[
https://issues.apache.org/jira/browse/HBASE-20830?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16833285#comment-16833285
]
Duo Zhang commented on HBASE-20830:
-----------------------------------
The problem is that, when implementing AMv2, no one had take enough attention
on the region replica feature, so we are not sure whether the region replica
feature is still stable enough after we completely rewrite the assignment
logic. And AFAIK, no big users have deployed 2.x cluster with region replica
enabled seriously in production, so if you have the ability to modify the
source code of HBase, you can try to use the region replica feature and report
back the problems and we can work together to improve it, if not, I do not
suggest you enable it as it may cause the cluster to stuck and not easy to
recover...
Sorry about this...
> Document that region replica does not work well with AMv2
> ---------------------------------------------------------
>
> Key: HBASE-20830
> URL: https://issues.apache.org/jira/browse/HBASE-20830
> Project: HBase
> Issue Type: Sub-task
> Components: documentation
> Reporter: Duo Zhang
> Assignee: Duo Zhang
> Priority: Major
> Fix For: 3.0.0
>
> Attachments: HBASE-20830.patch
>
>
> As we can see lots of TODOs for region replica feature in the
> AssignmentManager' code. And also, now we rely on the openSeqNum to determine
> whether a region has been successfully reopened, but the openSeqNum for a
> non-default replica region seems unstable...
> It is too late for the 2.1.0 release so let's just add a note in ref guide to
> say that region replica does not work well with AMv2 first, and then start to
> make it stable in follow on minor or major release lines.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
