[HBASE-24881] Fix flaky TestMasterAbortAndRSGotKilled for branch-2.2

• Type: Sub-task
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.2.6
• Assignee: Guanghao Zhang
• Reporter: Guanghao Zhang
• Created: 14/Aug/20 08:38
• Updated: 26/Aug/20 07:56
• Resolved: 26/Aug/20 07:56

Description

I meet this problem on branch-2.2 too. This case happened because the DelayCloseCP. The event execute order is:

1. Close regiong. But because the DelayCloseCP, it will close after 10 seconds.
2. Finish ut and shutdown cluster.
3. Shutdown master.
4. Shutdown RS. Call waitOnAllRegionsToClose method. But abortRequested is false now.
5. Close region and failed because master is down and report master error. Then abort RegionServer and set abortRequested to ture.
6. waitOnAllRegionsToClose hanged because the online regions cannot be empty.

 

waitOnAllRegionsToClose(final boolean abort) already consider the abort case but the problem is abortRequested is false when call this method. I thought the fix should be that keep to check the abortRequested in waitOnAllRegionsToClose method internal.

Issue Links

• links to: GitHub Pull Request #2258