Fix a bug where outer containers are never released, when a single
job process created both an inner container and outer container.
This gave container allocation errors on later job steps, or
a fatal rmdir error on parallel job's final release_container.
When releasing a job tasks' current container, step back to the
parent container that the job was executing within before the
current container was created.
Portions of class function cpuset.release() were used to destroy
old orphaned containers. These steps should not modify the
job.container cpuset object. They are now factored out as
non-class function cpuset.release_container().
Risk: low.
Signed-off-by: Duane Sand <duanes@google.com>
git-svn-id: http://test.kernel.org/svn/autotest/trunk@2190 592f7852-d20e-0410-864c-8624ca9c26a4
2 files changed