[
https://issues.apache.org/jira/browse/HBASE-22074?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16799027#comment-16799027
]
Duo Zhang commented on HBASE-22074:
-----------------------------------
Let's try again.
[~stack] The basic idea here is to make use of the existing 'fence' in our
WALProcedureStore, that's why I introduced a reportTranstion method for
OpenRegionProcedure/CloseRegionProcedure. In the method we will persist the
state into the procedure store, and wake up the procedure, and then when
executing the procedure, we will update the hbase:meta. So if we are not the
active master any more, we can not write to the WALProcedureStore any more, as
the new master will call recoverLease to close our writer. This could solve the
problem in HBASE-22060.
Thanks.
> Should use procedure store to persist the state in reportRegionStateTransition
> ------------------------------------------------------------------------------
>
> Key: HBASE-22074
> URL: https://issues.apache.org/jira/browse/HBASE-22074
> Project: HBase
> Issue Type: Bug
> Components: amv2, proc-v2
> Reporter: Duo Zhang
> Assignee: Duo Zhang
> Priority: Blocker
> Fix For: 3.0.0, 2.2.0, 2.3.0
>
> Attachments: HBASE-22074-v1.patch, HBASE-22074-v2.patch,
> HBASE-22074.patch
>
>
> For now we will update the meta region directly. This may cause lots of
> problems and after a bunch of fixes, we still can not solve the problem in
> HBASE-22060.
> So maybe the approach itself is not a good choice, let's try another way to
> see if it could work better.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
