Le 18/05/2012 07:22, Stan Hoeppner a écrit :
> On 5/17/2012 11:44 PM, /dev/rob0 wrote:
>> [snip]
>> Note that you must disable restrictions on the reinjection from
>> content_filter. You do NOT want to reject at that point, because it
>> is risking backscatter.
>
> Of course. But this is mostly a concern only when using SMTP for
> reinjection. I'm currently looking at using the sendmail command for
> reinjection.
but make sure you don't enable a "global" content filter. set a
-o content_filter=
under the pickup service.
> So I should only possibly need to disable header_checks in
> the master.cf service definition, though after a quick look they won't
> pose a problem if left enabled (but for tiny CPU burn).
>
> I've historically been opposed to using content filters for various
> reasons, as some here may recall me stating, probably mostly on other
> lists. That said, I've recently been toying with the idea of taking SA
> for a test drive, using spamc/spamd and sendmail reinjection-the "easy
> method" so some state.
>
currently, the "best" option is to use amavisd-new. this will avoid
forking a child for every message (amavisd-new embeds SA code, so it
load it once).
> A little OT, but I'd like to ask, as I've not been able to find real
> information via Google. What's the memory consumption of a single spamd
> process using the default SA configuration? Maybe a better question is
> how much real RAM is SA eating on systems folks here have in production?
>
it indeed depends on the configuration. I've found that you can
remove/disable many checks. but I never measured how much each config eats.
