You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@hbase.apache.org by "Michael Stack (Jira)" <ji...@apache.org> on 2020/03/19 18:14:00 UTC
[jira] [Comment Edited] (HBASE-23984) [Flakey Tests]
TestMasterAbortAndRSGotKilled fails in teardown
[ https://issues.apache.org/jira/browse/HBASE-23984?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17062833#comment-17062833 ]
Michael Stack edited comment on HBASE-23984 at 3/19/20, 6:13 PM:
-----------------------------------------------------------------
I've changed my approach. I made a new patch. It is the minimal changes needed to fix the two scenarios that can cause tests to hang on exit. I need this committed to make progress on test cleanup. In this new patch I do not address code duplication and cleanup to try and make all Handlers look and act about the same. This latter effort has brought on an interesting discussion. I created HBASE-24015 yesterday because I figured we'd see a reaction like the one we are having here. We can carry on there? Meantime I'd like to get the fix committed. Thanks.
was (Author: stack):
I made a new patch. It is the minimal changes needed to fix the two scenarios that can cause tests to hand on exit. I need this committed to make progress on test cleanup. In this new patch I do not address code duplication and cleanup to try and make all Handlers look and act about the same.
> [Flakey Tests] TestMasterAbortAndRSGotKilled fails in teardown
> --------------------------------------------------------------
>
> Key: HBASE-23984
> URL: https://issues.apache.org/jira/browse/HBASE-23984
> Project: HBase
> Issue Type: Bug
> Reporter: Michael Stack
> Priority: Major
> Attachments: Screen Shot 2020-03-17 at 9.46.49 PM.png
>
>
> Its failing with decent frequency of late in shutdown of cluster. Seems basic. There is an unassign/move going on. Test just checks Master can come back up after being killed. Does not check move is done. If on subsequent cluster shutdown, if the move can't report the Master because its shutting down, then the move fails, we abort the server, and then we get a wonky loop where we can't close because server is aborting.
> At the root, there is a misaccounting when the unassign close fails where we don't cleanup references in the regionserver local RIT accounting. Deeper than this, close code is duplicated in three places that I can see; in RegionServer, in CloseRegionHandler, and in UnassignRegionHandler.
> Let me fix this issue and the code dupe.
> Details:
> From https://builds.apache.org/job/HBase-Flaky-Tests/job/branch-2/5733/artifact/hbase-server/target/surefire-reports/org.apache.hadoop.hbase.master.TestMasterAbortAndRSGotKilled-output.txt
> Here is the unassign handler failing because master went down earlier (Its probably trying to talk to the old Master location)
> {code}
> ***** ABORTING region server asf905.gq1.ygridcore.net,32989,1584000644108: Failed to close region ede67f9f661acc1241faf468b081d548 and can not recover *****
> Cause:
> java.io.IOException: Failed to report close to master: ede67f9f661acc1241faf468b081d548
> at org.apache.hadoop.hbase.regionserver.handler.UnassignRegionHandler.process(UnassignRegionHandler.java:125)
> at org.apache.hadoop.hbase.executor.EventHandler.run(EventHandler.java:104)
> at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
> at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
> at java.lang.Thread.run(Thread.java:748)
> {code}
> ... then the cluster shutdown tries to close the same Region... but fails because we are aborting because of above....
> {code}
> 2020-03-12 08:11:16,600 ERROR [RS_CLOSE_REGION-regionserver/asf905:0-0] helpers.MarkerIgnoringBase(159): ***** ABORTING region server asf905.gq1.ygridcore.net,32989,1584000644108: Unrecoverable exception while closing region hbase:namespace,,1584000652744.78f4ae5beda711a9bebad0b6b8376cc9., still finishing close *****
> java.io.IOException: Aborting flush because server is aborted...
> at org.apache.hadoop.hbase.regionserver.HRegion.internalPrepareFlushCache(HRegion.java:2545)
> at org.apache.hadoop.hbase.regionserver.HRegion.internalFlushcache(HRegion.java:2530)
> at org.apache.hadoop.hbase.regionserver.HRegion.internalFlushcache(HRegion.java:2504)
> at org.apache.hadoop.hbase.regionserver.HRegion.internalFlushcache(HRegion.java:2495)
> at org.apache.hadoop.hbase.regionserver.HRegion.doClose(HRegion.java:1650)
> at org.apache.hadoop.hbase.regionserver.HRegion.close(HRegion.java:1552)
> at org.apache.hadoop.hbase.regionserver.handler.CloseRegionHandler.process(CloseRegionHandler.java:110)
> at org.apache.hadoop.hbase.executor.EventHandler.run(EventHandler.java:104)
> at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
> at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
> at java.lang.Thread.run(Thread.java:748)
> {code}
> ....
> And the RS keeps looping trying to close the Region even though we're aborted and there is handling in RS close Regions to deal with abort.
> Trouble seems to be because when UnassignRegionHandler fails its region close, it does not unregister the Region with rs.getRegionsInTransitionInRS().remove(encodedNameBytes, Boolean.FALSE);
--
This message was sent by Atlassian Jira
(v8.3.4#803005)