[
https://issues.apache.org/jira/browse/HBASE-25129?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17205271#comment-17205271
]
Duo Zhang commented on HBASE-25129:
-----------------------------------
First, this is not a bug, right? It could be an improvement.
Second, things do not go as what you expect. The barrier is written when
opening a region, and there is no fencing between opening a region and adding
or modifying a replication peer, so the only safe way is to always write the
barrier when replication is enabled. Think of when you are checking if there is
a serial replication peer(and maybe you also need to test whether the peer
includes this table?), another one is adding a new serial replication peer.
The design here is to clear the barrier by a background chore. And what is the
real problem here for you? It generates too many edits and slow down the region
replication? IIRC, in the design doc, you said that these families will not be
replicated?
Thanks.
> serial replication, addReplicationBarrier is writing to rep_barrier family
> even there is no serial replication peer.
> --------------------------------------------------------------------------------------------------------------------
>
> Key: HBASE-25129
> URL: https://issues.apache.org/jira/browse/HBASE-25129
> Project: HBase
> Issue Type: Bug
> Reporter: Huaxiang Sun
> Assignee: Huaxiang Sun
> Priority: Major
>
> We found that there are quite some data in rep_barrier family even there is
> no serial replication enabled. Checked the code, it is checking if table has
> replication enabled. Think there is another check needed (i.e, is there any
> serial replication peers configured).
> [https://github.com/apache/hbase/blob/master/hbase-server/src/main/java/org/apache/hadoop/hbase/master/assignment/RegionStateStore.java#L215]
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
