> On March 2, 2017, 4:40 p.m., Jiang Yan Xu wrote:
> > Did you run into issues without this patch? cgroups::destroy() *should* 
> > kill all tasks in it.
> 
> James Peach wrote:
>     On my reading of `cgroups::destroy`, processes would only be killed when 
> destroying the `freezer` cgroup, since that is the only one with the 
> `freezer.state` control that would trigger the `internal::Destroyer` to run.
>     
>     The linked bug 
> [MESO-7049](https://issues.apache.org/jira/browse/MESOS-7049) shows test 
> output where the cgroup teardown fails.

Your are right. But `cgroups::kill` is asynchronous and we are not waiting on 
it. It seems we should improve `cgroups::destroy` to optionally not require the 
freezer for this to be reliable?


- Jiang Yan


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/56730/#review167773
-----------------------------------------------------------


On March 7, 2017, 2:27 p.m., James Peach wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/56730/
> -----------------------------------------------------------
> 
> (Updated March 7, 2017, 2:27 p.m.)
> 
> 
> Review request for mesos, haosdent huang, Jie Yu, and Jiang Yan Xu.
> 
> 
> Bugs: MESOS-7049
>     https://issues.apache.org/jira/browse/MESOS-7049
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> If a test case fails, it may leave stray tasks in the cgroup which keeps
> us from tearing it down when the test completes. Kill any stray tasks
> before destroying the cgroup.
> 
> 
> Diffs
> -----
> 
>   src/tests/containerizer/cgroups_tests.cpp 
> 76fabce4530ccc0a1d685cd48d932ced5a64bc58 
>   src/tests/mesos.cpp 6a96fa51dfc2a62063c3154b256bdac707b009bb 
> 
> 
> Diff: https://reviews.apache.org/r/56730/diff/1/
> 
> 
> Testing
> -------
> 
> sudo make check (Fedora 25)
> 
> 
> Thanks,
> 
> James Peach
> 
>

Reply via email to