[01:48] I'm playing with upstart, but I can't login on tty1.. any idea how to make a job that just starts a shell on tty1 ? [01:49] damjan: start on runlevel 5 / exec /bin/bash [01:49] runlevel 3 rather [01:50] how about "start on startup"? [01:53] I'm getting "main processes ended" [02:16] damjan: Also add "session leader" [02:18] ar 10 02:13:52 nixos init: sshd pre-start process (25558) [02:18] MarcWeber: still no luck "init: rescue main process ended, respawning" [02:18] seeing this line and ps -p no longer showing any process pid 25558 [02:18] rescues is the name of the job (file) I created [02:19] Then this means that the kernel doesn't tell upstart the the pr-start process has finished? [02:20] damjan: Are you sure that upstart has noticed that your file has changed? [02:20] MarcWeber: I'm installing upstart on a non-Upstart system [02:20] in a virtual machine, so I reboot it all the time [02:21] damjan: Which virtual machine? vserver? [02:21] KVM [02:21] Or qemu.. [02:22] I can't even login to the VM with upstart [02:22] http://rafb.net/p/3MfPOS65.html from line 3 on. This works hree [02:22] the stupid thing is, obviously /bin/bash fails to start for some reason but where is the error output ?!? [02:24] Try exec &> /tmp/foo [02:25] well / is not rw at that point :) [02:25] .. I suspect [02:25] .. although I have a script that should do the basic system setup, I don't see it's output so I don't know if it worked ok [02:26] You have do create /dev/console [02:27] yeah, I know that .. hmm .. I think it should exist .. but let me confirm [02:28] .. (just what was the trick to get the real /dev/ beneath udevs tmpfs ?!?)... [02:28] mkdir and mknod use google or kernel Documentation [02:29] ok the trick is "mount --bind / /mnt" now I can see the real /dev/ in /mnt/dev/ [02:29] it does have /dev/console null and zero [02:33] I know that upstart has nice goals. However I think it should try to be at least as good as the old init* programs. [02:36] I feel like it's hard to beat the old init* programs, they've been baked to perfection with the years [02:38] It's not that hard. However not beeing able to set oom shouldn't make starting the job fail altogther. [02:39] Maybe there should be a simple /bin/sh test and if upstart doesn't get to know within 1/20s that it has failed assume that it has to use ps -p to check wether a process has died. I know that sucks. But it would work in 99% of all cases which is much better than 0% [02:43] Let's hope that my snapshot I'v taken month ago works fine. It alread did.. [03:23] No, didn't help. Maybe I'm just talking nonsense. I'm a little bit disappointed that I still haven't suceeded.. [10:32] Keybuk, sadmac Do you think there is any chance using an alternative less pretty implementation of getting notified when processes (such a spre-start) exit? [10:33] how do you mean? [10:34] As I've said. Currently I've the problem on the v-server that the sshd-pre start script gets started.. [10:35] init prints the pid to the log. But then nothing happens. I can no longer show that process using ps -p pid [10:35] So it has died (it contained a mkdir -p command only. .) [10:36] init doesn't get notified that the pre-start script has exited so it dosen't start sshd [10:36] why doesn't init get notified that the pre-start script has exited? [10:36] that sounds like your vserver has broken if it's not sending SIGCHLDs to init [10:39] I can try contacting the support service again and ask wether they can enable it. Don't know yet was is happening. All I now that pre-start doesn't cause the daemon to be started. [10:39] Can I test the SIGCHLDs without being PID 1 somehow? [10:39] sure, sigchld is pretty fundamental ;) [10:39] did you try init with --debug and capturing the output? [10:39] only -v [10:39] I'll retry [10:40] I've commented out the whole oom_adj part Because I don't have permissions to change them (for whatever reasons). [10:48] line 165 http://rafb.net/p/gdEQ1132.html [10:48] sshd should be started on event manual [10:50] The job and debugging output http://rafb.net/p/sXDAnh65.html [10:52] Would it make sense to add an upstart-test application which could be run on startup once to verify that all required features are availible? [10:55] I could also move the code from the pro-start scripts into the main script.. [11:02] Is --debug output sent to the logging daemon as well? [11:04] Keybuk: So does this anything to you? [11:42] Keybuk: Would you mind pingingn me if you're availible for some minutes again? [13:55] Keybuk: You've been right. the strace shows that no SIGCHD is sent. http://rafb.net/p/2VpHyE47.html. I've sent another ticket. Thank you! [13:57] MarcWeber: sorry, been having X server crashes all morning [13:57] lost your log [13:57] Ah never mind. :-) [13:57] I'll never understand why people use vserver [13:57] from what I can tell, it doesn't implement a POSIX system very well [13:57] Because they are cheep and except of those troubles they work well.. [13:57] so it's pot luck whether anything works [13:58] surely Xen has killed it by now? [13:58] Keybuk: I can't affort a dedicated server yet :-( [13:58] I don't mean virtualisation in general [13:58] just the particular implementation that is vserver [13:58] So which kind of hosting service do you recommend ? [13:58] http://www.linode.com/ do lovely cheap Xen instances ;) [13:58] $19,95/mo for basic [13:59] though I don't know how that compares with your vserver host [13:59] Xen is supposed to work? [13:59] Xen works brilliantly [14:00] I guess I've found another hosting company which is using xen. So I'll just give up and switch again :) [14:00] heh [14:00] I don't mean to be mean ;) [14:00] but I have been on the receiving end of what feels like dozens of vserver bugs now [14:00] (both for upstart and udev) [14:00] where it just flat-out does things that the book says it can't [14:01] I can't ping linode < 120ms [14:01] So maybe I should use a hosting service which is closer to Germany. [14:01] Keybuk: I think upstart is great. I think it will take off. [14:02] there is one, I can't think what it is [14:18] Keybuk: liked the talk [14:18] Keybuk: I've just added some debugging hints which have helped me: http://upstart.ubuntu.com/wiki/Debugging#preview [14:22] echo "Debug Message" >> /tmp/upstart will not succeed until / is remounted RW [14:27] damjan: I've added that note :) [14:30] Keybuk: May I also quote you telling that Xen systems are known to work much better? [14:38] also the naming sounds a little weird echo "Debug Message" >> /tmp/upstart looks like it some sort of magic to make upsart be in debug mode [14:39] echo "Debug Message" >> /tmp/my_job_debug sounds better to me [14:39] with 0.5.1 you can do initctl log-priority debug so see what is happening inside upstart [14:40] I would also say on what version this wiki page applies to [14:51] I hope that everyone setting up upstart has enough knowledge to know what echo "" >> foo means :-) [14:53] keesj: I've added the log-priority hint as well [15:52] Meh, no waitfd in jaunty’s kernel. [15:53] no [16:33] ion_: no waitfd in anybody's kernel I hope. [16:33] the patch isn't upstream [16:44] talking of which [16:44] * Keybuk fires the patch gun at lkml [16:45] Keybuk: you have a patch for it? [16:45] no [16:45] you have :p [16:45] these are different patches [16:46] Keybuk: oic [16:46] Keybuk: thought you got bored and just rewrote it for me :) [16:50] Keybuk: how will upstart process these "when dbus started" lines? is dbus an argument before the name? some other construct? [20:40] Any news about waitfd? Is it getting into Linus’ tree any time soon? [20:43] ion_: I think it needs another round of fixes, and impressions are mixed. It also needs to be better explained. [20:46] The next patch had better be good. Alan Cox is going to flip his shit when he sees the thing a third time. [20:50] Heh