[HBASE-4341] HRS#closeAllRegions should take care of HRS#onlineRegions's weak consistency

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.90.4
• Fix Version/s: 0.90.5
• Component/s: regionserver
• Hadoop Flags: Reviewed
• Assignee: Jieshan Bean Assign to me
• Reporter: Jieshan Bean
• Created: 07/Sep/11 02:42
• Updated: 20/Nov/15 11:54
• Resolved: 08/Sep/11 15:39

Description

This's the reason of why did "https://builds.apache.org/job/hbase-0.90/282" get failure . In this test, one case was timeout and cause the whole test process got killed.

[logs]
Here's the related logs(From org.apache.hadoop.hbase.mapreduce.TestTableMapReduce-output.txt):

2011-08-31 10:09:01,089 INFO  [RegionServer:0;vesta.apache.org,52257,1314785332968.leaseChecker] regionserver.Leases(124): RegionServer:0;vesta.apache.org,52257,1314785332968.leaseChecker closing leases
2011-08-31 10:09:01,089 INFO  [RegionServer:0;vesta.apache.org,52257,1314785332968.leaseChecker] regionserver.Leases(131): RegionServer:0;vesta.apache.org,52257,1314785332968.leaseChecker closed leases
2011-08-31 10:09:01,403 INFO  [RegionServer:0;vesta.apache.org,52257,1314785332968] regionserver.HRegionServer(709): Waiting on 1 regions to close
2011-08-31 10:09:01,403 DEBUG [RegionServer:0;vesta.apache.org,52257,1314785332968] regionserver.HRegionServer(713): {74a7a8befdf9561dc1d90c4241afeac7=mrtest,uuu,1314785328546.74a7a8befdf9561dc1d90c4241afeac7.}
2011-08-31 10:09:01,697 INFO  [Master:0;vesta.apache.org:50036] master.ServerManager(465): Waiting on regionserver(s) to go down vesta.apache.org,52257,1314785332968
2011-08-31 10:09:02,697 INFO  [Master:0;vesta.apache.org:50036] master.ServerManager(465): Waiting on regionserver(s) to go down vesta.apache.org,52257,1314785332968
2011-08-31 10:09:03,008 INFO  [vesta.apache.org:50036.timeoutMonitor] hbase.Chore(79): vesta.apache.org:50036.timeoutMonitor exiting
2011-08-31 10:09:03,697 INFO  [Master:0;vesta.apache.org:50036] master.ServerManager(465): Waiting on regionserver(s) to go down vesta.apache.org,52257,1314785332968
2011-08-31 10:09:04,697 INFO  [Master:0;vesta.apache.org:50036] master.ServerManager(465): Waiting on regionserver(s) to go down vesta.apache.org,52257,1314785332968
2011-08-31 10:09:05,698 INFO  [Master:0;vesta.apache.org:50036] master.ServerManager(465): Waiting on regionserver(s) to go down vesta.apache.org,52257,1314785332968
2011-08-31 10:09:06,698 INFO  [Master:0;vesta.apache.org:50036] master.ServerManager(465): Waiting on regionserver(s) to go down vesta.apache.org,52257,1314785332968
2011-08-31 10:09:07,698 INFO  [Master:0;vesta.apache.org:50036] master.ServerManager(465): Waiting on regionserver(s) to go down vesta.apache.org,52257,1314785332968

[Analysis]
One region was opened during the RS's stopping.
This is method of "HRS#closeAllRegions":

  protected void closeAllRegions(final boolean abort) {
    closeUserRegions(abort);
    -------------------------
    if (meta != null) closeRegion(meta.getRegionInfo(), abort, false);
    if (root != null) closeRegion(root.getRegionInfo(), abort, false);
  }

HRS#onlineRegions is a ConcurrentHashMap. So walk down this map may not get all the data if some entries are been added during the traverse. Once one region was missed, it can't be closed anymore. And this regionserver will not be stopped normally. Then the following logs occurred:

2011-08-31 10:09:01,403 INFO  [RegionServer:0;vesta.apache.org,52257,1314785332968] regionserver.HRegionServer(709): Waiting on 1 regions to close
2011-08-31 10:09:01,403 DEBUG [RegionServer:0;vesta.apache.org,52257,1314785332968] regionserver.HRegionServer(713): {74a7a8befdf9561dc1d90c4241afeac7=mrtest,uuu,1314785328546.74a7a8befdf9561dc1d90c4241afeac7.}
2011-08-31 10:09:01,697 INFO  [Master:0;vesta.apache.org:50036] master.ServerManager(465): Waiting on regionserver(s) to go down vesta.apache.org,52257,1314785332968