https://bugzilla.novell.com/show_bug.cgi?id=805420
https://bugzilla.novell.com/show_bug.cgi?id=805420#c3
L. A. Walsh changed:
What |Removed |Added
----------------------------------------------------------------------------
Summary|systemd doesn't order |12.1 boot doesn't correctly
|dependencies -- allowing |order dependencies:
|localfs to exec before lvm |specifically localfs has no
|finishes finding volumes |dependency on lvm
--- Comment #3 from L. A. Walsh 2013-03-11 11:14:56 PDT ---
I traced this down some more, and it looks like the problem is in the
/etc/init.d/boot.localfs script where boot.lvm (among others) is listed
under 'Should-start' instead of required-start. Apparently this allows
the boot-time software to run boot.localfs in parallel with the *start* of
boot.lvm (whereas boot.lvm should be finished before boot.localfs).
It looks like maybe the should-starts were designed to have been started
and finished before boot.localfs runs, but instead, they are simply being
scheduled for 'start'.
The device-mapper is part of the lvm package... it only starts .07s before
localfs tries to start.
service boot.device-mapper start
service boot.localfs start
Activating device mapper...done
service boot.device-mapper done
Waiting for /dev/Home+Space/Home
/dev/Home+Space/Share
/dev/Home+Space/Home.diff
/dev/Media/Media
/dev/Home+Space/Squid_Cache
/dev/Backups/Backups
/dev/Home+Space/Media_Back .............................. timeout!
====trimmed beginning of lines:...
<39:32.829649000> service boot.localfs done
<39:32.830172000> service boot.cleanup start
<39:32.842134000> service boot.klog start
<39:32.845642000> service boot.localnet start
<39:32.846799000> service boot.lvm start
<39:37.442276000> service boot.lvm done
--- Noticed that boot.lvm isn't even started until after boot.localfs is done.
and that boot.lvm takes almost 5 seconds by itself to finish.
=====================
I confused the problem with systemd, as whenever I tried to change/insert boot
scripts, I got the message that the boot scripts were shadowed by systemd
and that the request(s) for changes was handled by it.
It turns out, AFAICT, that message is bogus.
-----------------
Carrying this forward -- as systemd does replace these scripts -- it needs
to be sure it waits until all the local disks are mounted before it starts.
systemd should not assume that all disks are mounted when it starts unless
it *checks* for them (i.e. it has targets that can check for these things and
they should be used).
It shouldn't make the assumption that something else handled everything before
it was loaded -- else the same problems can occur.
Note, I've seen lvm take over 15-20 seconds to perform volume building when it
had multiple (3-4) snapshots going ON the same volume and >30 seconds when more
than 5-6 snapshots are active. Seems to go up geometrically or exponentially,
not sure which.
I don't do such because of the problems it causes, but even 1 active snap --
which well falls into normal usage can cause a 5 second delay...
--
Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.