IMAP/LMTP/Quota locking problem

[Scott Adkins]

--On Tuesday, May 18, 2004 9:13 AM +0200 Sebastian Hagedorn 
<Hagedorn at uni-koeln.de> wrote:

> thanks for the info. The patch was definitely added *after* 2.2.1 ...
> hmm, now that I think about it: seems like it was even added after 2.2.3!
> Is bug 1270 really fixed in CVS only?

Good question! :) I hope it is fixed :)

If I remember right, it has something to do with SSL.  Are you using SSL?
Basically, the quota file gets locked and stays locked because the IMAP
process (which is usually the one that has the quota file open for write)
is trying to do a read on an SSL socket (this is from memory and from what
I have seen in our traces).  The quota lock problem is just a side effect
of that issue...

>> Currently, we manage the problem by watching the sendmail syslog file for
>> System I/O errors on the Cyrus mailer (grep for
>> 'mailer=cyrus.*System.I.O')
>
> We don't get those. I wonder if we actually have separate problems. Or
> maybe not. However, in our case the sendmail processes get stuck as well,
> maybe due to our setting of DeliveryMode to interactive!?

I find that interesting... after an IMAP process locks the quota file, the
LMTP processes start to stack up.  After awhile, anything trying to be
delivered to that user generates a System I/O error.  It makes it easy to
catch, since as soon as we see a few errors for a particular user, we run
fuser on that user's quota file to get a list of processes, run lsof on
that list of processes to find the process that has the quota file open
for write and we kill it.  The next entry in the syslog should show that
a message was sent instead of deferred, which means we fixed the problem.

>> The problem occurs regularly, with some users more reliably triggering
>> the problem than others.
>
> What does "regularly" mean? We've been running the server in prdouction
> since March, 1, but it happened for the first time a little more than a
> week ago ...

On average, I would say at least once every day or two.  Sometimes it gets
worse and we get a few in a day, and sometimes, a single user may do it a
couple times in a day... It does seem like some users trigger it a lot more
than the majority.  This suggests some kind of interaction between the IMAP
process and the e-mail client, possibly back to that SSL read issue again.

Scott
-- 
 +-----------------------------------------------------------------------+
      Scott W. Adkins                http://www.cns.ohiou.edu/~sadkins/
   UNIX Systems Engineer                  mailto:adkinss at ohio.edu
        ICQ 7626282                 Work (740)593-9478 Fax (740)593-1944
 +-----------------------------------------------------------------------+
     PGP Public Key available at http://www.cns.ohiou.edu/~sadkins/pgp/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 231 bytes
Desc: not available
Url : https://lists.andrew.cmu.edu/mailman/private/info-cyrus/attachments/20040518/3152f42a/attachment.bin