[03:31] <pwnguin> is there a list of known upstart events in ubuntu 9.04?
[14:16] <ion_> http://wiki.debian.org/LSBInitScripts/DependencyBasedBoot “At debconf7 and FOSDEM 2008 (video link), Petter Reinholdtsen gave a talk about the dependency based boot system.” Not even a mention of Upstart. :-P
[14:20] <Keybuk> that's Petter for you
[14:21] <ion_> He mentioned that “Linux is becoming more and more event-based but the system startup is sequential, but that’s an issue for another talk” (paraphrased).
[14:24] <Keybuk> :)
[14:27] <ion_> Btw, how far along is Upstart 0.next? :-)
[14:30] <Keybuk> the upstart bits are still pretty much untouched :p
[14:43] <Keybuk> today I am mostly hating D-Bus
[14:44] <sadmac> Keybuk: what did it do? Show us on the bear where glib touched you
[14:49] <Keybuk> ;)
[14:49] <Keybuk> just damned hard to do signals properly
[14:50] <sadmac> yeah, they didn't look fun
[14:50]  * sadmac has been thinking about thread-safe linked lists for libnih
[14:50] <sadmac> I don't know why, we don't need them right now
[14:51] <sadmac> but massively-concurrent linked lists are hard.
[14:57] <Keybuk> indeed
[15:00] <sadmac> and it makes the list heads bigger (need a semaphore). Might want the thread-safe ones to be a different type.
[15:00] <Keybuk> put the mutex in the list head
[15:00] <Keybuk> that's the usual pattern for locking linked list changes
[15:00] <Keybuk> it's not record-level, but it's enough
[15:00] <Keybuk> record-level, you're just asking for deadlocks
[15:02] <sadmac> I can do nearly record level with no deadlocks
[15:02] <sadmac> the trick is don't lock the nodes. Lock the links
[15:02] <sadmac> each node has a lock that protects its next pointer and /the next node's prev pointer/
[15:03] <sadmac> you only have to acquire two locks for each structural change. its juuust this side of safe
[15:05] <sadmac> it only can lock if its possible to call remove() on the list head itself (dining philosophers). I think I can fix that.
[15:59] <Plouj> hi
[15:59] <Plouj> does upstart have the ability to log to syslog all of the boot messages instead of to the console?
[16:02] <Plouj> well, not necessarily to syslog, but somewhere permanent and not initially user-visible
[17:31] <pwnguin> is there a list of known upstart events in ubuntu 9.04?
[17:38] <sadmac> pwnguin: Fedora has a manpage. I think its available on die.net
[17:46] <pwnguin> hmm. no network event?
[17:47] <sadmac> pwnguin: no. stuff like that should be emitted by other system services. you might want to look into having network manager do it.
[17:47] <sadmac> anything that can call `initctl emit` can emit an event
[17:48] <pwnguin> well, i really dont care if upstart itself emits an event versus given system service
[17:48] <pwnguin> i will look into NetworkManager though
[17:51] <sadmac> Plouj: there's nothing like that just yet. We've planned it though. Even tried it a few different ways. Its not as easy as it would seem to get right.
[17:51] <sadmac> Keybuk: do you have any plans as to making logging work?
[17:51] <Plouj> sadmac: can you tell me of some of the difficulties that you encountered?
[17:53] <sadmac> Plouj: well, we didn't want init to funnel all of the text traffic from all the services for resource reasons (that was Keybuk's logic. He can explain that) so we had our own logging service to transfer the info. The problem was when that went down it ended up SIGPIPEing everything connected to it, resulting in a pretty spectacular system failure.