Hey. On Tue, 2014-11-11 at 20:01 +0100, Michael Biebl wrote: > We discussed this a bit more yesterday, and we came to the conclusion, > that for jessie, it's probably the simplest solution, if we explicitly > call "udevadm settle" in /etc/init.d/networking before it ifup's any > devices. > A "udevadm settle" will block, until all events in the udev event queue > have been processed. This doesn't necessarily guarantee that all networ > interfaces actuall do exist, it's behvaiour is more like the one in > wheezy under sysvinit. Well not a perfect long term solution... but for the moment perhaps really an idea.
> Attached is a patch against /etc/init.d/networking. > While we discussed yesterday, to only run "udevadm settle" if there are > any auto interfaces, I changed it, to also cover allow-hotplug. > I also changed the init script to handle allow-hotplug interfaces. Which probably explains why with allow-hotplug, I no longer saw services which couldn't bind to their respective addresses. Doesn't that basically mak the ifup@.service useless? > I'll followup with more details later. okay... > First, it would be great if you Christop could test this patch. Both > using allow-hotplug and auto. Done (with the v2 patch, of course). I did about 5-7 boots for each of the two, the network always came up by "itself" (i.e. my cron script never needed to repair things), and in both cases, it seemed that all the services could bind to their addresses. > P.S: The patch also has a downside, i.e. it will serialize early boot > and make it slower. Sure :-( Thanks so far :) Chris.
smime.p7s
Description: S/MIME cryptographic signature
