-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Andy Jezierski writes:
> Andy Jezierski <[EMAIL PROTECTED]> wrote on 09/13/2004 10:13:01 AM:
> 
> > I've got a question as to whether you think this is a milter error or a
> > spamd error.  I'm thinking milter.
> >
> > I'm running milter-spamc 0.24 along with sendmail 8.13 on one box, which
> is
> > calling spamd 3.0rc3 (haven't had a chance to go to rc4 yet) running on
> > another box.  Everything works just fine, except, when I run an sa-learn
> on
> > the box running spamd.
> >
> [snip]
> 
> I got a reply from the milter author saying this is a problem with
> spamd....
> 
> > Sep 13 09:36:31 viper milter-spamc[40625]: 34482 i8DEYOCH096305: timeout
> > before input from SPAMD server
> > Sep 13 09:36:31 viper milter-spamc[40625]: 34482 i8DEYOCH096305: SPAMD
> > status line failure
> > Sep 13 09:36:44 viper milter-spamc[40625]: 34486 i8DEYake096312: timeout
> > before input from SPAMD server
> > Sep 13 09:36:44 viper milter-spamc[40625]: 34486 i8DEYake096312: SPAMD
> > status line failure
> 
> This is the problem with spamd, its not truely threaded and gets hung up
> when ever more than one process (spamd, sa-learn, spamassassin) access
> the token/database files.
> 
> The second set up errors from sendmail could be avoided by increasing
> the sendmail-milter timeout in the INPUT_MAIL_FILTER line, but this is
> not guaranteed solve overall problems with SA.
> 
> Essentially, the problem lies with SA perl bloat and lack of decent
> threading.
> 
> You could try increasing the milter-spamd timeout, but if you do this
> you must increase the sendmail-milter timeout to account for this.
> 
> Should I open a bugzilla on this or is there no way around the problem for
> now?

what's the time spent waiting for spamd?  what's the milter's default
timeout?

It sounds a lot like SpamAssassin's potential latency (up to 10 seconds if
a db operation is in progress, waiting for the db to be unlocked; possibly
higher if network tests are being used; plus a few seconds for the non-db,
non-network tests) is higher than the milter expects to have to deal with. 

This is not a bug; spamc, for example, can wait for up to 10 minutes for
spamd to complete before giving up.  that's pretty extreme though.
I don't think I've ever seen spamd take that long.

The correct behaviour of the milter should be to either:
(a) wait for spamd to return a response although it may take a while,
(b) pass the message through and deliver as nonspam, or
(c) queue the message for later checking, probably not possible
from a milter.

(I'm disregarding the ignorant comments about "bloat" and threading.)

- --j.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (GNU/Linux)
Comment: Exmh CVS

iD8DBQFBRy82QTcbUG5Y7woRAvK4AKC6fe4Mz819THlGlCfcQu2rlqnj6gCcDEP/
E6X5LLpjm/vdTl024tMxdxw=
=DhOT
-----END PGP SIGNATURE-----

Reply via email to