[00:27] mbiebl: the shutdown messages are normal [00:27] that's unrelated [00:27] at -> atd - good catch [00:27] udev-finish - good catch [00:28] the directly after upgrade stuff isn't ideal [00:28] not sure why you have a running dbus though [00:28] but it's because those are services are can't shutdown, so we can't "inherit" them [00:28] they're still running under sysv conditions until you reboot [01:23] Keybuk: I'd also suggest to remove dbus-reconnect.conf and run telinit q in dbus.conf's post-start [02:23] mbiebl: doesn't work ;) [02:25] Why is that? [02:29] because telinit q reloads all the configs [02:29] and throws away the accumulated "start on" events ;) [02:29] it was a bad choice for dbus reload, in hindsight [02:30] So, there should be a way to tell Upstart just to reconnect to dbus? [02:30] yes [03:39] keybuk: rc-sysinit starts again if one runs mountall from a system that has already booted up. Any way to fix that? [03:40] The job starts running rcS scripts [04:11] keybuk: Why is anacron a task? [06:35] can someone point me to some docs on what events i can emit [06:36] like how do i start stop services [06:36] using initctl emit [06:37] or like... cant i get a list of services this way? [06:39] ion: anacron is not a daemon [06:40] It's triggered by cron (and other event sources) [06:41] and throws away the accumulated "start on" events ;) [06:43] Do you intend to approach it differently then or keep the current solution? [13:33] mbiebl: Ah. In that case, ‘task’ is right, but something else must be done to prevent ‘start anacron’ from blocking for ages, especially during the installation of the package. [13:39] oh, is that what it's doing? [13:39] hah [13:39] to053953 < ion> keybuk: rc-sysinit starts again if one runs mountall from a system that has already booted up. Any way to fix that? [13:39] to054029 < ion> The job starts running rcS scripts [13:40] no, I mean anacron [13:41] I mean, that’s another issue i talked to you about. [13:41] oh [13:41] Don't Do That Then [13:41] Hehe [13:42] mountall will emit events like filesystem, which will start other bits [13:42] Indeed [13:42] I guess rc-sysinit should have "start on startup and filesystem" [13:42] because it should be only run-once [13:45] I think there’s a slight design problem with my fsck-queue patch. I’m just installing a virtual system to test. If there’s a LVM VG that resides on /sda1 *and /sda2, the code will probably output {"8:0", "8:0", NULL} which then will get both locked (i.e. added to the lock hash). [13:46] That is, i’ll have to make the get-physical-dev-ids code not output duplicates. [13:52] *nods* [14:43] mbiebl: you were saying something about /forcefsck not working ? [14:44] so anybody has a good dbus-deamon starting service?(that waits for the service to be started?) [14:44] keesj: ? [14:45] not sure what you mean [14:45] something that checks if the daemon is ready [14:45] you don't need to [14:46] I need to sue fork for that? [14:46] right, "expect fork" and "exec dbus-daemon --system --fork" are sufficient [14:47] thanks [14:48] Keybuk: I have noticed myself as well that /forcefsck does not work since I switched to upstart [14:48] Ah, Mount *mnt isn’t nih_alloc’ed. No wonder it segfaulted. :-P [14:58] Keybuk: Yesterday you said that in the expression on foo started that started was a sub-state of foo (or sub-job or whatever)? [14:58] Keybuk: I was wondering if you'd thought of a way to create more of these sub-states === h\h is now known as haraldh [14:58] in terms of syntax etc. [15:00] syntax is the wishy bit [15:00] but yes, you can create them [15:00] you can also create them deliberately inside logs [15:00] err inside jobs [15:00] e.g. apache logrotate [15:01] I have an other meta physical question, suppose I have a dbus server that I want to launch using dbus activation, would'nt it be nice if is wat a upstart server? [15:01] service [15:02] yes [15:02] (so I have respawn and such) [15:02] it's relatively trivial to patch d-bus to use Upstart to start services [15:03] Keybuk: heres a few thoughts then: Suppose the existence of a stanza called "exclude" which takes as its argument a few sub-states. It prevents those sub-states from being true at the same time [15:04] why would you want that? [15:04] and how would such a thing be resolved? [15:04] Keybuk: if you tried to start one state it would stop any of the others in the list [15:04] Keybuk: the "why" I'm getting to :) [15:04] isn't that the same as "while not " ? [15:04] starting that implies that has to be stopped [15:05] (for the same reason that starting "while " implies that has to be started) [15:05] Keybuk: for two states yes. for N or more its a bit more difficult. Plus this is a push maneuver. It implicitly /makes/ it true [15:05] assuming you're talking about the physical "start" command [15:05] which is the only case where we want that kind of behaviour [15:05] I don't think we ever want auto behaviour like that [15:06] Keybuk: ah, but we do, let me continue [15:06] could you explain on the ML? [15:06] I'm really busy at the moment [15:06] Keybuk: sure. It'll be out of context there since we started the discussion here, but that's our fault to begin with anyway :) [15:07] you can provide context ;) [15:09] Keybuk: not to mention I'll see you in two weeks anyway :) === robbiew-afk is now known as robbiew [15:19] indeed [15:34] keybuk: Okay, now everything seems to work – i have tested the fsck queue as well (in a virtual machine). Patches updated. [15:35] awesome [15:36] Actually, i just noticed a missing nih_assert that should be there. Patches updated again. :-P [15:36] ion: can you get rid of the LIKELY/UNLIKELY bits [15:36] Okay [15:37] also no need to nih_warn() on repeating update_mount_dev_ids [15:37] if it's not supposed to happen, assert [15:37] if it's safe, just return [15:38] actually, it's quite easy for that to be called multiple times [15:38] e.g. a disk change for an md device [15:38] (would you need to recalculate then, I wonder?) [15:40] update_mount_dev_ids is called just before fsck is spawned, and the dev_ids are only used for fsck locking. [15:40] it looks like you call it in the udev device watcher [15:40] in which case [15:40] say the fsck is queued [15:40] and a new update for the udev device comes in [15:40] you'd need to dequeue the existing fsck (I don't think you even check for it right now) [15:40] rescan the ids [15:40] then queue again [15:41] Do md or lvm partitions ever become visible if they are not ready/fully initialized? [15:41] yes [15:41] though the watcher thing checks for that [15:41] but they can *change* remember [15:41] imagine: [15:41] boot, mdadm initialised, in degraded [15:42] user notices while fsck of / is happening [15:42] (md is /var maybe) [15:42] and pulls the disk out and replaces it [15:42] you'll get a change for the queued md [15:43] forcefsck - not sure this hook is right [15:43] mountall gets passed this as --force-fsck on the command-line, it doesn't check for the file [15:43] removing that should go in post-stop of the conf I think [15:44] Ok === robbiew1 is now known as robbiew [15:50] so is there anyway to start a service using 'initctl emit'? [15:50] PuffTheMagic: yes. initctl emit where the service is defined with start on [15:51] oh, but no equiv of 'initctl start ' [15:51] PuffTheMagic: no [15:51] :( [15:51] PuffTheMagic: why? [15:52] cause stupid palm dev had to use upstart 0.3.x on the Pre.... [15:52] so there is no dbus interface [15:52] and i wanted to make a gui [15:52] for starting stopping services [15:53] so i have to make a service that wrapps around upstart/initctl [15:53] and relays stuff over dbus [15:53] and i wanted to just use events [15:53] but i guess i could just call initctl start/stop [15:53] just seems ugly [15:55] is there a libupstart or anything in the older upstarts that will give me more functionality? [15:55] PuffTheMagic: there was a time when such a thing as libupstart existed. Not sure if its still rattling around in that code. [15:56] well im using 0.3.8 [15:56] PuffTheMagic: still. [15:57] yes, there's a libupstart in 0.3.8 [15:58] (it's how initctl talks to upstart) [15:58] good [15:58] so i can use that to do start/stop :D [15:58] sadmac2: *cough* isn't "that code" what you're going to be supporting for the next seven years? :p [15:58] haha [15:59] Keybuk: 7? You're some crazy kinda optimist. I'm betting on it being around in 2040. [16:00] I'll retire before 0.3 does. [16:00] you could always ship 0.6 instead :p [16:00] then, when you have some problem with it, I probably won't just point and laugh [16:00] so the palm devs must have static build initctl then cause i see no libupstart on this userland [16:00] how lame [16:01] PuffTheMagic: upstart always static-builds initctl and init [16:01] Keybuk: you overestimate my authority on these issues. notting is the one to bug. maybe plautrba. [16:01] oh [16:01] Keybuk: 8ball says don't count on it though :( [16:01] Keybuk: I'm support engineering now though. I don't build the product. I just get stuck with the consequences. [16:02] Keybuk: didn't land in time (heck, i still haven't seen any patches for integrating it yet, and i do not have time to look at it myself right now) [16:03] I guess you're not using it any differently than in F10? No plans to convert more jobs over? [16:03] nope. but i'd assume it's still not a no-op to land it [16:03] notting: its missing the state transfer thing, which means the ttys-don't-respawn thing will come back. other than that its not hard. [16:04] not a no-op, but not hard. [16:04] Keybuk: I wasn't planning on moving anything until 1.0 [16:04] just gonna have to fiddle with them again then anyway. [16:04] 1.0/0.10 depending on how stable we feel. [16:04] * Keybuk is moving things now, it's fun [16:38] keybuk: Patches updated. [16:46] Shutdown seems to be delayed by perhaps 10 seconds by this: http://heh.fi/tmp/upstart-rc-stop-cont-stop-cont [16:50] LVM seems to be nice enough to compute which physical partitions each logical volume spans, and only report those partitions in /sys/.../slaves instead of e.g. every single partition the volume group uses. [18:25] keybuk: Did you receive the above messages? [18:26] ion: I can't answer that question ;) [18:27] shutdown: the "10 seconds" is killall5 running [18:27] the upstart output is just a side-effect of what killall 5 is doing [18:27] some other process is the one not stopping [18:29] Ok [18:29] Why SIGSTOP? [19:04] Can you review the new patches, btw? === robbiew1 is now known as robbiew [22:26] ion, Keybuk: who is responsible for mountall? [22:26] Dunno about responsible, but Keybuk is the author. :-P [22:27] ok, there is a bug wrt to ntfs mounts [22:27] I have a ntfs-3g auto mount in my fstab and subsequently mountall fails [22:27] which stops the boot process [22:28] I think there are two issues here: 1.) failing to mount ntfs(-3g) mount points [22:29] 2.) the mount point is non-critical (mounted as a shared partition under /mnt/data) [22:29] the boot should not fail because mount failed for a non-critical mount point [22:34] Keybuk: and a more cosmetic issue: the output of mount differs between the old-style mounting using the sysv initscripts and mountall [22:34] http://pastebin.ca/1561541 [22:50] Keybuk: btw, is there a bug tracker for this sort of issues or should I send it to the m-l or via pm? === robbiew1 is now known as robbiew === robbiew is now known as robbiew-afk