[
https://issues.apache.org/jira/browse/HBASE-15424?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15185245#comment-15185245
]
Ashish Singhi commented on HBASE-15424:
---------------------------------------
When wal write fails (then any way we will not be able replicate the hfile) we
will not be adding the hfile-refs node in zk.
But there is another existing issue when the hfile is successfully loaded but
writing to wal is failed so in this case source cluster will have the data but
peer will not have. IMO we can discuss this in a separate jira.
> Add bulk load hfile-refs for replication in ZK after the event is appended in
> the WAL
> -------------------------------------------------------------------------------------
>
> Key: HBASE-15424
> URL: https://issues.apache.org/jira/browse/HBASE-15424
> Project: HBase
> Issue Type: Bug
> Components: Replication
> Affects Versions: 2.0.0, 1.3.0, 1.4.0
> Reporter: Ashish Singhi
> Assignee: Ashish Singhi
> Priority: Minor
>
> Currenlty hfile-refs znode used for tracking the bulk loaded data replication
> is added first and then the bulk load event in appended in the WAL. So this
> may lead to a issue where the znode is added in ZK but append to WAL is
> failed(due to some probelm in DN), so this znode will be left in ZK as it is
> and will not allow hfile to get deleted from archive directory.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
