[HBASE-4028] Hmaster crashes caused by splitting log.

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.90.3
• Fix Version/s: 0.90.4
• Component/s: master
• Hadoop Flags: Reviewed
• Assignee: gaojinchao
• Reporter: gaojinchao
• Due: 24/Jun/11
• Created: 24/Jun/11 06:14
• Updated: 20/Nov/15 11:55
• Resolved: 26/Jun/11 00:34

Description

In my performance cluster(0.90.3), The Hmaster memory from 100 M up to 4G when one region server crashed.
I added some print in function doneWriting and found the values of totalBuffered is negative.

10:29:52,119 WARN org.apache.hadoop.hbase.regionserver.wal.HLogSplitter: gjc:release Used -565832
hbase-root-master-157-5-111-21.log:2011-06-24 10:29:52,119 WARN org.apache.hadoop.hbase.regionserver.wal.HLogSplitter: gjc:release Used -565832release size25168

void doneWriting(RegionEntryBuffer buffer) {
synchronized (this)

{ LOG.warn("gjc1: relase currentlyWriting +biggestBufferKey " + buffer.encodedRegionName ); boolean removed = currentlyWriting.remove(buffer.encodedRegionName); assert removed; }

long size = buffer.heapSize();

synchronized (dataAvailable)

{ totalBuffered -= size; LOG.warn("gjc:release Used " + totalBuffered ); // We may unblock writers dataAvailable.notifyAll(); }

LOG.warn("gjc:release Used " + totalBuffered + "release size"+ size);
}