| geofft | is upstart.at known to be down? if it's permanently down, should links in the cookbook point to e.g. netsplit.com instead? | 02:20 |
|---|---|---|
| SpamapS | geofft: I hadn't noticed that it was down, but perhaps we should correct that | 03:51 |
| geofft | yeah, not all the posts are on netsplit.com, looks like. | 03:52 |
| geofft | should I consider this reported, or should I email upstart-devel? | 03:53 |
| alexbligh | I want to start an upstart job after the root filesystem has been remounted r/w, but before udev is started. What is the best way to achieve this? | 12:02 |
| alexbligh | (the job is a task) | 12:03 |
| yatesy | Hi folks, is it actually possible to trigger upstart to re-read the definition of an upstart service without having stop and start the service? Ideally I'd love the ability to restart a running service and it act on changes in the definition | 12:54 |
| jodh | alexbligh: mountall is going to emit the 'mounted' event when '/' is remounted r/w but udev starts on 'virtual-filesystems' which occurs before local disks are mounted. See http://upstart.ubuntu.com/cookbook/#mountall-events. | 13:04 |
| jodh | alexbligh: 'start on started virtual-filesystems' (from #ubuntu-devel) is incorrect - 'virtual-fileystems' is an event, not a job. | 13:04 |
| jodh | alexbligh: see upstart-events(7). | 13:05 |
| alexbligh | jodh, Does udev actually run with / mounted readonly? What I need to do is write a file before persistent-net-rules is used (in essence I need to store the state somewhere else because / is on a ramdisk) | 13:14 |
| jodh | alexbligh: theoretically yes since the 'virtual-filesystems' event gets emitted before any local filesystems are (re-mounted), but realistically the disk might be mounted r/w by the time udevd starts. | 13:16 |
| alexbligh | Given udevd writes stuff to / (I believe) how does it avoid getting disk read only errors? | 13:17 |
| jodh | alexbligh: I don't think it does write to / - it writes to /dev/.udev/ and /run/udev, both being on virtual filesystems. | 13:20 |
| alexbligh | hmmm.... | 13:22 |
| jodh | alexbligh: maybe you could write a udev rule to run a program that would "find" the state you're talking about? | 13:23 |
| alexbligh | jodh, the state is the contents of persistent net rules. But I think you are saying the persistent net rules udev script could run when / is read only. | 13:24 |
| jodh | alternatively, if you will always have an initramfs, you could call whatever program you want from the initramfs, *then* start Upstart. This is how friendly-recovery works on Ubuntu (see /etc/init/friendly-recovery.conf and penultimate line of /usr/share/initramfs-tools/init). | 13:24 |
| alexbligh | jodh, will upstart work if I mount / read-write? | 13:25 |
| jodh | yes - mountall just sees that it doesn't need to remount /. | 13:25 |
| alexbligh | jodh, ok, so if I make a task on the starting event for udev, I could remount / readwrite in there? | 13:26 |
| jodh | alexbligh: however, take care wrt fsck | 13:26 |
| alexbligh | fsck isn't needed because it's a ram disk we've just tftpbooted :-) | 13:27 |
| jodh | alexbligh: yes, but then you'll be potentially fighting against mountall which will also be trying the same operation around the same time. | 13:27 |
| alexbligh | jodh, Can I make this start on mountall starting OR udev starting (so it will start before either of them)? In which case any idea how I do that? | 13:28 |
| alexbligh | jodh, i.e. 'start on starting udev or starting mountall' (or something similar) | 13:31 |
| jodh | alexbligh: that will potentially run 2 instances of the job (although not if the time the 2nd event arrives occurs before the job has finished). | 13:32 |
| alexbligh | jodh, that won't be an issue because I will make the job idempotent | 13:33 |
| alexbligh | jodh, but is the syntax right? | 13:33 |
| jodh | alexbligh: yes. | 13:35 |
| alexbligh | jodh, fantastic, thanks! | 13:36 |
| SpamapS | jodh: hey. I was wondering.. are we expecting any upstart features in 12.10 ? | 14:05 |
| jodh | SpamapS: yeah. Currently working on a biggie: https://wiki.ubuntu.com/FoundationsTeam/Specs/QuantalUpstartStatefulReexec | 14:24 |
| jMCg | Can I disable "respawn" in an .override file? -- This is sort of relevant for pacemaker jobs, which are the ones that do the management here. | 14:24 |
| jodh | jMCg: no as you can't meaningfully create an override file without that stanza and have it override an already defined 'respawn' in the .conf. | 14:26 |
| jodh | jMCg: I don't follow why you want to do this. | 14:26 |
| jMCg | jodh: because if you have the job running in a cluster, you don't want the cluster think that the job is dead, start it on a different node, while upstart restarts it on the same node again, that would be not very good in most cases. | 14:27 |
| SpamapS | jodh: Oh cool! | 14:42 |
| jodh | SpamapS: it's pretty close right now. | 14:42 |
| SpamapS | jMCg: you could effectively disable it by setting the respawn limits very low | 14:43 |
| SpamapS | jMCg: I don't know if upstart will allow it, but you could try 'respawn limit 0 1' | 14:44 |
| SpamapS | jMCg: actually 'respawn limit 0 999' would be better.. meaning "if it dies more than 0 times in 999 seconds" | 14:45 |
| jMCg | SpamapS: so long as I can put that in an .override file. | 14:51 |
| SpamapS | jMCg: yes you can | 15:06 |
| SpamapS | jMCg: 'respawn' and 'respawn limit' are two different stanzas | 15:06 |
| jMCg | ACK | 15:14 |
| === JanC_ is now known as JanC | ||
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!