| jMCg | how do I tell upstart's stop to kill all children of a process too? | 00:59 |
|---|---|---|
| jMCg | https://git-wip-us.apache.org/repos/asf?p=trafficserver.git;a=blob_plain;f=rc/trafficserver.conf.in;hb=HEAD | 01:00 |
| jMCg | HRM. I don't have expect fork in my version. Maybe that's it. | 01:01 |
| jMCg | Oy, no answer here :-/ | 16:46 |
| SpamapS | jMCg: IIRC upstart will kill the entire "process group" | 16:57 |
| jMCg | SpamapS: but it doesn't | 16:59 |
| SpamapS | jMCg: actually no, I see now, it just kills the tracked pid | 17:01 |
| jMCg | SpamapS: sometimes I miss SMF | 17:01 |
| SpamapS | jMCg: really though, a parent that doesn't kill its children off on SIGTERM is misbehaving. ;) | 17:05 |
| SpamapS | jMCg: ok, so if I understand my "process group" mechanics correctly, kill() to a pid that is a process group leader will be delivered to all process group members. | 17:08 |
| SpamapS | hrm no I misread that | 17:23 |
| SpamapS | have to send to -pid to get the process group | 17:23 |
| jMCg | SpamapS: thing is, this is a controll process, and it's children, so really, what I need is a way to tell upstart how *really* to reload or restart the. thing. | 17:34 |
| SpamapS | jMCg: you can change the normal kill signal | 17:34 |
| SpamapS | unfortunately, not the reload signal | 17:34 |
| SpamapS | jMCg: so if SIGTERM is too violent/handled poorly by that process.. then you canchange to whatever is the more graceful way to shutdown | 17:35 |
| SpamapS | but ultimately, SIGHUP needs to be handled well | 17:35 |
| jMCg | SpamapS: I'll have to discuss that with my peers, the thing is: We start traffic_cop in upstart, whic starts traffic_manager, which starts traffic_server which oh fuck me. | 17:35 |
| jMCg | I mean, gimme a sec, I might have done something wrongishly. | 17:36 |
| jMCg | https://issues.apache.org/jira/browse/TS-755 | 17:36 |
| jMCg | Okay, that doesn't really affect anything here. | 17:37 |
| jMCg | Maybe I need to tell upstart to expect fork? | 17:37 |
| jMCg | Lemme try that. | 17:38 |
| * SpamapS reads upstart code again | 17:38 | |
| jMCg | Yes, I do need to do that. As I correctly do in the code I have up on apache's git ;) | 17:39 |
| jMCg | So why did I do it wrong in my installation? I don't know. | 17:39 |
| jMCg | Of course this is a really not very graceful way of restarting the thing. | 17:40 |
| SpamapS | jMCg: ok, I can confirm, upstart kills the whole process group / session | 17:40 |
| SpamapS | if (kill (pgid > 0 ? -pgid : pid, signal) < 0) | 17:40 |
| SpamapS | nih_return_system_error (-1); | 17:40 |
| jMCg | SpamapS: with expect fork, or even without? | 17:40 |
| SpamapS | jMCg: no matter what | 17:41 |
| SpamapS | expect fork just helps it figure out the pid | 17:41 |
| SpamapS | jMCg: actually, reload does not send to the group | 17:43 |
| SpamapS | if (kill (processes[0]->item1, SIGHUP) < 0) { | 17:43 |
| SpamapS | which makes sense | 17:43 |
| SpamapS | its not trying to clean up, its trying to tell the main process to reload its configuration | 17:43 |
| jMCg | SpamapS: I need to test if this will work, and if does, adapt a new workflow. | 17:45 |
| jMCg | SpamapS: it would appear as though it works! | 17:47 |
| SpamapS | jMCg: \o/ | 17:49 |
| jMCg | We did something right in our code! | 17:51 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!