b53f41., starting to roll back the global memstore size.
java.lang.OutOfMemoryError: Java heap space
        at java.nio.HeapByteBuffer.<init>(HeapByteBuffer.java:57)
        at java.nio.ByteBuffer.allocate(ByteBuffer.java:335)
        at org.apache.hadoop.hbase.regionserver.OnheapChunk.allocateDataBuffer(OnheapChunk.java:41)
        at org.apache.hadoop.hbase.regionserver.Chunk.init(Chunk.java:104)
        at org.apache.hadoop.hbase.regionserver.ChunkCreator.getChunk(ChunkCreator.java:226)
        at org.apache.hadoop.hbase.regionserver.ChunkCreator.getChunk(ChunkCreator.java:180)
        at org.apache.hadoop.hbase.regionserver.ChunkCreator.getChunk(ChunkCreator.java:163)
        at org.apache.hadoop.hbase.regionserver.MemStoreLABImpl.getOrMakeChunk(MemStoreLABImpl.java:273)
        at org.apache.hadoop.hbase.regionserver.MemStoreLABImpl.copyCellInto(MemStoreLABImpl.java:148)
        at org.apache.hadoop.hbase.regionserver.MemStoreLABImpl.copyCellInto(MemStoreLABImpl.java:111)
        at org.apache.hadoop.hbase.regionserver.Segment.maybeCloneWithAllocator(Segment.java:178)
        at org.apache.hadoop.hbase.regionserver.AbstractMemStore.maybeCloneWithAllocator(AbstractMemStore.java:287)
        at org.apache.hadoop.hbase.regionserver.AbstractMemStore.add(AbstractMemStore.java:107)
        at org.apache.hadoop.hbase.regionserver.HStore.add(HStore.java:706)
        at org.apache.hadoop.hbase.regionserver.HRegion.restoreEdit(HRegion.java:5494)
        at org.apache.hadoop.hbase.regionserver.HRegion.replayRecoveredEdits(HRegion.java:4608)
        at org.apache.hadoop.hbase.regionserver.HRegion.replayRecoveredEditsIfAny(HRegion.java:4404)

After this exception, the memstore did not roll back, and since MSLAB is used, all the chunk allocated won't release for ever. Those memory is leak forever...

We need to rollback the memory if open region fails(For now, only global memstore size is decreased after failure).

Another problem is that we use replayEditsPerRegion in RegionServerAccounting to record how many memory used during replaying. And decrease the global memstore size if replay fails. This is not right, since during replaying, we may also flush the memstore, the size in the map of replayEditsPerRegion is not accurate at all!

Issue Links

• relates to: HBASE-21029 Miscount of memstore's heap/offheap size if same cell was put
• links to: ReviewBoard