Performance issues with MS and SA

[Steve Freegard]

On 30/03/11 18:00, donald.dawson at bakerbotts.com wrote:
> I would like to get some input from the community on our performance issue.
>
> Running 'MailScanner --debug --debug-sa' points to DCC as a bottleneck
> and Bayes.
>
> Turning off Bayes and DCC makes a big difference, but I would like to
> keep DCC.
>
> Here's are our settings in spam.assassin.prefs.conf:
>
> dns_available yes
> bayes_path /var/spool/spamassassin/bayes
> bayes_ignore_header X-BakerBotts-MailScanner
> bayes_ignore_header X-BakerBotts-MailScanner-SpamCheck
> bayes_ignore_header X-BakerBotts-MailScanner-SpamScore
> bayes_ignore_header X-BakerBotts-MailScanner-Information
> lock_method flock

Add:

bayes_learn_to_journal 1

This will give a big speed boost for file-based bayes.

Also add:

bayes_auto_expire 0

And see my comment below about running expiry as a cronjob to improve 
bayes performance.

> ifplugin Mail::SpamAssassin::Plugin::DCC
> dcc_path /usr/local/bin/dccproc
> endif

Read up on 'dccifd' and use that instead; it's *far* quicker if you want 
DCC running.

> Always Include SpamAssassin Report = yes

Consider setting this to 'No'; it will save some considerable cycles. 
e.g. for whitelisted messages etc.

> Rebuild Bayes Every = 86400
> Wait During Bayes Rebuild = yes

Doing a rebuild like this will cause some serious performance issues if 
your mail servers are busy, leading to backlogs.

Set this to no and disable this function entirely; then create a cronjob 
to run at a quiet hour that runs once per day:

sa-learn --force-expire

Hope that helps.

Kind regards,
Steve.