[
https://issues.apache.org/jira/browse/HBASE-12136?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14154490#comment-14154490
]
Virag Kothari commented on HBASE-12136:
---------------------------------------
If useMulti is set to true, then the operations of adding a peer znode and its
children such as peer -state and tableCf will be atomic making the server
change redundant. However if the useMulti is not configured, then we need to
have a nodeCreated() event on the tracker so that even on race condition, the
nodeCreated() will make sure to refresh the tableCf config.
The TestPerTableCFReplication.testPerTableCFReplication() should no longer be
flakey.
> Race condition between client adding tableCF replication znode and server
> triggering TableCFsTracker
> -----------------------------------------------------------------------------------------------------
>
> Key: HBASE-12136
> URL: https://issues.apache.org/jira/browse/HBASE-12136
> Project: HBase
> Issue Type: Bug
> Reporter: Virag Kothari
> Assignee: Virag Kothari
> Attachments: HBASE-12136.patch
>
>
> In ReplicationPeersZKImpl.addPeer(), there is a race between client creating
> tableCf znode and the server triggering TableCFsTracker. If the server wins,
> it wont be able to read the data set on tableCF znode and replication will
> be misconfigured
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
