[
https://issues.apache.org/jira/browse/HADOOP-5729?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12702187#action_12702187
]
Wang Xu commented on HADOOP-5729:
---------------------------------
I only did it on 0.19 branch...
AFAIK processIOError() intends to deal with errors of editStreams
and remove bad ones. But in open(), no editStream should
exists in advance. Thus I think it should be handled here.
If no editStreams opened here, NameNode will still keep alive,
but logSync() cannot written anything and cause no exception.
I think if editStreams is empty, it should be dealt with at once,
ether in open() or in processIOError().
> FSEditLog.open should stop going on if cannot open any directory
> ----------------------------------------------------------------
>
> Key: HADOOP-5729
> URL: https://issues.apache.org/jira/browse/HADOOP-5729
> Project: Hadoop Core
> Issue Type: Bug
> Components: dfs
> Affects Versions: 0.19.1
> Environment: CentOS 5.2, jdk 1.6, hadoop 0.19.1
> Reporter: Wang Xu
> Assignee: Wang Xu
> Fix For: 0.19.2
>
> Attachments: fseditlog-open.patch
>
> Original Estimate: 1h
> Remaining Estimate: 1h
>
> FSEditLog.open will be invoked when SecondaryNameNode doCheckPoint,
> If no dir is opened successfully, it only prints some WARN messages in log,
> and goes on running.
> However, it causes the editStreams becomes empty and cannot by synced
> in. And if editStreams were decreased to 0 when exceptions occured during
> logsync, NameNode would print FATAL log message and halt itself. Hence,
> we think it should also stopped itself at that time.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
