[jira] [Commented] (MESOS-375) Keep the stats for tasks around after the tasks are archived

["Benjamin Mahler (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MESOS-375?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13591919#comment-13591919 ] 

Benjamin Mahler commented on MESOS-375:
---------------------------------------

If a task finishes within a second, the current implementation of the resource monitoring will not have _any_ data associated with it. This is because we monitor resource usage every 5 seconds (by default). This can be changed with --resource_monitoring_interval on the slave if finer granularity is desired.

As for keeping the stats around, the time series are currently stored inside the statistics module. It's only accessible via /statistics/series.json, which is clunky at the moment as it requires the knowledge of the framework id and executor id.

E.g. localhost:5051/statistics/series.json?context=monitor&name=<framework_id>/<executor_id>/cpu_usage

Adding a nicer time-series style endpoint for usage is possible, but requires keeping around ExecutorInfo structs for completed executors.
                
> Keep the stats for tasks around after the tasks are archived
> ------------------------------------------------------------
>
>                 Key: MESOS-375
>                 URL: https://issues.apache.org/jira/browse/MESOS-375
>             Project: Mesos
>          Issue Type: Bug
>            Reporter: Florian Leibert (flo)
>            Priority: Blocker
>
> We would like to ship chronos with task stats, however many of our tasks finish within a second or so and thus requests to <slave>/monitor/usage.json often no longer show the task. 
> Could we get these stats for a variable window even after the task got scheduled? 

--
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