[HBASE-9810] Global memstore size will be calculated wrongly if replaying recovered edits throws exception

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 0.98.0, 0.96.1
• Fix Version/s: 0.98.0, 0.96.1
• Hadoop Flags: Reviewed
• Assignee: Chunhui Shen
• Reporter: Chunhui Shen
• Created: 21/Oct/13 02:55
• Updated: 28/Apr/15 16:18
• Resolved: 14/Nov/13 02:19

Description

Recently we encountered such a case in 0.94-version:

Flush is triggered frequently because:

DEBUG org.apache.hadoop.hbase.regionserver.MemStoreFlusher: Flush thread woke up because memory above low water=14.4g

But, the real global memstore size is about 1g.

It seems the global memstore size has been calculated wrongly.

Through the logs, I find the following root cause log:

ERROR org.apache.hadoop.hbase.regionserver.handler.OpenRegionHandler: Failed open of region=notifysub2_index,\x83\xDC^\xCD\xA3\x8A<\x
E2\x8E\xE6\xAD!\xDC\xE8t\xED,1379148697072.46be7c2d71c555379278a7494df3015e., starting to roll back the global memstore size.
java.lang.NegativeArraySizeException
        at org.apache.hadoop.hbase.KeyValue.getFamily(KeyValue.java:1096)
        at org.apache.hadoop.hbase.regionserver.HRegion.replayRecoveredEdits(HRegion.java:2933)
        at org.apache.hadoop.hbase.regionserver.HRegion.replayRecoveredEditsIfAny(HRegion.java:2811)
        at org.apache.hadoop.hbase.regionserver.HRegion.initializeRegionInternals(HRegion.java:583)
        at org.apache.hadoop.hbase.regionserver.HRegion.initialize(HRegion.java:499)
        at org.apache.hadoop.hbase.regionserver.HRegion.openHRegion(HRegion.java:3939)
        at org.apache.hadoop.hbase.regionserver.HRegion.openHRegion(HRegion.java:3887)
        at org.apache.hadoop.hbase.regionserver.handler.OpenRegionHandler.openRegion(OpenRegionHandler.java:332)
        at org.apache.hadoop.hbase.regionserver.handler.OpenRegionHandler.process(OpenRegionHandler.java:108)
        at org.apache.hadoop.hbase.executor.EventHandler.run(EventHandler.java:169)
        at java.util.concurrent.ThreadPoolExecutor$Worker.runTask(ThreadPoolExecutor.java:886)
        at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:908)
        at java.lang.Thread.run(Thread.java:662)

Browse the code of this part, seems a critial bug about global memstore size when replaying recovered edits.

(RegionServerAccounting#clearRegionReplayEditsSize is called for each edit file, it means the roll back size is smaller than actual when calling RegionServerAccounting#rollbackRegionReplayEditsSize)

Anyway, the solution is easy as the patch.