hadoop-mapreduce-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Siddharth Seth (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (MAPREDUCE-5384) Races in DelegationTokenRenewal
Date Tue, 20 Aug 2013 21:01:52 GMT

    [ https://issues.apache.org/jira/browse/MAPREDUCE-5384?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13745421#comment-13745421
] 

Siddharth Seth commented on MAPREDUCE-5384:
-------------------------------------------

Karthik, apologies, I haven't been able to get to this earlier.

With the latest patch, I'm not sure which race is being fixed. It looks like a cancel while
a RenewalTimer is running, will still lead to an additional Renewal being scheduled for the
same token, which is the same behaviour without the patch. Have some concerns with synchronization
/ thread safety in the patch as well - leaving those out.

Instead of the relatively large changes in the patch, I think it'll be a lot simpler to just
associate a 'cancel'/'intentToCancel' flag with the token to prevent renewal attempts after
a cancel is called. The in-process renew can check these till the last moment before invoking
the actual renew, and subsequent renewals will not attempt a renew (maybe even not schedule
a renew). The changes to the unit test will likely still be required - to allow the one extra
in-flight renew.

Do you know if this problem exists in the 2.x renewer as well ?
                
> Races in DelegationTokenRenewal
> -------------------------------
>
>                 Key: MAPREDUCE-5384
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-5384
>             Project: Hadoop Map/Reduce
>          Issue Type: Bug
>    Affects Versions: 1.2.0, 1.1.2, 1.2.1
>            Reporter: Karthik Kambatla
>            Assignee: Karthik Kambatla
>         Attachments: mr-5384-0.patch, mr-5384-1.patch, mr-5384-2.patch
>
>
> There are a couple of races in DelegationTokenRenewal. 
> One of them was addressed by MAPREDUCE-4860, which introduced a deadlock while fixing
this race. Opening a new JIRA per discussion in MAPREDUCE-5364, since MAPREDUCE-4860 is already
shipped in a release.
> Races to fix:
> # TimerTask#cancel() disallows future invocations of run(), but doesn't abort an already
scheduled/started run().
> # In the context of DelegationTokenRenewal, RenewalTimerTask#cancel() only cancels that
TimerTask instance. However, it has no effect on any other TimerTasks created for that token.


--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Mime
View raw message