> On Dec. 15, 2015, 7:36 p.m., Ian Downes wrote:
> > Initial CFS parameters should be specified to Docker using the appropriate 
> > flags, not tacked onto the end of launch() where we don't yet know the 
> > cgroup. Subsequent updates done by Mesos in update().

As per our offline conversation, in this code path by the time the code is 
called we do have the executor PID, and the executor is in the correct cgroup 
(eg inside the docker container).  I really would rather have a single way to 
set/update the CFS parameters, much like how the cgroup isolator calls 
update(...) from its prepare(...) method.

However, I agree the other case does need to be addressed, either in this 
review or a follow up, where tasks launched with the docker executor still 
don't get their CFS parameters updated.


- Steve


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/33174/#review110547
-----------------------------------------------------------


On Dec. 15, 2015, 8:14 p.m., Steve Niemitz wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/33174/
> -----------------------------------------------------------
> 
> (Updated Dec. 15, 2015, 8:14 p.m.)
> 
> 
> Review request for mesos, Ian Downes, Jie Yu, and Timothy Chen.
> 
> 
> Bugs: MESOS-2617
>     https://issues.apache.org/jira/browse/MESOS-2617
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> Fix for docker containerizer not configuring CFS quotas correctly.
> 
> It would be nice to refactor all this isolation code in a way that can be 
> shared between all containerizers, as this is basically just copied from the 
> CgroupsCpushareIsolator, but that's a much bigger undertaking.
> 
> 
> Diffs
> -----
> 
>   src/slave/containerizer/docker.cpp 2d2dd4e0df36207c5f3cbb4fe2c50df51c0f3e9e 
> 
> Diff: https://reviews.apache.org/r/33174/diff/
> 
> 
> Testing
> -------
> 
> 
> Thanks,
> 
> Steve Niemitz
> 
>

Reply via email to