[
https://issues.apache.org/jira/browse/HBASE-24924?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17181740#comment-17181740
]
Duo Zhang commented on HBASE-24924:
-----------------------------------
So the corruption is from outside our control, this means there is no simple
way for us to determine how to fix it.
Please see the discussions on this PR
https://github.com/apache/hbase/pull/2237
In general, I think we could add a check in InitMetaProcedure to not delete the
meta region with data in it, and fail the initialization of master. But I do
not think we should 'recover' automatically. There is something wrong, only
human what is the real problem and use HBCK2 to fix it.
> stuck InitMetaProcedure in finishActiveMasterInitialization
> -----------------------------------------------------------
>
> Key: HBASE-24924
> URL: https://issues.apache.org/jira/browse/HBASE-24924
> Project: HBase
> Issue Type: Bug
> Components: master
> Affects Versions: 3.0.0-alpha-1, 2.2.3
> Reporter: Bo Cui
> Assignee: Bo Cui
> Priority: Major
> Attachments: image-2020-08-21-09-12-33-894.png, masterLog.gif
>
>
> if procWAL has InitMetaProcedure, and InitMetaProcedure state is finished,
> and meta znode does not exist， finishActiveMasterInitialization will stuck
> because during startup，If InitMetaProcedure exists, InitMetaProcedure
> recreates a new CountDownLatch.
> master jstack
> !image-2020-08-21-09-12-33-894.png!
> master log
> !masterLog.gif!
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
