Skip net checks if local tests score is high enough?

[Scott Truman <Sc...@ncs.co.nz>]

Hi there,
	I was wondering if there was any option to skip the "net" tests
(RBLs, DCC etc) if the email was scored as spam sufficiently by the
local tests? If there is not, would this be considered a good option for
future releases. i.e. if for example the bayes and regex tests are
enough to score an email over a certain threshold, say 15, then any net
tests are basically a waste of time and resources IMO. "net" tests with
negative values would still get run however, as these may still have
some bearing on the final score.

Also, am I correct in thinking that net tests with a score of 0.0 are
not run at all?

Cheers
Scott

P.S Spamassassin 3.0 is great, thanks.

Re: Skip net checks if local tests score is high enough?

[Daniel Quinlan <qu...@pathname.com>]

Theo Van Dinter <fe...@kluge.net> writes:

> I was actually chatting with another developer in IM about this very
> idea last night.  It sounds like a good idea, except, that it's more
> efficient to launch the RBL tests in the background at the start, then
> run the local tests, then check the network test results.

Sure, but we don't have to wait even longer on network test results if
the score exceeds some higher threshold.  More importantly, we don't
have to run DCC, Razor, and Pyzor if the score exceeds some threshold --
those tests can't be started in the background, at least not in our
current implementation (which needs to be revisited anyway).

> The likelihood is that you'll have to do network checks on more of your
> mail than the amount the short cut would be used on, so you're likely
> to end up spending more time per message since you'll always be waiting
> to start the network checks.

Nah, that's not the problem, we'll always want to start RBL queries
early, but even now, we don't wait for the slowest ones.  The problem we
do have to be more careful about is that the repeated check to see if
you can exit early can end up costing more than the benefit.  The
original sort by score code had that problem, mostly because it was
poorly designed.

Daniel

-- 
Daniel Quinlan
http://www.pathname.com/~quinlan/

Re: Skip net checks if local tests score is high enough?

[Theo Van Dinter <fe...@kluge.net>]

On Sun, Jul 04, 2004 at 02:09:31PM +1200, Scott Truman wrote:
> 	I was wondering if there was any option to skip the "net" tests
> (RBLs, DCC etc) if the email was scored as spam sufficiently by the
> local tests? If there is not, would this be considered a good option for

Nope.  There is no short-cutting code at all right now.

> enough to score an email over a certain threshold, say 15, then any net
> tests are basically a waste of time and resources IMO. "net" tests with

I was actually chatting with another developer in IM about this very
idea last night.  It sounds like a good idea, except, that it's more
efficient to launch the RBL tests in the background at the start, then
run the local tests, then check the network test results.

The likelihood is that you'll have to do network checks on more of your
mail than the amount the short cut would be used on, so you're likely
to end up spending more time per message since you'll always be waiting
to start the network checks.

> Also, am I correct in thinking that net tests with a score of 0.0 are
> not run at all?

Right.  Any test with a 0 score is not run.

-- 
Randomly Generated Tagline:
"And the next time you consider complaining that running Lucid Emacs
 19.05 via NFS from a remote Linux machine in Paraguay doesn't seem to
 get the background colors right, you'll know who to thank."
 (By Matt Welsh)

Re: Skip net checks if local tests score is high enough?

[Matt Kettler <mk...@comcast.net>]

At 02:09 PM 7/4/04 +1200, Scott Truman wrote:
>         I was wondering if there was any option to skip the "net" tests
>(RBLs, DCC etc) if the email was scored as spam sufficiently by the
>local tests?

No, but it gets mentioned at least 4 times a year.

The biggest problem is that SA launches the DNS queries before it runs the 
local body tests. After all, they take the longest, and you may as well do 
something useful while waiting for the results to come back.

SA could be re-coded to abort waiting after the body is done and score is 
over xx points, but I don't know that this would buy you very much.


>Also, am I correct in thinking that net tests with a score of 0.0 are
>not run at all?

Yes, you are correct. All tests with score of 0.0 are not run, including 
network rules. (there is a minor exception of rules named with double 
underscore at the start (__), which are unscored sub-tests for meta rules)