[
https://issues.apache.org/jira/browse/HBASE-19633?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16305269#comment-16305269
]
Duo Zhang commented on HBASE-19633:
-----------------------------------
[~zghaobac] FYI. I think one more thing is that we need to revisit the order at
RS side to confirm that after refresh succeed there will be no moving nor
creating for the given peer any more.
> Clean up the replication queues in the postPeerModification stage when
> removing a peer
> --------------------------------------------------------------------------------------
>
> Key: HBASE-19633
> URL: https://issues.apache.org/jira/browse/HBASE-19633
> Project: HBase
> Issue Type: Sub-task
> Components: proc-v2, Replication
> Reporter: Duo Zhang
> Assignee: Duo Zhang
> Attachments: HBASE-19633-HBASE-19397-v1.patch,
> HBASE-19633-HBASE-19397.patch
>
>
> In the previous implementation, we can not always cleanly remove all the
> replication queues when removing a peer since the removing work is done by RS
> and if an RS is crashed then some queues may left there forever. That's why
> we need to check if there are already some queues for a newly created peer
> since we may reuse the peer id and causes problem.
> With the new procedure based replication peer modification, I think we can do
> it cleanly. After the RefreshPeerProcedures are done on all RSes, we can make
> sure that no RS will create queue for this peer again, then we can iterate
> over all the queues for all Rses and do another round of clean up.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
