hadoop-yarn-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Xuan Gong (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (YARN-275) Make NodeManagers to NOT blindly heartbeat irrespective of whether previous heartbeat is processed or not.
Date Thu, 20 Dec 2012 06:13:15 GMT

     [ https://issues.apache.org/jira/browse/YARN-275?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Xuan Gong updated YARN-275:
---------------------------

    Attachment: YARN-270.1.patch

1.In RMNode,store the NodeUpdatedDone in a field and make the scheduler set the status of
the event as processed once it is done processing
2.Then in RTS, before it sends another status update to RMNode, check if RMNode's last update
was processed, if not,ask the NM to back off
3.implemented the back off by sending the next heartbeat interval to remote NMs.
Originally NodeStatusUpdater pings every hard coded 1 second, now change it so that the next
heartbeat interval comes from RM (now hard code as 5S. Maybe we need find another way to do
it)
                
> Make NodeManagers to NOT blindly heartbeat irrespective of whether previous heartbeat
is processed or not.
> ----------------------------------------------------------------------------------------------------------
>
>                 Key: YARN-275
>                 URL: https://issues.apache.org/jira/browse/YARN-275
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>          Components: nodemanager, resourcemanager
>            Reporter: Vinod Kumar Vavilapalli
>            Assignee: Xuan Gong
>         Attachments: YARN-270.1.patch
>
>
> We need NMs to back off. The event handler mechanism is very scalable but not infinitely
so :)

--
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