=== jpabq is now known as jpabq_afk | ||
=== jpabq_afk is now known as jpabq | ||
=== jpabq is now known as jpabq_afk | ||
=== django|away is now known as django | ||
superm1 | dekarl: thanks for committing that patch for the exec stuff | 15:22 |
---|---|---|
superm1 | i'm glad to hear that sorts things out | 15:22 |
superm1 | i'm still a bit perplexed on why it's necessary though. isn't upstart supposed to track forks? | 15:22 |
tgm4883 | superm1, upstart only tracks forks if we tell it to (eg. 'expect fork') | 15:27 |
tgm4883 | but inside of that, IDK exactly why we're using exec instead of expect fork except that is what I was told :/ | 15:27 |
superm1 | oh i thought we had expect fork in the upstart job, but now that i look i see it's commented out | 15:28 |
tgm4883 | yea, IDK why that is | 15:28 |
superm1 | tgm4883: dekarl okay i think i'm following what's happening. upstart is tracking the PID of 'test' and the shell that sourced /etc/default/locale NOT the mythbackend PID | 15:47 |
superm1 | so when the kill signal is sent it gets sent to that shell not the backend PID | 15:47 |
superm1 | and then if you run it with exec it tracks the mythbackend PID at the end | 15:47 |
superm1 | mythbackend itself doesn't actually fork at all or anything | 15:47 |
superm1 | sorry if that's the same conclusion you came to, but if so it makes sense to me now :) | 15:48 |
dekarl | superm1: no worries.. I came to no conclusion at all, I just tested and pushed the fix that tgm4883 came up with :) | 15:50 |
superm1 | so the other remaining question is whether we bump the kill timeout up too | 15:51 |
superm1 | tgm4883 determined that the default between SIGTERM and SIGKILL signals is 5 seconds | 15:51 |
dekarl | I'm happy with using exec, thats what daemontools has in its run scripts, too. | 15:53 |
dekarl | If I understood correctly the 5 seconds are ok since the bug in shutdown was solved in parallel to us fixing the upstart job | 15:53 |
superm1 | i think that is the case too | 15:55 |
dekarl | searched for 5 minutes in irclog, commits and tickets didn't find the mention and giving up... only candidate being https://github.com/MythTV/mythtv/commit/0bb3bc2d9ca25528944429e84c21743db57b15a1 | 15:59 |
tgm4883 | superm1, sorry, yea I came to that conclusion previously but didn't explain all that | 16:13 |
tgm4883 | superm1, dekarl so while mythbackend is stopping in 2-3 seconds in our tests, should we be accounting for slower/complex systems? | 16:13 |
tgm4883 | superm1, dekarl I mean, if we set the timeout to 10 seconds, it's still going to only wait 2-3 seconds on most systems, but also give these other systems a little more time to shutdown before killing it | 16:14 |
superm1 | tgm4883: well looking through /etc/init there aren't many apps using a custom timeout | 16:15 |
superm1 | don't know that it would do any harm though | 16:16 |
tgm4883 | I don't think it would cause any harm. I'd think it would exist on precise (we should check) and thats the oldest we build for | 16:17 |
tgm4883 | superm1, I doubt many people would notice the 10 second timeout | 16:17 |
superm1 | true | 16:18 |
tgm4883 | superm1, I suppose the flip side of it is, what is the backend doing during shutdown that might be harmed by killing it early | 16:19 |
tgm4883 | superm1, that is probably the more important question | 16:19 |
superm1 | well maybe contacting mysql | 16:19 |
superm1 | but that might be killed early too | 16:19 |
tgm4883 | superm1, right, I'd just like a safe number to set as default | 16:22 |
tgm4883 | superm1, don't want to kill the backend if it's in the process of some mission critial stuff | 16:22 |
superm1 | OK well lets set it for 10 and see if we still get any complaints related to shutdown | 16:23 |
superm1 | it'll be better than what it is now | 16:23 |
tgm4883 | yea | 16:23 |
tgm4883 | sounds good | 16:23 |
tgm4883 | superm1, I'd have to see some other logs, but IIRC we're getting better messages in the mythbackend logs about shutting down (which is nice) | 16:25 |
tgm4883 | superm1, we could also turn logging way up, issue shutdown and see what it logs | 16:26 |
tgm4883 | all I get now is 4 lines | 16:26 |
superm1 | yeah if we have people complaining, that's definitely what we'll need to do | 16:26 |
superm1 | okay well i committed to master, 0.26, and 0.25. so we'll see in the next autobuilds what peoples says | 16:29 |
=== jpabq_afk is now known as jpabq | ||
=== Steve_Goodey is now known as Steve-Goodey |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!