[HBASE-14897] TestTableLockManager.testReapAllTableLocks is flakey

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 1.3.2, 1.4.3, 2.0.0, 1.2.7
• Assignee: Heng Chen
• Reporter: Heng Chen
• Created: 30/Nov/15 15:36
• Updated: 01/Feb/19 19:54
• Resolved: 20/Feb/18 15:49

Description

It comes from email list which stack post.

This is the some relates QA information.

https://builds.apache.org/view/H-L/view/HBase/job/HBase-Trunk_matrix/512/jdk=latest1.8,label=Hadoop/testReport/org.apache.hadoop.hbase.master/TestTableLockManager/testReapAllTableLocks/

The reason is here.

    writeLocksObtained.await();
    writeLocksAttempted.await();

writeLocksAttempted maybe count down to 0 before created node on ZK, and main thread will go on to run lockManager.reapWriteLocks(), And after that node was created on ZK, so relates lock acquire will timeout.

I upload a patch which can reproduce this issue.