Re: Performance: the "ms" number in the amavisd log

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

Dale Carstensen wrote:

> And there's a second digression.  As I was browsing
through the
> amavisd log to see what recent "ms" times
have been (many over 130000,
> some even higher) I noticed a "SA TIMED OUT"
line.  Hmm, isn't that
> by default 300 seconds (5 minutes?) 

It's 30 seconds by default.  You can increase that by
adjusting the
$sa_timeout parameter in your amavisd.conf file.  Increasing
that to 60
seconds is reasonable, but if you have to boost it much
higher than that
to avoid the timeouts you should probably consider more
capable hardware
to handle the load.


> So I ran it with -D.  It says there's something invalid
about storing
> data in type bytea columns.

Make sure, first of all, that you're using PostgreSQL 8.x
(the minimum
requirement), and that you're using the PostgreSQL-specific
engine for
SpamAssassin (Mail::SpamAssassin::BayesStore::PostgreSQL
rather than the
generic SQL engine).

Second, make sure you've added the put_tokens(),
greatest_int(), and
least_int() functions, as found in the bayes_pg.sql schema:

<http://search.cpan.org/src/JMASON/
Mail-SpamAssassin-3.1.7/sql/bayes_pg.sql>

This was supposed to be added to the maia-pgsql.sql schema
prior to
1.0.2, but evidently got overlooked.  I've opened a ticket
(#420) to
make sure it doesn't get missed again.

- --
Robert LeBlanc <rjlrenaissoft.com>
Renaissoft, Inc.
Maia Mailguard <http://www.maiamail
guard.com/>

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)

iD8DBQFFu9s4GmqOER2NHewRAijpAKCCCHgNZnbBxohhDuoY+SSvuImHHQCf
Y+Bc
eA9LWY37J6+3yhSU9PGxBRE=
=bDUP
-----END PGP SIGNATURE-----
_______________________________________________
Maia-users mailing list
Maia-usersrenaissoft.com
http://www.renaissoft.com/mailman/listinfo/maia-users

All I've done to tell spamassassin to use postgres are the
_dsn
settings in local.cf.  I get the impression that isn't
sufficient.
What did I miss?

The 3 functions (greatest_int, least_int, put_tokens) for
postgres
are in the database.  I think I got them from the apache
spamassassin
3.1.x page about DB AWL/Bayes, and I went by the README
files, there,
too, or at least I read them and re-read them as I performed
the
surgery.

 ht
tp://svn.apache.org/repos/asf/spamassassin/tags/
  spamassassin_current_release_3.1.x/sql/

However, the last usage timestamp for the perl file
Mail/SpamAssassin/BayesStore/PgSQL.pm is the installation
time, January
7, while its sibling SQL.pm is less than twenty minutes ago
(although
right now the Bayesian is running on Berkeley DB, I thought,
so I don't
know why it wouldn't be early this morning.)

The postgresql version is 8.1.4 on the slower system where
I'm
attempting this first.  On the busier but faster one, it's
8.1.3.

I also checked the log on the busier but faster server for
"SA TIMED OUT" lines, and there seem to be just
about 2 per day,
and all the tracebacks show that it's in BayesStore. 
There's
probably a simple explanation for such consistency, like
they
all happen because of a certain cron job or two.  Hmm,
they're
all 12 hours apart, one around 2 AM, give or take 30
minutes,
the other within a minute or two of 12 hours later, around 2
PM,
although one day they were close to 6, not 2.  The rules
from
apache and openprotect are refreshed and amavisd restarted
at
the end of the script that starts at 1:30 AM, but I don't
see
anything that runs 12 hours later.  A minor mystery.

>Dale Carstensen wrote:
>
>> So I ran it with -D.  It says there's something
invalid about storing
>> data in type bytea columns.
>
>Make sure, first of all, that you're using PostgreSQL
8.x (the minimum
>requirement), and that you're using the
PostgreSQL-specific engine for
>SpamAssassin (Mail::SpamAssassin::BayesStore::PostgreSQL
rather than the
>generic SQL engine).
>Robert LeBlanc <rjlrenaissoft.com>


_______________________________________________
Maia-users mailing list
Maia-usersrenaissoft.com
http://www.renaissoft.com/mailman/listinfo/maia-users