[
https://issues.apache.org/jira/browse/HBASE-26538?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17453421#comment-17453421
]
Duo Zhang commented on HBASE-26538:
-----------------------------------
OK, finally I found that, the problem is SCP. In SCP, we have a claim queue
step for transfering the replication queue of a dead server to other region
servers. And now, since the replication peer region_replica_replication is
gone, on region server side we will just ignore the request. But no one will
actually delete the queue, the SCP will retry forever.
I think we could add a logic in SCP, to try to delete the replication queue.
Will add comments to explain why these queues can not be deleted in normal case.
> Should find a way to clear the replication queue for a legacy
> region_replica_replication peer
> ---------------------------------------------------------------------------------------------
>
> Key: HBASE-26538
> URL: https://issues.apache.org/jira/browse/HBASE-26538
> Project: HBase
> Issue Type: Sub-task
> Components: read replicas, Replication
> Reporter: Duo Zhang
> Priority: Major
>
> When rolling upgrading, we will delete the legacy region_replica_replication
> peer. But since the old region servers still use this peer for replicating,
> we can not delete all the replication queues.
> We need to find a way to deal with these legacy replication queues after
> upgrading.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
