10.6.3 - Apparent spontaneous unmount of network volumes

Question

Since updating my server to 10.6.3 I am experiencing spontaneous unmounts of AFP volumes from clients running both 10.6.2 and 10.6.3. In both cases the Console on the client systems reports the following messages:
02/04/2010 3:07:53 am kernel AFP_VFS afpfsCacheLookup: got 2 on root vnode, unmounting vol
02/04/2010 3:07:53 am kernel AFP_VFS afpfs_unmount: /Volumes/XXX, flags 524288, pid 626
(Obviously the PID on the second system is different.)
In both cases there was a Time Machine backup happening during this occurrence which commenced prior to the unmount, but there had been many previous backups which did not trigger an unmount so I am unsure whether this is significant or not.
The only other clue is that on the 10.6.3 client (but not the 10.6.2 client) the following two lines appear in the console immediately prior to the unmount event:
02/04/2010 3:07:53 am KernelEventAgent[40] tid 00000000 type 'afpfs', mounted on '/Volumes/XXX', from 'afp_000000004oMw0oYHtK3gzaWf-2.2e000004', dead
02/04/2010 3:07:53 am com.apple.KernelEventAgent[40] KernelEventAgent: sysctlunmount(afp000000004oMw0oYHtK3gzaWf-2.2e000004): Resource busy
This is definitely becoming an issue for me which I desperately need to resolve.
One other symptom I am experiencing is that both clients seem to repeatedly "lose" their mail POP3 passwords and these need to be re-entered to retrieve mail. I suspect what's actually happening is that authentication is incorrectly failing on the server, triggering the request to re-enter the password, and I wonder whether this is a factor in the AFP unmounts. What's significant here is that this is happening on both the 10.6.2 and 10.6.3 clients; the only common factor is that the server has been updated to 10.6.3.

Answer 1

Latest situation is that the problem is still ocurring regularly. I have tried repairing permissions on the server but this has not changed anything. It is very clear to me that this is some sort of interaction between Time Macine and mounted volumes. Two clients have both just unmounted their network volumes in exactly the same way - the Time Capsule volume has been mounted and then immediately the network volumes are unmounted. Here is a transcript from one of them:
28/04/2010 2:36:48 pm com.apple.backupd[40394] Starting standard backup
28/04/2010 2:36:48 pm KernelEventAgent[41] tid 00000000 type 'afpfs', mounted on '/Volumes/Vol1', from 'afp_000000004oMw0oYHtK3gzaWf-2.2e000047', dead
28/04/2010 2:36:48 pm com.apple.backupd[40394] Attempting to mount network destination using URL: afp://[email protected]/OurTimeCapsule
28/04/2010 2:36:48 pm com.apple.KernelEventAgent[41] KernelEventAgent: sysctlunmount(afp000000004oMw0oYHtK3gzaWf-2.2e000047): Resource busy
28/04/2010 2:36:49 pm com.apple.backupd[40394] Mounted network destination using URL: afp://[email protected]/OurTimeCapsule
28/04/2010 2:36:49 pm kernel AFP_VFS afpfsCacheLookup: got 13 on root vnode, unmounting vol
28/04/2010 2:36:49 pm kernel AFP_VFS afpfsCacheLookup: got 13 on root vnode, unmounting vol
28/04/2010 2:36:49 pm kernel AFP_VFS afpfs_unmount: /Volumes/Vol2, flags 524288, pid 629
28/04/2010 2:36:49 pm kernel AFP_VFS afpfsCacheLookup: got 13 on root vnode, unmounting vol
28/04/2010 2:36:49 pm kernel AFP_VFS afpfs_unmount: /Volumes/Vol1, flags 524288, pid 629
28/04/2010 2:36:49 pm kernel AFP_VFS afpfs_mount: /Volumes/OurTimeCapsule, pid 40400
One other thing that I have noticed recently (and seen on other systems) is that systems on the network seem to regularly get new names assigned to them with -1 or -2 appended. I don't know if this is connected to the problem or not, but it certainly could be part of the same syndrome. I get the impression that there's something going on somewhere in the code that handles network resources, maybe within Bonjour?
One thing that I find highly frustrating is that even if you remount the network volume many apps (such as the MS Office 2008 and Adobe CS4 apps) crash and burn if they have a file open from it, and often in a somewhat untidy way - eg Word displays a dialog saying that the file may be corrupt with an OK button but if you press this the dialog disappears and reappears again straight away. there is no way to get out of this except by force-quitting, even though the app is actually running fine apart from being in an apparent deadlock.