You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Till Rohrmann (Jira)" <ji...@apache.org> on 2020/07/27 07:37:00 UTC
[jira] [Commented] (FLINK-18715) add cpu usage metric of
jobmanager/taskmanager
[ https://issues.apache.org/jira/browse/FLINK-18715?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17165503#comment-17165503 ]
Till Rohrmann commented on FLINK-18715:
---------------------------------------
Thanks for creating this ticket [~1026688210]. Isn't the CPU usage something you already get from K8s or other resource management systems? The problem I see with Flink trying to measure the CPU usage is that it neglects other processes running on the same machine. This could lead to Flink reporting a very low CPU usage even though the machine on which it is running is super busy.
> add cpu usage metric of jobmanager/taskmanager
> -------------------------------------------------
>
> Key: FLINK-18715
> URL: https://issues.apache.org/jira/browse/FLINK-18715
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Metrics
> Affects Versions: 1.11.1
> Reporter: wgcn
> Priority: Major
> Fix For: 1.12.0, 1.11.2
>
>
> flink process add cpu usage metric, user can determine that their job is io bound /cpu bound ,so that they can increase/decrese cpu core in the container (k8s,yarn). If it's nessary
> . you can assign it to me ,I come up with a idea calculating cpu usage ratio using ManagementFactory.getRuntimeMXBean().getUptime() and ManagementFactory.getOperatingSystemMXBean().getProcessCpuTime over a period of time . it can get a value in single cpu core environment. and user can use the value to calculate cpu usage ratio by dividing num of container's cpu core.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)