[
https://issues.apache.org/jira/browse/HBASE-5745?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13263130#comment-13263130
]
Uma Maheswara Rao G commented on HBASE-5745:
--------------------------------------------
I have just committed the HDFS issue in trunk and branch-2.
For supporting older released versions, we may have to implement the similar
logic in Hbase SFLR, to ensure the correct length.Otherwise this is a dataloss
case from Hbase perspective.
> SequenceFileLogReader#getPos may get wrong length on DFS restart
> ----------------------------------------------------------------
>
> Key: HBASE-5745
> URL: https://issues.apache.org/jira/browse/HBASE-5745
> Project: HBase
> Issue Type: Bug
> Components: wal
> Affects Versions: 0.96.0
> Reporter: Uma Maheswara Rao G
> Assignee: Uma Maheswara Rao G
> Priority: Critical
>
> This is actually a kind of integration bug from Hbase perspective.
> Currently HDFS will count the partial block length as 0, if there are no
> locations found for the partial block. This can happend opn DFS restart,
> before DNs completely reports to NN.
> Explained the scenario in HDFS-3222. Actually this is a bug in HDFS. we may
> solve this in latest versions.
> So, whatever the versions Hbase using may have this bug. HMaster may not be
> able to replay the complete edits if there is an Hmaster switch also at the
> same time.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
