[
https://issues.apache.org/jira/browse/MESOS-2655?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14573410#comment-14573410
]
Benjamin Mahler commented on MESOS-2655:
----------------------------------------
After chatting with Vinod, the updated plan is to generalize the existing
long-lived-framework rather than introducing this special case one. If we make
the command configurable (rather than a fixed random sleep) and the task
resources configurable, it can be used for running a particular command in a
long-lived fashion against a cluster. For example, if one wishes to consume the
revocable cpus:
{noformat}
./long-lived-framework --master=[master] --command="while true; do true; done"
--task_resources="mem:32;disk:32" --task_revocable_resources="cpus:1"
{noformat}
> Implement a stand alone test framework that uses revocable cpu resources
> ------------------------------------------------------------------------
>
> Key: MESOS-2655
> URL: https://issues.apache.org/jira/browse/MESOS-2655
> Project: Mesos
> Issue Type: Task
> Reporter: Vinod Kone
> Assignee: Benjamin Mahler
> Labels: twitter
>
> Ideally this would be an example framework (or stand alone binary like load
> generator framework) that helps us evaluate oversubscription in a real
> cluster.
> We need to come up with metrics that need to be exposed by this framework for
> evaluation (e.g., how many revocable offers, rescinds, preemptions etc).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
