[
https://issues.apache.org/jira/browse/HBASE-21465?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16685971#comment-16685971
]
Duo Zhang commented on HBASE-21465:
-----------------------------------
Or maybe the v4 is better, where we do not change the logic of writing sequence
id file. FWIW, we always need the logic in v4 patch even if we go v3, as we
need to consider rolling upgrading, where the region servers in old version
still run the old code, where we write sequence id file before writing open
marker.
> Retry on reportRegionStateTransition can lead to unexpected errors
> ------------------------------------------------------------------
>
> Key: HBASE-21465
> URL: https://issues.apache.org/jira/browse/HBASE-21465
> Project: HBase
> Issue Type: Sub-task
> Components: amv2
> Reporter: Duo Zhang
> Assignee: Duo Zhang
> Priority: Major
> Fix For: 3.0.0, 2.2.0
>
> Attachments: HBASE-21465-UT.patch, HBASE-21465-v1.patch,
> HBASE-21465-v2.patch, HBASE-21465-v3.patch, HBASE-21465-v4.patch,
> HBASE-21465.patch
>
>
> It is possible that the reportRegionStateTransition method is succeeded at
> master side, but before returning the the region server, the rpc connection
> is broken, or the master restarts. So next when the region server try
> again,we will find that the state for the region and the TRSP is not correct,
> and can lead to a RS abort or something even worse.
> We should be able to determine whether a reportRegionStateTransition call is
> just a retry and has already been succeeded, and just ignore it.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
