Comment 3 for bug 703800

Yeah there are quite a few bugs related to pre-stop scripts, they don't
really work like any other, so there are definite gaps in the coverage

On Thu, Feb 24, 2011 at 8:00 AM, scm <email address hidden> wrote:

> I can confirm I see this behavior in upstart (0.6.5-8) on lucid. On
> several of our custom tasks, if I change pre-stop to post-stop, then
> restart behaves normally. I will see if I can craft an upstart job to
> mimic this behavior using stock tools.
>
> --
> You received this bug notification because you are a member of Upstart
> Developers, which is subscribed to upstart .
> https://bugs.launchpad.net/bugs/703800
>
> Title:
> restart command fails to restart main process when pre-stop stanza
> exists
>
> Status in Upstart:
> New
> Status in “upstart” package in Ubuntu:
> New
>
> Bug description:
> While testing the portmap daemon's recent changes in Ubuntu which
> cause statd to properly follow it on stop/start, I tried to restart
> portmap, only to find that it was not in fact restarted.
>
> This happens with any job that has a pre-stop. The reason is that in
> job_restart(), the code does this:
>
>
> job_change_goal (job, JOB_STOP);
> job_change_goal (job, JOB_START);
>
> job_change_goal will return as soon as the *first* state change has
> been completed. If there is no pre-stop, that is the change from
> JOB_RUNNING to JOB_KILLED, which does dutifully kill the main process.
>
> However, if there is a pre-stop, the pre-stop is run, but then
> job_change_state returns to job_change_goal, which then returns to
> job_restart, which then changes the goal back to start, which makes
> the new job_next_state() one that will bypass the change to
> JOB_KILLED.
>
> This was found on upstart v0.6.7-3 in Ubuntu, but also exists in the
> code in the current trunk.
>
> TEST CASE
>
> 1. create job file /etc/init/test-restart-prestop.conf with this content:
> # test-restart-prestop
>
> pre-stop exec /bin/true
>
> exec /bin/sleep 3600
> 2. run "start test-restart-prestop" -- record pid of job
> 3. immediately run "restart test-restart-prestop" -- if bug is present,
> pid remains the same when it should be a new pid.
>