[Dovecot] Correct usage of NFS mounted Maildirs

[Fran Fabrizio]

Dovecot 1.0RC15 running on Fedora Core 5 kernel 2.6.18-1.2257.fc5smp.
A SLES10 server (kernel  2.6.16.21-0.8-smp) has the home directories on
an ext3 partition mounted with options (rw,acl,user_xattr,usrquota)
which is then exposed as an NFS mount with server side options
(rw,no_root_squash,no_acl,async).  The home directories include Maildirs
to which postfix delivers.
The Fedora Core 5 dovecot&postfix machine then mounts this NFS mount
with options (rw,hard,intr)

Dovecot variables:
mmap_disable=yes
lock_method=dotlock
default_mail_env =
maildir:~/Maildir:INDEX=/usr/local/dovecot/indexes/%u   (only one
dovecot server and wanted to see if local indices helped)

I didn't turn off the attribute caching of NFS since I only have one
dovecot server, I think the wiki said it was safe to leave on in this case.

With this configuration, I am seeing a lot of NFS Stale Filehandle
errors, even after moving the index files to a local disk.  Here's an
excerpt from the log:

dovecot: Dec 22 11:23:06 Info: Dovecot v1.0.rc15 starting up
dovecot: Dec 22 11:23:22 Info: imap-login: Login: user=<nfsuser1>,
method=plain, rip=138.26.64.50, lip=138.26.64.5
dovecot: Dec 22 11:23:24 Info: imap-login: Login: user=<nfsuser1>,
method=plain, rip=138.26.64.50, lip=138.26.64.5
dovecot: Dec 22 11:24:48 Info: imap-login: Login: user=<nfsuser1>,
method=plain, rip=138.26.64.50, lip=138.26.64.5
dovecot: Dec 22 11:26:43 Error: IMAP(nfsuser1):
inotify_add_watch(/nethome/facstaff/nfsuser1/Maildir/.Trash/new) failed:
Stale NFS file handle
dovecot: Dec 22 11:26:43 Error: IMAP(nfsuser1):
inotify_add_watch(/nethome/facstaff/nfsuser1/Maildir/new) failed: Stale
NFS file handle

To be honest, I'm not 100% sure this is a dovecot problem, because I
also see that I am getting stale filehandles when postfix tries to
deliver the mail to the Maildir, but I'm curious as to whether this is
because of something dovecot did because I have it misconfigured.

Does anything about my dovecot config provide any clues as to what I
might have done wrong?  The NFS mounts are completely stable everywhere
except within the Maildirs.  From the shell, I can run a loop to call
'ls' 1000 times outside of a Maildir and it works fine.  Inside a
Maildir, I will get Stale NFS Filehandle errors before long.  Does this
sound like it could have something to do with my Dovecot config, given
that it is happening even after I moved indices to the local disk (still
getting stale filehandle errors on files like dovecot-keywords).  I
performed the following procedure:

- shut down dovecot
- did a rm-rf Maildir on my test account
- unmounted the NFS mount /nethome/facstaff
- remounted the NFS mount /nethome/facstaff
- set the index files to be local in dovecot.conf and created that
directory (/usr/local/dovecot/index/)
- sent a few test mails to the user (to recreate the Maildir)
- started dovecot
- logged in (via Thunderbird) to account through dovecot.
- read test message
- deleted test message
- opened shell to dovecot server
- cd'ed into user's Maildir
- ran 'ls' a few times.  Got Stale Filehandle errors after a few.
- back in Thunderbird, tried to read/move/delete other test mails, got
the above errors in the log.

If it's not a dovecot issue I know it's off-topic, but has anyone else
seen similar?  I might be imagining this, but I think the problem
started when I moved from SLES9 to SLES10 for my fileserver (newer
kernel, newer version of NFS, perhaps an incompatibility here?) 

Thanks,
Fran