You are viewing a plain text version of this content. The canonical link for it is here.
Posted to mapreduce-issues@hadoop.apache.org by "Jason Lowe (JIRA)" <ji...@apache.org> on 2014/01/10 17:44:51 UTC

[jira] [Updated] (MAPREDUCE-5717) Task pings are interpreted as task progress

     [ https://issues.apache.org/jira/browse/MAPREDUCE-5717?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Jason Lowe updated MAPREDUCE-5717:
----------------------------------


Prior to MAPREDUCE-5196, tasks could ping the AM or provide a status update to the AM.  The former was a sanity check to make sure the AM is still present, while the latter was used to indicate some progress had been made on the task's end.  Therefore a task that was still pinging but taking too long to send status updates would be detected as a hung task by the AM and the task killed.

Now the AM treats both pings and status updates as progress on the tasks side, which means a task that hangs while processing data but stays alive enough to keep pinging will just keep running and the AM will never timeout the task.

> Task pings are interpreted as task progress
> -------------------------------------------
>
>                 Key: MAPREDUCE-5717
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-5717
>             Project: Hadoop Map/Reduce
>          Issue Type: Bug
>    Affects Versions: 3.0.0
>            Reporter: Jason Lowe
>




--
This message was sent by Atlassian JIRA
(v6.1.5#6160)