Pre-start and stop

[Clint Byrum]

Excerpts from Steve Langasek's message of 2014-01-07 10:25:12 -0800:
> On Mon, Jan 06, 2014 at 10:45:31PM -0500, Stéphane Graber wrote:
> > > It seems like most of the examples in the Cookbook recommend that if you
> > > want to stop a job in the pre-start stanza, you should call "stop" which
> > > will handle stopping the job for you.  But, it seems looking at the
> > > Upstart code I can just as easily return a negative return value to
> > > cause the job to stop:
> 
> > > http://bazaar.launchpad.net/~upstart-devel/upstart/trunk/view/head:/init/job.c#L417
> 
> > > Is there a reason I should call "stop" over just returning a negative
> > > value?
> 
> > Calling { stop; exit 0; } also means the job won't be considered as failed.
> 
> > Returning non-zero will log an error in the log file and prevent any job
> > depending on the current job from starting (which may or may not be what
> > you want).
> 
> > So in short, exitting non-zero or calling stop+exit are two different
> > things with different behaviours, you have to choose which you want.
> 
> A job depending on the current job won't start if you call 'stop' from the
> pre-start, either - because a depending job will have 'start on started
> foo', and the 'started' event will not be emitted for a job stopped from the
> pre-start.
> 

I did the following test:

/etc/init/test-block.conf
pre-start script
  sleep 5
  exit 1
end script

/etc/init/test-block-slave.conf
start on starting test-block
script
  env > /tmp/env.$$
end script

/tmp/env.$$ ended up as:

UPSTART_INSTANCE=
INSTANCE=
UPSTART_JOB=test-block-slave
TERM=linux
PATH=/usr/local/sbin:/usr/local/bin:/usr/bin:/usr/sbin:/sbin:/bin
UPSTART_EVENTS=starting
PWD=/
JOB=test-block

Consequently when I changed it from 'exit 1' to 'stop' (and thus, exit 0)
the slave started exactly the same.

So if you need something _started_ you should use the started event. If
you need to start before something, you should use the _starting_ event.
But I don't think you can depend on something "after pre-start succeeds
but before started". For that you need the other job to emit an event
in pre-start on success. Note that it probably wouldn't be hard to have
upstart emit a pre-started, but I question the need for such a thing.