[
https://issues.apache.org/jira/browse/HBASE-17657?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15873298#comment-15873298
]
Ted Yu commented on HBASE-17657:
--------------------------------
Looking at ZKAsyncRegistry#getMetaRegionLocation() (where the WARN message from
test output came from) :
{code}
if (stateAndServerName.getFirst() != RegionState.State.OPEN) {
LOG.warn("Meta region for replica " + replicaId + " is in state "
+ stateAndServerName.getFirst());
locs[replicaId] = null;
{code}
Meaning, when location for any replica is not known, null would be returned.
This was what happened during failed test (breakpoint on the above line in
Eclipse isn't hit for successful test run).
This means, the test should exercise getMetaRegionLocation() more than once to
deal with the situation where location for some replica is not known
temporarily.
> TestZKAsyncRegistry is flaky
> -----------------------------
>
> Key: HBASE-17657
> URL: https://issues.apache.org/jira/browse/HBASE-17657
> Project: HBase
> Issue Type: Test
> Reporter: Ted Yu
> Assignee: Ted Yu
> Attachments: 17657.v1.txt
>
>
> TestZKAsyncRegistry showed up in failed tests several times.
> On
> https://builds.apache.org/job/HBASE-Find-Flaky-Tests/lastSuccessfulBuild/artifact/dashboard.html
> , TestZKAsyncRegistry is reported flaky 33% of the time.
> e.g.
> https://builds.apache.org/job/PreCommit-HBASE-Build/5708/testReport/org.apache.hadoop.hbase.client/TestZKAsyncRegistry/test/
> Toward the end of test output:
> {code}
> 2017-02-14 20:23:20,779 WARN [main-EventThread] client.ZKAsyncRegistry(198):
> Meta region for replica 2 is in state PENDING_OPEN
> 2017-02-14 20:23:20,800 INFO [98410feec74d:45445.activeMasterManager]
> hbase.MetaTableAccessor(1767): Updated table hbase:meta state to ENABLED in
> META
> 2017-02-14 20:23:20,804 DEBUG [PostOpenDeployTasks:534574363]
> regionserver.HRegionServer(2034): Finished post open deploy task for
> hbase:meta,,1_0001.534574363
> 2017-02-14 20:23:20,811 DEBUG [RS_OPEN_META-98410feec74d:50745-0]
> handler.OpenRegionHandler(126): Opened hbase:meta,,1_0001.534574363 on
> 98410feec74d,50745,1487103793930
> {code}
> Looks like some replica of the hbase:meta table might not have finished
> opening by the time the test asserted region location not being null.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
