-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/70863/#review215947
-----------------------------------------------------------


Fix it, then Ship it!





src/slave/containerizer/mesos/isolators/cgroups/cgroups.cpp
Lines 665-666 (original), 665-666 (patched)
<https://reviews.apache.org/r/70863/#comment302878>

    Tech debt, I would suggest to put this part of codes here.
    
    ```
      // We currently can't call `subsystem->isolate()` on nested
      // containers, because we don't call `prepare()`, `recover()`, or
      // `cleanup()` on them either. If we were to call `isolate()` on
      // them, the call would likely fail because the subsystem doesn't
      // know about the container. This is currently OK because the only
      // cgroup isolator that even implements `isolate()` is the
      // `NetClsSubsystem` and it doesn't do anything with the `pid`
      // passed in.
      //
      // TODO(klueska): In the future we should revisit this to make
      // sure that doing things this way is sufficient (or otherwise
      // update our invariants to allow us to call this here).
      if (containerId.has_parent()) {
        return Nothing();
      }
    ```


- Gilbert Song


On June 16, 2019, 7:47 p.m., James Peach wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/70863/
> -----------------------------------------------------------
> 
> (Updated June 16, 2019, 7:47 p.m.)
> 
> 
> Review request for mesos, Gilbert Song, Jie Yu, and Qian Zhang.
> 
> 
> Bugs: MESOS-9805
>     https://issues.apache.org/jira/browse/MESOS-9805
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> Currently, the PID targeted by the cgroups isolator is moved into
> the cgroup before the subsystem runs to apply any type-specific
> cgroup configuration. We should reverse the order of this so that
> the PID is only moved once the cgroup is fully configured by the
> subsystem.
> 
> A specific case that can happen is where a PID is assigned to a net_cls
> cgroup before that cgroup has its class ID set.  This intermediate
> process state can be observed by system monitoring process, causing
> confusion that is hard to debug.
> 
> 
> Diffs
> -----
> 
>   src/slave/containerizer/mesos/isolators/cgroups/cgroups.hpp 
> 4a1871b3b06b54a02dfe09289f7fb304a3f7f24c 
>   src/slave/containerizer/mesos/isolators/cgroups/cgroups.cpp 
> e7819d732172bdbd215106e3b781588c1f78b2ec 
> 
> 
> Diff: https://reviews.apache.org/r/70863/diff/1/
> 
> 
> Testing
> -------
> 
> sudo make check (Fedora 30)
> 
> 
> Thanks,
> 
> James Peach
> 
>

Reply via email to