hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Eli Collins (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-3048) Small race in BlockManager#close
Date Wed, 15 Aug 2012 22:30:38 GMT

    [ https://issues.apache.org/jira/browse/HDFS-3048?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13435578#comment-13435578
] 

Eli Collins commented on HDFS-3048:
-----------------------------------

Forgot to mention, per Todd's comment above ("I seem to remember trying to fix this once,
but ran into a deadlock issue with the join() call.") I looked and I don't see a deadlock
issue with the join call, eg I don't see how it could be blocked on the BlockManager#close
path that's waiting on it. Because the DN doesn't catch sig STOP we don't actually run the
shutdown path in normal execution.
                
> Small race in BlockManager#close
> --------------------------------
>
>                 Key: HDFS-3048
>                 URL: https://issues.apache.org/jira/browse/HDFS-3048
>             Project: Hadoop HDFS
>          Issue Type: Bug
>          Components: name-node
>    Affects Versions: 2.0.0-alpha
>            Reporter: Eli Collins
>            Assignee: Andy Isaacson
>         Attachments: hdfs-3048.txt, hdfs-3048.txt, hdfs-3787-2.txt
>
>
> There's a small race in BlockManager#close, we close the BlocksMap before the replication
monitor, which means the replication monitor can NPE if it tries to access the blocks map.
We need to swap the order (close the blocks map after shutting down the repl monitor).

--
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

        

Mime
View raw message