[
https://issues.apache.org/jira/browse/SPARK-13433?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15157050#comment-15157050
]
lichenglin commented on SPARK-13433:
------------------------------------
It's something like deadlock.
driver use all cores> application has no cores to be finished>the driver won't
be finish>driver keep the core>driver use all cores;
if the application won't be finished, the driver won't be finished too.
I mean that We should not let the driver use all the cores,
to make sure that the application have at least one core to do it's job.
> The standalone server should limit the count of cores and memory for
> running Drivers
> --------------------------------------------------------------------------------------
>
> Key: SPARK-13433
> URL: https://issues.apache.org/jira/browse/SPARK-13433
> Project: Spark
> Issue Type: Improvement
> Components: Scheduler
> Affects Versions: 1.6.0
> Reporter: lichenglin
>
> I have a 16 cores cluster.
> A Running driver at least use 1 core may be more.
> When I submit a lot of job to the standalone server in cluster mode.
> all the cores may be used for running driver,
> and then there is no cores to run applications
> The server is stuck.
> So I think we should limit the resources(cores and memory) for running driver.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
