[jira] [Commented] (MESOS-9315) Adding support for implicit allocation of mandatory custom resources in Mesos

["Clément Michaud (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MESOS-9315?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16764380#comment-16764380 ] 

Clément Michaud commented on MESOS-9315:
----------------------------------------

Hello Benjamin,

I've done the patch for this feature: [https://reviews.apache.org/r/69938/.|https://reviews.apache.org/r/69938/]

I had another use case for it lately: I tried to run Spark on our Mesos cluster and since we installed the XFS isolator with hard limit enforced, every task should have a given amount of disk. However, there is currently no way to allocate disk to a Spark executor. I'm patching the framework to introduce it but in the meantime this review should temporarily solve this issue if we put let say 5GB of disk by default for every tasks that do not reserve any amount like Spark.

 

Best regards,

Clément

> Adding support for implicit allocation of mandatory custom resources in Mesos
> -----------------------------------------------------------------------------
>
>                 Key: MESOS-9315
>                 URL: https://issues.apache.org/jira/browse/MESOS-9315
>             Project: Mesos
>          Issue Type: Improvement
>            Reporter: Clément Michaud
>            Priority: Minor
>         Attachments: mesos-community-email.txt
>
>
> I sent a an email (attached) few days ago to propose the introduction of a new hook to append resources implicitly to tasks for mandatory resources. This would allow Mesos to support mandatory resources like network bandwidth or disk IO for instance. 
> In a nutshell, we propose to add a hook with the following signature
> {code:java}
> Result<Resources> masterLaunchTaskResourceDecorator(
>   const Resources& slaveResources,
>   TaskInfo& task)
> {code}
> and call it in the master in the ACCEPT message handler.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)