[
https://issues.apache.org/jira/browse/HBASE-20117?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16974196#comment-16974196
]
Duo Zhang commented on HBASE-20117:
-----------------------------------
Maybe the problem is here.
{code}
if (peerIds.isEmpty()) {
// no serial replication, only keep the newest barrier
Cell cell =
result.getColumnLatestCell(HConstants.REPLICATION_BARRIER_FAMILY,
HConstants.SEQNUM_QUALIFIER);
metaTable.delete(new
Delete(regionName).addFamily(HConstants.REPLICATION_BARRIER_FAMILY,
cell.getTimestamp() - 1));
cleanedRows++;
deletedBarriers += barriers.length - 1;
continue;
}
{code}
If there are no serial peers we will always keep the newest barrier, but here
we do not check whether the region is gone or not.
Let me check if I can provide a UT for this.
But anyway, [~stack] I think HBCK should just ignore this family? Replication
is asynchoronous, so the barrier family will remain there for a while after the
info family is gone.
> Cleanup the unused replication barriers in meta table
> -----------------------------------------------------
>
> Key: HBASE-20117
> URL: https://issues.apache.org/jira/browse/HBASE-20117
> Project: HBase
> Issue Type: Sub-task
> Components: master, Replication
> Reporter: Duo Zhang
> Assignee: Duo Zhang
> Priority: Major
> Fix For: 3.0.0, 2.1.0
>
> Attachments: HBASE-20117-v1.patch, HBASE-20117-v2.patch,
> HBASE-20117.patch
>
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
