>From: Reindl Harald <h.rei...@thelounge.net>
>Sent: Saturday, March 28, 2015 6:13 AM
>To: users@spamassassin.apache.org
>Subject: Re: Uptick in spam

>Am 28.03.2015 um 12:04 schrieb David Jones:
>> I know that but I choose to use the "traditional" method in the Postfix
>> smtpd_recipient_restrictions so I can specify the order.  I have such a
>> high volume of mail for more than 100,000 mailboxes, I want to check
>> in a specific order using my local rbldnsd feed to prevent abuse of other
>> RBLs further down the list

Thank you for the recommendation and I will research the impact that
my high volume mail filters would cause to other RBLs that I do not
have a local rbldnsd feed for.  I have a local caching DNS server pointed
to a set of private DNS servers hosting my rbldnsd zones so the impact
should be as low as possible to the "external" RBL lookups.  I have to be
mindful of their free use limitations and abuse policies.  (I have received
emails from a few of them for excessive usage and had to discontinue
using those.)

>the problem with this approach is that with each RBL you raise the
>false-positive rates extremely, until now i did not see any RBL without
>FP be it Zen, Barracuda or Spamcop

You are correct.  This method does give complete power to each RBL
to reject a message.  If there were a way to specify the order of RBL
checks then I could eliminate this problem.  I will research this.

>another thing is performance: "smtpd_recipient_restrictions" is
>sequential while postscreen asks all RBLs parallel, if one or more have
>a timeout it don't block, they are just not taken into account at that
>moment, when you have enough RBL's the result is still good

I have very fast, low latency connections to the Internet so speed is not
my problem.  My typical batch processing time (30 emails) is under 5
seconds in MailScanner which is very good running 2 AV scanners.
Postfix is a tiny fraction of that processing time and most of it is AV
and SA.  In SA, I have DCC (local DCC peer), Razor, Pyzor, Bayes in a
redis DB, CRM114, and BOGOFILTER enabled.  I have tuned SA from
taking around 30 seconds to under 4 seconds per batch using safe
shortcircuit rules and safe whitelist_from_* entries.

The only spam I have a problem with is from compromised accounts
for the first 30 minutes or so until RBLs kick in.  I am still able to block
most of the compromised account spam.  I know I could turn on grey-
listing and help with this but I feel that greylisting is not worth the delay
_in our environment_ for the small gain that I would get.  I want to look
into selective greylisting when I get some time to build it out properly
for our environment that is acceptable for our customers.

Reply via email to