[
https://issues.apache.org/jira/browse/HADOOP-3877?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12620140#action_12620140
]
Allen Wittenauer commented on HADOOP-3877:
------------------------------------------
While it is clear that the name node should be a bit more protective (what
happens when edits get processed and that overflows?), but data loss can be
prevented in the case of a single failure by having multiple name node
directories listed in the config...
> disk full
> ---------
>
> Key: HADOOP-3877
> URL: https://issues.apache.org/jira/browse/HADOOP-3877
> Project: Hadoop Core
> Issue Type: Bug
> Components: dfs
> Affects Versions: 0.16.4
> Environment: ubuntu linux
> Reporter: Torsten Curdt
> Priority: Critical
>
> On our development cluster the namenode's disk filled up. This corrupted
> both: the fsimage and edits file. This should never happen!
> The files were truncated and I was able to fix them with a hex editor. Of
> course some data must have been lost.
> (see also possibly related https://issues.apache.org/jira/browse/HADOOP-2550 )
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
