[00:00] is it safe to say that the stuff in utils/ directory is depreciated and unmaintained? [00:13] no [00:13] not at all [00:20] Keybuk: anything else in libnih/upstart need some code love? [00:22] sadmac: haven't merged my stuff from the last week or two back in yet [00:23] Keybuk: what are we keeping for 0.10? RH is going to need init to do SELinux, and I'm wondering if that's worth doing now. [00:23] why do you need init to do it? [00:24] you can set the policy in the initramfs [00:24] that's what we're doing now. [00:24] there's push to move it back though. [00:25] I'm not going to have a "drop everything" flag day like we did 0.3->0.5 [00:25] every thing in 0.5 today will be in 0.10, or will be gradually changed [00:27] that's good to know. [00:27] timetable's been pushed up a bit too [00:27] oh? [00:28] I'm probably going to be working full-time on Upstart from a week or two's time through to June [00:28] rather than later [00:29] http://people.ubuntu.com/~scott/stig-20090208-0940.png [00:29] (my first attempt at a native Upstart boot) [00:29] 18s. Too slow. [00:29] :D [00:29] no [00:29] you're mis-reading the chart [00:29] it's 8.5s [00:30] you did early login? [00:30] everything after that is me getting a terminal opening, and fumbling to stop the bootchart collector ;) [00:30] oh [00:30] no, that's all-up [00:30] X in 3s [00:31] If the service management stuff is stable enough (and why wouldn't it be) I'm pretty sure I can write some decent dynamic ionice stuff that'd make early login work well. [00:31] define "early login" ? [00:32] popping up gdm before we've loaded everything (in the event driven sense: before the running set has stabilized) [00:33] that's bringing up gdm once its dependencies are met [00:33] maybe that's the same thing? [00:33] keybuk: Cool×2 (the timetable and the native boot) :-) [00:33] if you installed other services like apache, they'd start up about the same time [00:33] Keybuk: its that + not making gdm depend on some "the system is done booting" trigger [00:34] oh, there's no such thing [00:34] Keybuk: right [00:34] gdm depends on d-bus, and /usr being mounted [00:34] Keybuk: doing anything /but/ early login is a pain in the ass now. [00:34] hal is system-bus activated [00:34] and depends on acpid [00:34] (so X blocks on it) [00:35] also depends on d-bus and udev [00:36] Keybuk: so the idea of this would be that upstart could be set in a kind of "background mode" such that when a new service's goal changed, it would be ioniced into the "idle" class. It would be ioniced back into the class specified by its configuration (or the default) when it reached the "started" or "stopped" states. [00:38] And this would be done to ~everything else than the services that lead to gdm getting started? [00:39] ion_: it would be turned off until gdm was started [00:39] ion_: the idea is "get non-interactive stuff out of the way of the user" [00:39] Ah [00:57] sadmac: I've been looking into upstart in Fedora CVS. the devel branch doesn't seem to be building (missing dbus-devel as a BuildDep), and I was wondering if it was that or some other reason rawhide is using upstart 0.3.9 from F10 [00:58] sadmac: (im looking to get some rawhide builds working on my OLPC XO) [00:58] nteon: we aren't going to 0.5.1 [00:58] nteon: neither is ubuntu. [00:58] nteon: next release shipped by any major distro (afaik) will be 0.10 [01:01] sadmac: good to know, thanks :) [01:01] I suppose I can stop trying to get my rawhide box booting with 0.5.1... [01:06] heh [01:09] around when is 0.10 expected to come out? [01:09] nteon: in time for the next Ubuntu release... damned if I can remember all the dates [01:09] should be F12 material [01:23] sadmac: sounds good [01:25] sadmac: I'm cautiously interested in getting more involved in Fedora development, so if theres anything I can do to help with upstart packaging or patching, let me know (not that I'm intimately familiar with the codebase or anything) [06:06] Keybuk: did you go back to a custom control protocol for initctl? [06:06] Keybuk: there's no reference to dbus in initctl in trunk === sadmac_ is now known as sadmac [08:08] sadmac: SUBDIRS = m4 intl nih nih-dbus dbus init doc po # util [08:09] util/ in trunk isn't ported (yet) afaics [10:27] what would be the right place to increase the timeout between initctl and upstart? [10:42] we lost a lot of functionality while swichting to dbus [10:42] while adding more code :p [10:43] and dependencies [10:44] I hope there will be something in return [10:52] More secure IPC. Upstart can determine when a service is running based on when it appears on D-Bus. [11:12] is there kerberdos support in dbus. otherwise i don't why it's more secure [11:12] More eyes looking at the networking code. [11:13] Especially relevant when normal users gain access to Upstart. [11:14] I see. [11:14] I did not look at how a user is stopped from sending events yet [11:16] Also, the D-Bus dependency should be a non-issue in most places. For an embedded system with no need for IPC and extremely limited resources, perhaps the D-Bus code can be even disabled. If not yet, the one needing it is free to share a patch. :-) [11:17] my kind of patches and ideas generaly don't get accepted in upstart. [11:18] BADCODE i guess [14:08] sadmac2: 0.5 is the current branch [14:09] I haven't ported the util stuff over yet [15:42] Keybuk: ah