You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@mesos.apache.org by "sujz.buaa" <su...@qq.com> on 2015/12/11 15:52:40 UTC
Doubt on implementation of collecting container resource usage
Hi, all:
I find that periodical collection of resource usage is implemented by reading '/proc/pid/stat' file once per second, instead of running htop command in shell and parsing output of it, and currently it seems that mesos can only monitor usage of per-process, necessary extention must be made to achieve fine-grained monitoring of each task if wanted, however, we can get usage from htop output directly no matter for a process or task.
Could anybody explain why reading '/proc/pid/stat' file scheme is selected, is there some reasons?
Thank you and best regards.
Re: Doubt on implementation of collecting container resource usage
Posted by zhou weitao <zh...@gmail.com>.
htop is an extension to OS, or htop is too high to use for Mesos, IMO.
2015-12-11 22:52 GMT+08:00 sujz.buaa <su...@qq.com>:
> Hi, all:
>
> I find that periodical collection of resource usage is implemented by
> reading '/proc/pid/stat' file once per second, instead of running htop
> command in shell and parsing output of it, and currently it seems that
> mesos can only monitor usage of per-process, necessary extention must be
> made to achieve fine-grained monitoring of each task if wanted, however, we
> can get usage from htop output directly no matter for a process or task.
>
> Could anybody explain why reading '/proc/pid/stat' file scheme is
> selected, is there some reasons?
>
> Thank you and best regards.
>
>
--
-------------------------------------------------------------
周伟涛
blog: http://vitan.github.io/
github: https://github.com/vitan
LinkedIn: https://cn.linkedin.com/in/zhouweitao
Re: Doubt on implementation of collecting container resource usage
Posted by tommy xiao <xi...@gmail.com>.
sysdig
2015-12-11 22:52 GMT+08:00 sujz.buaa <su...@qq.com>:
> Hi, all:
>
> I find that periodical collection of resource usage is implemented by
> reading '/proc/pid/stat' file once per second, instead of running htop
> command in shell and parsing output of it, and currently it seems that
> mesos can only monitor usage of per-process, necessary extention must be
> made to achieve fine-grained monitoring of each task if wanted, however, we
> can get usage from htop output directly no matter for a process or task.
>
> Could anybody explain why reading '/proc/pid/stat' file scheme is
> selected, is there some reasons?
>
> Thank you and best regards.
>
>
--
Deshi Xiao
Twitter: xds2000
E-mail: xiaods(AT)gmail.com