Comment 101 for bug 818177
Revision history for this message
| Steve Langasek (vorlon) wrote Re: [Bug 818177] Re: boot failures because 'udevadm exit' times out while udevd waits for an already-dead thread | #101 |
On Sat, Oct 08, 2011 at 11:56:50AM -0000, Eduard Hasenleithner wrote:
> Many thanks. My LVM system now starts successfully, albeit still showing
> the 60s hang occasionally. Even the supposed-
> volume is mounted correctly. Just the "/sbin/lvm vgchange -a y" persists
> in the process list, but this apparently does not harm.
Wow, I didn't expect that to be the case! I still want to keep the other
bug report open and pursue the "hung vgchange" problem, anyway; leaving
stray processes running in the initramfs after boot is untidy.
> One additional note I want to make: I'd set the udevadm-control timeout
> at least to 70 seconds. The epoll_wait timeout is reset every time
> something happens. So if the "settle time" is bigger than 1 second, the
> total timeout would be more than 61 seconds of udevadm-control. Setting
> the timeout to 70 seconds, gives us at least 10 seconds "settle time".
Ah, that's a good point. Of course, 70 seconds isn't guaranteed to be
enough time either; in the pathological case, you need 60*num_workers
seconds, because if you get one worker finishing every 59.9 seconds, the
udevd timeout will continue to be extended and we'll still have the udevadm
control problem.
But 61 seconds should be enough for the 99.99% case, and if we start to see
reports of problems in the other .01% we can adapt by raising the timeout.
(I would prefer not to just '|| true' the udevadm call, though, since I
think having to occasionally reboot the machine again is better than being
left with extra udev processes running around doing who-knows-what on the
system.)
> My conclusion is that "udevadm control --exit" (introduced in april) is
> still a bit immature, and further races might wait to happen.
It's true that there might be further races, but I don't think it's due to
immaturity of udevadm control --exit... rather, the races were always there
and we were blissfully ignoring them with pkill (and that's where the bug
report came from that led us to switch to udevadm in the first place).
--
Steve Langasek Give me a lever long enough and a Free OS
Debian Developer to set it on, and I can move the world.
Ubuntu Developer http://
<email address hidden> <email address hidden>