Hi!

On Wed, 2009-07-22 at 15:22 +0200, Bernd Petrovitsch wrote:
[...]
> I'm in the process of upgrading cyrus-imapd-2.2.13 (from
> Debian-3.1/Sarge) to cyrus-imapd-2.3.7 (from RHEL-5/CentOS-5) on a
> system with approx. 25k mailboxes.
> 
> We are using quite simple sieve scripts for vacation and 2.3.7
> duplicates (into the local inbox) all emails which trigger an outgoing
> vacation mail.
> Glancing over the source diff'ing the old and new imap/lmtp_sieve.c file
> (and diff'ing to 2.3.14) doesn't reveal anything remotely strange (to my
> eyes - I'm not experienced with cyrus-imapd's source code. So I may well
> looked into the wrong file.).
It was the correct file as it calls "sendmail" (configureable
in /etc/imapd.conf) there to send out the vacation mail.

> Googling and searching in the bug/issue trackers from above also doesn't
> give any useful hint - and I didn't even found someone else having that
> problem.
Please find the solution at the end:

> FWIW, we have "duplicatesuppression: false" in /etc/imapd.conf - as it
> saved a lot of I/O load on the old systems (avoiding storing and
> checking the for duplicates in the DB).
This and Cyrus-Imapd had nothing to do with it - at least for the root
cause. The symptoms are IMHO strange nonetheless.

Debugging the issue leads to: the above mentioned "sendmail" call is in
my case the (real) sendmail from the RHEL/5/CentOS-5
sendmail-8.13.8-2.el5 package. "sendmail" returned with an exitcode of
65 - "data format error" according to sendmail's source and the erros
was reproducable on the shell. It was *not* caused by illegal characters
in headers or some other fault in the to-be-sent mail (at least I and an
other pair of eyes didn't find anything faulty).
The root cause was (to the best of my understanding - please correct me
of/when/where I'm wrong): The called sendmail acted as a MSA and - thus
- used /etc/mail/submit.mc which had at the end
----  snip  ----
FEATURE(`msp', `[$IP_ADDRESS_OF_SMARTHOST]')dnl
----  snip  ----
to deliver the mail directly to the resident smart host.
Changing that to
----  snip  ----
FEATURE(`msp', `[127.0.0.1]')dnl
----  snip  ----
returns 0 (and made everything else magically work as expected). Adding
`MSA' to use port 587 instead of 25 doesn't change anything BTW.
The main difference is that the MSA now connects to the local MTA which
in turn forwards the mail so the smart host (as the sendmail.mc has an
equivalent SMART_HOST() line there).

In short: Obviously I either mis/ab/used sendmails features or something
else is somewhere wrong.
Sorry, but I don't feel like I's like to debug sendmail as such further
(and that would belong to some sendmail ML anyways).

Thanks for all the answers,
	Bernd
-- 
Firmix Software GmbH                   http://www.firmix.at/
mobil: +43 664 4416156                 fax: +43 1 7890849-55
          Embedded Linux Development and Services