[
https://issues.apache.org/jira/browse/MESOS-2769?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14560091#comment-14560091
]
Ian Downes commented on MESOS-2769:
-----------------------------------
It'll use the perf_event cgroup to profile all contained processes/tasks which,
in the MesosContainerizer, is the same for all cgroups for the container. I
expect the actual code to do the aggregation and statistics will be separate to
the isolator so it could easily be re-used.
Yes, the plan is to expose through ResourceStatistics.
We're working on test code for this and determining if the overhead is
sufficiently low to support optionally enabling this continuously.
> Metric for cpu scheduling latency from all components
> -----------------------------------------------------
>
> Key: MESOS-2769
> URL: https://issues.apache.org/jira/browse/MESOS-2769
> Project: Mesos
> Issue Type: Improvement
> Components: isolation
> Affects Versions: 0.22.1
> Reporter: Ian Downes
> Assignee: Cong Wang
> Labels: twitter
>
> The metric will provide statistics on the scheduling latency for
> processes/threads in a container, i.e., statistics on the delay before
> application code can run. This will be the aggregate effect of the normal
> scheduling period, contention from other threads/processes, both in the
> container and on the system, and any effects from the CFS bandwidth control
> (if enabled) or other CPU isolation strategies.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
