Details
-
Test
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
2.3.0, 2.4.0
-
None
-
None
-
Reviewed
Description
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:
Here is the unassign handler failing because master went down earlier (Its probably trying to talk to the old Master location)
***** 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$Worker.run(ThreadPoolExecutor.java:624) at java.lang.Thread.run(Thread.java:748)
... then the cluster shutdown tries to close the same Region... but fails because we are aborting because of above....
2020-03-12 08:11:16,600 ERROR [RS_CLOSE_REGION-regionserver/asf905:0-0] helpers.MarkerIgnoringBase(159): ***** ABORTING region server asf905.gq1.ygridcore.net,32989,1584000644108: Unrecoverable exception while closing region hbase:namespace,,1584000652744.78f4ae5beda711a9bebad0b6b8376cc9., still finishing close ***** java.io.IOException: Aborting flush because server is aborted... at org.apache.hadoop.hbase.regionserver.HRegion.internalPrepareFlushCache(HRegion.java:2545) at org.apache.hadoop.hbase.regionserver.HRegion.internalFlushcache(HRegion.java:2530) at org.apache.hadoop.hbase.regionserver.HRegion.internalFlushcache(HRegion.java:2504) at org.apache.hadoop.hbase.regionserver.HRegion.internalFlushcache(HRegion.java:2495) at org.apache.hadoop.hbase.regionserver.HRegion.doClose(HRegion.java:1650) at org.apache.hadoop.hbase.regionserver.HRegion.close(HRegion.java:1552) at org.apache.hadoop.hbase.regionserver.handler.CloseRegionHandler.process(CloseRegionHandler.java:110) at org.apache.hadoop.hbase.executor.EventHandler.run(EventHandler.java:104) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624) at java.lang.Thread.run(Thread.java:748)
....
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);
Attachments
Attachments
Issue Links
- is duplicated by
-
HBASE-23985
[flakey test] TestZooKeeper
-
- Resolved
-
- is related to
-
HBASE-24015
Coverage for Assign and Unassign of Regions on RegionServer on failure
-
- Resolved
-
- relates to
-
HBASE-24026 Revisit cluster shutdown
-
- Open
-
- links to