[
https://issues.apache.org/jira/browse/HBASE-5099?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13177351#comment-13177351
]
[email protected] commented on HBASE-5099:
------------------------------------------------------
bq. On 2011-12-29 19:02:17, Ted Yu wrote:
bq. > src/main/java/org/apache/hadoop/hbase/master/HMaster.java, line 1475
bq. > <https://reviews.apache.org/r/3323/diff/3/?file=65929#file65929line1475>
bq. >
bq. > Timeout value should be included in the exception message.
It may not because of timeout.
bq. On 2011-12-29 19:02:17, Ted Yu wrote:
bq. > src/test/java/org/apache/hadoop/hbase/master/TestMasterRecovery.java,
line 2
bq. > <https://reviews.apache.org/r/3323/diff/3/?file=65930#file65930line2>
bq. >
bq. > Year is not needed.
I copied it from another test case. I can remove it.
bq. On 2011-12-29 19:02:17, Ted Yu wrote:
bq. > src/test/java/org/apache/hadoop/hbase/master/TestMasterRecovery.java,
line 34
bq. > <https://reviews.apache.org/r/3323/diff/3/?file=65930#file65930line34>
bq. >
bq. > This shouldn't be small test since mini cluster is involved.
Ok.
bq. On 2011-12-29 19:02:17, Ted Yu wrote:
bq. > src/test/java/org/apache/hadoop/hbase/master/TestMasterRecovery.java,
line 41
bq. > <https://reviews.apache.org/r/3323/diff/3/?file=65930#file65930line41>
bq. >
bq. > We shouldn't pass 1 here since that means 1 master.
It means 1 region server. Actually, I do want 1 master so it can recover
itself. Otherwise, the backup master will take over and the active master
doesn't have a chance to recover in this scenario.
bq. On 2011-12-29 19:02:17, Ted Yu wrote:
bq. > src/main/java/org/apache/hadoop/hbase/master/HMaster.java, line 1452
bq. > <https://reviews.apache.org/r/3323/diff/3/?file=65929#file65929line1452>
bq. >
bq. > Line is 88 chars long.
bq. > If ExecutorService is imported, the line should be much shorter.
We can't do this since there is already an ExecutorService (from hbase)
imported. I can't use the hbase ExecutorService because it doesn't fit.
bq. On 2011-12-29 19:02:17, Ted Yu wrote:
bq. > src/test/java/org/apache/hadoop/hbase/master/TestMasterRecovery.java,
line 50
bq. > <https://reviews.apache.org/r/3323/diff/3/?file=65930#file65930line50>
bq. >
bq. > Since the test doesn't involve standby master, I think we should use
a different name.
There is test case called TestMasterFailover which involves standby master.
That's why I called it TestMasterRecovery. How about
TestMasterZKSessionRecovery?
- Jimmy
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/3323/#review4148
-----------------------------------------------------------
On 2011-12-29 18:38:54, Jimmy Xiang wrote:
bq.
bq. -----------------------------------------------------------
bq. This is an automatically generated e-mail. To reply, visit:
bq. https://reviews.apache.org/r/3323/
bq. -----------------------------------------------------------
bq.
bq. (Updated 2011-12-29 18:38:54)
bq.
bq.
bq. Review request for hbase, Ted Yu and Michael Stack.
bq.
bq.
bq. Summary
bq. -------
bq.
bq. Per discussion with Ted (on issues), I put up a patch to run
tryRecoveringExpiredZKSession() in a separate thread and time it out and fail
the recovery if it is stuck somewhere.
bq.
bq. I added a test to test the abort method. However, for the mini cluster,
becomeActiveMaster() doesn't succeed so the abort method ends up always
aborted. So the actually success recovery is not tested.
bq.
bq.
bq. This addresses bug HBASE-5099.
bq. https://issues.apache.org/jira/browse/HBASE-5099
bq.
bq.
bq. Diffs
bq. -----
bq.
bq. src/main/java/org/apache/hadoop/hbase/master/HMaster.java a5935a6
bq. src/test/java/org/apache/hadoop/hbase/master/TestMasterRecovery.java
PRE-CREATION
bq.
bq. Diff: https://reviews.apache.org/r/3323/diff
bq.
bq.
bq. Testing
bq. -------
bq.
bq. mvn -PlocalTests -Dtest=TestMaster* clean test
bq.
bq.
bq. Thanks,
bq.
bq. Jimmy
bq.
bq.
> ZK event thread waiting for root region while server shutdown handler waiting
> for event thread to finish distributed log splitting to recover the region
> sever the root region is on
> ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: HBASE-5099
> URL: https://issues.apache.org/jira/browse/HBASE-5099
> Project: HBase
> Issue Type: Bug
> Affects Versions: 0.92.0, 0.94.0
> Reporter: Jimmy Xiang
> Assignee: Jimmy Xiang
> Attachments: ZK-event-thread-waiting-for-root.png,
> distributed-log-splitting-hangs.png, hbase-5099-v2.patch,
> hbase-5099-v3.patch, hbase-5099.patch
>
>
> A RS died. The ServerShutdownHandler kicked in and started the logspliting.
> SpliLogManager
> installed the tasks asynchronously, then started to wait for them to complete.
> The task znodes were not created actually. The requests were just queued.
> At this time, the zookeeper connection expired. HMaster tried to recover the
> expired ZK session.
> During the recovery, a new zookeeper connection was created. However, this
> master became the
> new master again. It tried to assign root and meta.
> Because the dead RS got the old root region, the master needs to wait for the
> log splitting to complete.
> This waiting holds the zookeeper event thread. So the async create split
> task is never retried since
> there is only one event thread, which is waiting for the root region assigned.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
