[HBASE-12766] TestSplitLogManager#testGetPreviousRecoveryMode sometimes fails due to race condition

• Type: Test
• Status: Closed
• Priority: Minor
• Resolution: Duplicate
• Assignee: Unassigned
• Reporter: Ted Yu
• Created: 28/Dec/14 15:14
• Updated: 17/Jun/22 18:43
• Resolved: 13/Mar/15 15:47

Description

From https://builds.apache.org/job/HBase-1.0/614/testReport/junit/org.apache.hadoop.hbase.master/TestSplitLogManager/testGetPreviousRecoveryMode/ :

java.lang.AssertionError: Mode4=LOG_SPLITTING
	at org.junit.Assert.fail(Assert.java:88)
	at org.junit.Assert.assertTrue(Assert.java:41)
	at org.apache.hadoop.hbase.master.TestSplitLogManager.testGetPreviousRecoveryMode(TestSplitLogManager.java:656)
...
2014-12-27 19:04:56,576 INFO  [Thread-8] coordination.ZKSplitLogManagerCoordination(594): found orphan task testRecovery
2014-12-27 19:04:56,577 INFO  [Thread-8] coordination.ZKSplitLogManagerCoordination(598): Found 1 orphan tasks and 0 rescan nodes
2014-12-27 19:04:56,578 DEBUG [main-EventThread] coordination.ZKSplitLogManagerCoordination(464): task not yet acquired /hbase/splitWAL/testRecovery ver = 0
2014-12-27 19:04:56,578 INFO  [main-EventThread] coordination.ZKSplitLogManagerCoordination(548): creating orphan task /hbase/splitWAL/testRecovery
2014-12-27 19:04:56,578 INFO  [main-EventThread] coordination.ZKSplitLogManagerCoordination(178): resubmitting unassigned orphan task /hbase/splitWAL/testRecovery
2014-12-27 19:04:56,578 INFO  [main-EventThread] coordination.ZKSplitLogManagerCoordination(229): resubmitting task /hbase/splitWAL/testRecovery
2014-12-27 19:04:56,582 INFO  [Thread-8] master.TestSplitLogManager(650): Mode1=LOG_SPLITTING
2014-12-27 19:04:56,584 DEBUG [main-EventThread] zookeeper.ZooKeeperWatcher(313): split-log-manager-tests58920b37-7850-44e5-8b97-871caff81fdb-0x14a8d231db70000, quorum=localhost:60541, baseZNode=/hbase Received ZooKeeper Event, type=NodeDataChanged, state=SyncConnected, path=/hbase/splitWAL/testRecovery
2014-12-27 19:04:56,584 INFO  [Thread-8] master.TestSplitLogManager(653): Mode2=LOG_SPLITTING
2014-12-27 19:04:56,584 DEBUG [Thread-8] coordination.ZKSplitLogManagerCoordination(870): Distributed log replay=true
2014-12-27 19:04:56,585 WARN  [main-EventThread] coordination.ZKSplitLogManagerCoordination$GetDataAsyncCallback(996): task znode /hbase/splitWAL/testRecovery vanished or not created yet.
2014-12-27 19:04:56,585 INFO  [main-EventThread] coordination.ZKSplitLogManagerCoordination(472): task /hbase/splitWAL/RESCAN0000000001 entered state: DONE dummy-master,1,1

From the above log we can see that by the time the following is called (line 654 in test):

    slm.setRecoveryMode(false);

the split task was not done - it entered done state 1 millisecond later.
So ZKSplitLogManagerCoordination#hasSplitLogTask was true and isForInitialization parameter is false, leading to the execution of the following branch:

      } else if (!isForInitialization) {
        // splitlogtask hasn't drained yet, keep existing recovery mode
        return;

Thus recoveryMode was left in LOG_SPLITTING state.