[jira] [Commented] (HBASE-14840) Sink cluster RS reports data replication as success though the data it is not replicated

["Ashish Singhi (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HBASE-14840?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15013155#comment-15013155 ] 

Ashish Singhi commented on HBASE-14840:
---------------------------------------

*Analysis:*
I found that if {{HRegionServer#replicationSinkHandler}} is not yet initialized and before that  replication request has come then we will simply return back to the source cluster. Source cluster assumes that the call was success as there was no exception and hence will go ahead and update the WAL position in ZK node.
{code}
  public ReplicateWALEntryResponse replicateWALEntry(final RpcController controller,
      final ReplicateWALEntryRequest request) throws ServiceException {
    try {
      if (regionServer.replicationSinkHandler != null) {
        checkOpen();
        requestCount.increment();
        List<WALEntry> entries = request.getEntryList();
        CellScanner cellScanner = ((PayloadCarryingRpcController)controller).cellScanner();
        regionServer.getRegionServerCoprocessorHost().preReplicateLogEntries(entries, cellScanner);
        regionServer.replicationSinkHandler.replicateLogEntries(entries, cellScanner);
        regionServer.getRegionServerCoprocessorHost().postReplicateLogEntries(entries, cellScanner);
      }
      return ReplicateWALEntryResponse.newBuilder().build();
    } catch (IOException ie) {
      throw new ServiceException(ie);
    }
  }
{code}

Will provide a patch fixing this in some time.

> Sink cluster RS reports data replication as success though the data it is not replicated
> ----------------------------------------------------------------------------------------
>
>                 Key: HBASE-14840
>                 URL: https://issues.apache.org/jira/browse/HBASE-14840
>             Project: HBase
>          Issue Type: Bug
>            Reporter: Y. SREENIVASULU REDDY
>            Assignee: Ashish Singhi
>
> *Scenario:*
> Sink cluster is down
> Create a table and enable table replication
> Put some data
> Now restart the sink cluster
> *Observance:*
> Data is not replicated in sink cluster but still source cluster updates the WAL log position in ZK, resulting in data loss in sink cluster.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)