[Ocfs2-users] Postfix cluster with OCFS2 mboxes?

[Robert Edmonds]

Hi,

I'd appreciate hearing from anyone who is running a Postfix cluster with
shared mboxes mounted on an OCFS2 volume.  (It looks like directory
entries aren't indexed/hashed on OCFS2, so this rules out Maildirs --
please correct me if this has changed.)  Here follows some of my
notes/questions:

I have a four node VMware team set up running a fairly untuned Postfix
2.2 setup on Debian testing.  (Presumably there's nothing wrong with
running OCFS2 on non-commercial distros that haven't been blessed by
Oracle.)

Access to shared storage is via AoE (vblade on another machine on the
LAN).

On Debian's 2.6.16-2 kernel I was able to trigger a BUG on one of the
nodes while stress-testing Postfix, see:

    http://edmonds.ath.cx/ocfs2_node1_log.txt

Is this anything dangerous?  I've upgraded to Debian's 2.6.17-2 kernel
and I haven't been able to trigger it so far.

Under stress-testing (i.e., mailbombing) I see a certain percentage of
mail getting deferred but eventually delivered:

Aug 20 20:23:00 node2 postfix/virtual[3696]: 3D04C54BE3:
to=<user1 at test.localdomain>, relay=virtual, delay=9240, status=deferred
(mailbox /var/mail/test.localdomain/user1: unable to lock for exclusive
access: Resource temporarily unavailable)

Is there any way to reduce this lock contention?  I assume this will be
less of an issue when running on non-virtual hardware with faster I/O
and a far smaller mail load?

I also see to a lesser degree messages in syslog of this form:

    (2703,0):ocfs2_extent_map_insert:603 ERROR: status = -17

Where the only variant is the first number in the tuple.  What does this
mean?  Is it harmless?

Thanks!

-- 
Robert Edmonds