mesos-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ben Mahler" <benjamin.mah...@gmail.com>
Subject Re: Review Request: Initial support for CPU limits via CFS Bandwidth Control.
Date Tue, 19 Feb 2013 22:24:34 GMT


> On Feb. 19, 2013, 7:38 p.m., Ben Mahler wrote:
> >

On second thought, after discussing this with ben and vinod. Let's add a flag
to enable CFS bandwidth control. This is going to have to be the short term way
we roll this out.

Once you add the flag, then we can fail fast in the cgroups isolation module if it
doesn't exist. There's also no more need to check for existence later when trying
to set the control.

Sound good?


- Ben


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/9464/#review16744
-----------------------------------------------------------


On Feb. 18, 2013, 11:51 p.m., David Mackey wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/9464/
> -----------------------------------------------------------
> 
> (Updated Feb. 18, 2013, 11:51 p.m.)
> 
> 
> Review request for mesos, Benjamin Hindman, Vinod Kone, and Ben Mahler.
> 
> 
> Description
> -------
> 
> Initial support for CPU "hard limits" via CFS Bandwidth Control in cgroups.
> 
> CFS is unique relative to existing Mesos cgroups support in that it is a "subfeature"
of an already supported cgroups subsystem, cpu. Also, there are two "tunables" for configuring
CFS bandwidth limiting.
> 
> There are 4 approaches one could take:
> 1) Use the CFS bandwidth limiting if the feature is present
> 2) Expose as separate flag, eg "cpu,cfs,memory,freezer"
> 3) Add feature flag support to subsystems via an additional delimiter, eg "cpu+cfs,memory,freezer".

> 4) Add an additional control flag via some other means
> 
> Option 2's downside breaks the 1:1 mapping between cgroups subsystems and a cgroups resource
flag.
> Option 3's downside is it greatly increases complexity of parsing cgroups subsystem flags.
> 
> This diff takes option 1. 
> 
> 
> This addresses bug MESOS-315.
>     https://issues.apache.org/jira/browse/MESOS-315
> 
> 
> Diffs
> -----
> 
>   src/slave/cgroups_isolation_module.cpp 14f549e 
> 
> Diff: https://reviews.apache.org/r/9464/diff/
> 
> 
> Testing
> -------
> 
> make check + additional testing
> 
> 
> Thanks,
> 
> David Mackey
> 
>


Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message