[RFC] stopsignal stanza

Marc - A. Dahlhaus mad at wol.de
Wed May 4 20:51:43 UTC 2011


Am 04.05.2011 18:29, schrieb Clint Byrum:

> Excerpts from Marc - A. Dahlhaus's message of Wed May 04 03:52:35 -0700 2011:
>> Am Montag, den 02.05.2011, 16:28 -0700 schrieb Scott James Remnant:
>>> Looks to me like you're on exactly the right track here.
>>>
>>>
>>> Just needs some more test coverage. I have no further comments ;-)
>>>
>>
>> Version with tests attached...
>>
>> IMO the naming of the kill stanza is now slightly inappropriate:
>>
>> kill timeout VAL:
>> Was clear for defining the timeout before we send a SIGKILL after
>> stopping a job.
>>
>> kill signal SIG:
>> We are not changing the SIGKILL signal here but the SIGNAL we send to a
>> Job that should be stopped before we send a kill signal if the job
>> doesn't stop before it stops longer as the defined timeout allows it to
>> do.
>>
>> I suggest changing the name of the stanza from kill to stopping and add
>> an alias kill for backward compatibility and document it as such. It
>> would be a better and not misleading name as we send "stopping signal"
>> when we stop the job and we would send SIGKILL to it after the "stopping
>> timeout" is reached.
>>
>> What do you think?
> 
> IMO, kill timeout X should remain, it is very clear what it does
> 
> Since upstart stanzas have leaned toward readability, I think this works
> quite nicely:
> 
> kill signal SIGINT
> reload signal SIGUSR1
> kill exec postfix stop
> reload exec apachectl graceful
> 
> SIGKILL is special, and must never be overridden, so it should not even
> be considered. The 'kill timeout' is the timeout waiting for the kill
> action before SIGKILL is sent. Its always going to be confusing, because
> of the symbolic name of SIGKILL. No matter what the special action is,
> we're always going to send SIGKILL to the pid after kill timeout.


Didn't know about a to be implemented kill exec. Makes sense then...

Marc



More information about the upstart-devel mailing list