[jira] [Created] (HBASE-18158) Two running in-memory compaction threads may lose data

["Chia-Ping Tsai (JIRA)" <ji...@apache.org>]

Chia-Ping Tsai created HBASE-18158:
--------------------------------------

             Summary: Two running in-memory compaction threads may lose data
                 Key: HBASE-18158
                 URL: https://issues.apache.org/jira/browse/HBASE-18158
             Project: HBase
          Issue Type: Bug
    Affects Versions: 2.0.0
            Reporter: Chia-Ping Tsai
             Fix For: 2.0.0


{code:title=CompactingMemStore.java|borderStyle=solid}
  private void stopCompaction() {
    if (inMemoryFlushInProgress.get()) {
      compactor.stop();
      inMemoryFlushInProgress.set(false);
    }
  }
{code}
The stopCompaction() set inMemoryFlushInProgress to false so there may be two in-memory compaction thread are executed simultaneously. If there are two running InMemoryFlushRunnable, the later InMemoryFlushRunnable may change the versionedList. 
{code:title=MemStoreCompactor.java|borderStyle=solid}
  public boolean start() throws IOException {
    if (!compactingMemStore.hasImmutableSegments()) { // no compaction on empty pipeline
      return false;
    }

    // get a snapshot of the list of the segments from the pipeline,
    // this local copy of the list is marked with specific version
    versionedList = compactingMemStore.getImmutableSegments();
{code}
And the first InMemoryFlushRunnable will use the chagned versionedList to remove the corresponding segments.
{code:title=MemStoreCompactor#doCompaction|borderStyle=solid}
      if (!isInterrupted.get()) {
        if (resultSwapped = compactingMemStore.swapCompactedSegments(
            versionedList, result, (action==Action.MERGE))) {
          // update the wal so it can be truncated and not get too long
          compactingMemStore.updateLowestUnflushedSequenceIdInWAL(true); // only if greater
        }
      }
{code}
In conclusion, first InMemoryFlushRunnable will remove the worng segment. And the later InMemoryFlushRunnable will introduce NPE because first InMemoryFlushRunnable set versionedList to null after compaction.
{code}
Exception in thread "RpcServer.default.FPBQ.Fifo.handler=3,queue=0,port=45712-inmemoryCompactions-1496563908038" java.lang.NullPointerException
        at org.apache.hadoop.hbase.regionserver.CompactionPipeline.swap(CompactionPipeline.java:119)
        at org.apache.hadoop.hbase.regionserver.CompactingMemStore.swapCompactedSegments(CompactingMemStore.java:283)
        at org.apache.hadoop.hbase.regionserver.MemStoreCompactor.doCompaction(MemStoreCompactor.java:212)
        at org.apache.hadoop.hbase.regionserver.MemStoreCompactor.start(MemStoreCompactor.java:122)
        at org.apache.hadoop.hbase.regionserver.CompactingMemStore.flushInMemory(CompactingMemStore.java:388)
        at org.apache.hadoop.hbase.regionserver.CompactingMemStore$InMemoryFlushRunnable.run(CompactingMemStore.java:500)
        at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
{code}



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)