=== bradleyayers_ is now known as bradleyayers [18:09] Hrm.. jenkins forks and exits when you send it HUP [18:09] meaning the job appears to die, and upstart loses track of the pid [18:10] May have to drop pid tracking for it then. Bummer. [22:56] what's a good late-in-boot-as-possible event normally? runlevel 2? [22:56] SpamapS, ^ [22:56] Heh, we just had a big discussion about this yesterday [22:57] there isn't a good one [22:57] :) [22:57] the best you have is 'stopped rc RUNLEVEL=[2345]' [22:57] Which is when tty1 pops up [22:58] SpamapS, brings to mind the famous question... what's your sleep(1) number? [22:58] marrusl: we need an intermediate job for mid-boot things that you want to track the state of. this goes back to the 'network-services' job. [22:58] lol [22:58] marrusl: what class of things do you want to be after? [22:59] If its "everything" .. we're probably going to transition nearly everything that is 'start on runlevel [2345]' to be 'start on starting network-services' [22:59] and then if you want to be "after that stuff" you will be 'start on started network-services' [22:59] SpamapS, in this case networking indeed. [22:59] and it's on lucid. :-/ [22:59] marrusl: for networking, runlevel 2 is good 11.10 and later [23:00] marrusl: for lucid, your best bet is to delay runlevel 2 for all of the network devices [23:00] I wonder.. [23:00] perhaps we should backport the static-network-up event [23:00] not the change to /etc/init/rc.conf [23:01] SpamapS, basically so far your idea to use "start on started udev" seems ok, but since the problem condition is so rare (~3-5% of boots) we have to script checking for the problem conditions... [23:01] and if none, reboot. [23:02] but when I had an invalid upstart job starting infiniband (well failing to start) [23:03] the script didn't catch the error condition and kept rebooting. [23:04] marrusl: burn-in test :) [23:04] SpamapS, haha. basically yup.