Two reasons:

1) Files in a cgroup are not regular files. You cannot 'rm' them. You can
only remove the directory using 'rmdir'

2) We want the creation/removal of a cgroup to be atomic. For example,
assume the cgroup hierarchy is like the following:

/cgroup -- hierarchy root
/cgroup/a
/cgroup/a/1
/cgroup/a/2

You want to remove the cgroup 'a'. If you use the recursive version, here
is what will potentially happen:

  1. cgroup 'a/1' is removed
  2. cgroup 'a/2' cannot be removed because some tasks are still in the
cgroup
  3. cgroup 'a' cannot be removed since 'a/2' is not removed yet.

In that case, we have a partial removal, which we don't expect.

- Jie

On Fri, Sep 28, 2012 at 2:10 PM, Vinod Kone <[email protected]> wrote:

>
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/7338/#review12026
> -----------------------------------------------------------
>
>
>
> src/linux/cgroups.cpp
> <https://reviews.apache.org/r/7338/#comment25646>
>
>     I don't understand why recursive is not good here?
>
>
>
> src/linux/cgroups.cpp
> <https://reviews.apache.org/r/7338/#comment25645>
>
>     ditto
>
>
> - Vinod Kone
>
>
> On Sept. 28, 2012, 5:33 a.m., Jie Yu wrote:
> >
> > -----------------------------------------------------------
> > This is an automatically generated e-mail. To reply, visit:
> > https://reviews.apache.org/r/7338/
> > -----------------------------------------------------------
> >
> > (Updated Sept. 28, 2012, 5:33 a.m.)
> >
> >
> > Review request for mesos and Benjamin Hindman.
> >
> >
> > Description
> > -------
> >
> > The recent refactor changes break the assumptions in the cgroups code.
> >
> >
> > Diffs
> > -----
> >
> >   src/linux/cgroups.cpp cdafe6e
> >   third_party/libprocess/include/stout/os.hpp 13dbc71
> >
> > Diff: https://reviews.apache.org/r/7338/diff/
> >
> >
> > Testing
> > -------
> >
> > make check.
> >
> > Tested on my vm (latest ubuntu 12.04)
> >
> >
> > Thanks,
> >
> > Jie Yu
> >
> >
>
>

Reply via email to