[jira] Commented: (HBASE-11) [hbase] Unexpected exits corrupt DFS

["Jim Kellerman (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HBASE-11?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12588612#action_12588612 ] 

Jim Kellerman commented on HBASE-11:
------------------------------------

Although HLog.splitLog now properly skips zero length files (and they are all deleted after the log(s) are split, it probably does not handle empty files well (files that were properly closed but have no content - these are not zero length because of the metadata written to the beginning of the file. In this case, the reader will throw an eof exception, which will issue a warning about possible data loss. We should explicitly check for eof exception before generic IOException and ignore it.

As for improperly closed files, there's not much we can do about that until we have (at least a subset of) HADOOP-1700.

> [hbase] Unexpected exits corrupt DFS
> ------------------------------------
>
>                 Key: HBASE-11
>                 URL: https://issues.apache.org/jira/browse/HBASE-11
>             Project: Hadoop HBase
>          Issue Type: Bug
>          Components: regionserver
>            Reporter: Bryan Duxbury
>            Assignee: Jim Kellerman
>
> When a RegionServer exits unexpectedly, it often leaves its DFS files open. In the case of the redo log, this can result in a zero-length file, which a newly started RegionServer will be unable to read, causing it to exit again. It also causes DFS corruption, requiring the admin to run a dfs fsck -delete before HBase can be restarted.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.