Re: Review Request 14043: Added memory threshold notifications for inducing OOMs.

[Eric Biederman <eb...@xmission.com>]

> On Sept. 27, 2013, 1:01 a.m., Ian Downes wrote:
> > src/slave/flags.hpp, line 183
> > <https://reviews.apache.org/r/14043/diff/2/?file=357626#file357626line183>
> >
> >     How does the kernel behave if the cgroup memory limit is higher than the resource allocation we actually want? 
> >     
> >     Specifically, will the kernel start reclaiming pages early enough to keep the memory used below the soft limit? 
> >

I looked at it quickly and in 3.4 the kernel does reclaim memory and try to keep it below the soft limit.  How successful that code is I can't say.


- Eric


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/14043/#review26425
-----------------------------------------------------------


On Sept. 27, 2013, 12:49 a.m., Ben Mahler wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/14043/
> -----------------------------------------------------------
> 
> (Updated Sept. 27, 2013, 12:49 a.m.)
> 
> 
> Review request for mesos, Benjamin Hindman, Eric Biederman, Ian Downes, David Mackey, and Vinod Kone.
> 
> 
> Bugs: MESOS-662
>     https://issues.apache.org/jira/browse/MESOS-662
> 
> 
> Repository: mesos-git
> 
> 
> Description
> -------
> 
> This adds onto tdmackey's patch here: https://reviews.apache.org/r/14024/
> 
> In addition to having the OOM killer enabled, this adds memory threshold notifications in order to induce our own OOMs upon reaching the requested memory limit. This allows us to capture the relevant memory.stat information at the time of the OOM event for the majority of OOMs.
> 
> When a process is allocating quickly enough to hit the hard limit, this will still behave correctly by allowing the OOM killer to correct the OOM. In this case, we will report memory information that excludes the offending process, but that's the best we can do for now.
> 
> 
> Diffs
> -----
> 
>   src/slave/cgroups_isolator.hpp e86062e 
>   src/slave/cgroups_isolator.cpp a1f5b32 
>   src/slave/flags.hpp ea1e4f7 
> 
> Diff: https://reviews.apache.org/r/14043/diff/
> 
> 
> Testing
> -------
> 
> make check, added tests in https://reviews.apache.org/r/14044
> 
> 
> Thanks,
> 
> Ben Mahler
> 
>