You are viewing a plain text version of this content. The canonical link for it is here.
Posted to common-dev@hadoop.apache.org by "Arun C Murthy (JIRA)" <ji...@apache.org> on 2008/07/29 07:00:32 UTC

[jira] Created: (HADOOP-3849) MapEventsFetcherThread doesn't wait for the heartbeat-interval if it doesn't have events to fetch

MapEventsFetcherThread doesn't wait for the heartbeat-interval if it doesn't have events to fetch
-------------------------------------------------------------------------------------------------

                 Key: HADOOP-3849
                 URL: https://issues.apache.org/jira/browse/HADOOP-3849
             Project: Hadoop Core
          Issue Type: Bug
          Components: mapred
    Affects Versions: 0.18.0
            Reporter: Arun C Murthy
            Assignee: Arun C Murthy
             Fix For: 0.19.0


The 'notify' done by FetchStatus.getMapEvents causes the MapEventsFetcherThread to immediately run to the JobTracker (getTaskCompletionEvents). 

On a 3500 node cluster, I saw that each TaskTracker calls JobTracker.getTaskCompletionEvents multiple times per-second. This caused the JobTracker's RPC queues to back-up resulting in each RPC spending more than 120s in the queue - leading to shuffle proceeding very very slowly.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

[jira] Resolved: (HADOOP-3849) MapEventsFetcherThread doesn't wait for the heartbeat-interval if it doesn't have events to fetch

Posted by "Arun C Murthy (JIRA)" <ji...@apache.org>.
     [ https://issues.apache.org/jira/browse/HADOOP-3849?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Arun C Murthy resolved HADOOP-3849.
-----------------------------------

    Resolution: Invalid

Invalid, FetchStatus will regulate itself to ensure it doesn't slam the JobTracker.

> MapEventsFetcherThread doesn't wait for the heartbeat-interval if it doesn't have events to fetch
> -------------------------------------------------------------------------------------------------
>
>                 Key: HADOOP-3849
>                 URL: https://issues.apache.org/jira/browse/HADOOP-3849
>             Project: Hadoop Core
>          Issue Type: Bug
>          Components: mapred
>    Affects Versions: 0.18.0
>            Reporter: Arun C Murthy
>            Assignee: Arun C Murthy
>             Fix For: 0.19.0
>
>
> The 'notify' done by FetchStatus.getMapEvents causes the MapEventsFetcherThread to immediately run to the JobTracker (getTaskCompletionEvents). 
> On a 3500 node cluster, I saw that each TaskTracker calls JobTracker.getTaskCompletionEvents multiple times per-second. This caused the JobTracker's RPC queues to back-up resulting in each RPC spending more than 120s in the queue - leading to shuffle proceeding very very slowly.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.