[Gllug] Configuring SPF to cope with secondary incoming mail servers

[Nix]

On 4 Dec 2006, Jason Clifford outgrape:
> On Mon, 4 Dec 2006, Karanbir Singh wrote:
>> not every rule, but a _lot_ of them yes. We make it a point to specially 
>> look at all expensive ones ( cpu / network ).
>
> You consider DNS lookups to be expensive? Perhaps that's an indication of 
> an underspecified platform.

Network lookups are very time-expensive due to the need to wait for
responses. Without net lookups enabled it takes well under a second for
SA to process a message (assuming Bayes is SQL-backed which on any big
site it will be). With net lookups enabled that can stretch to ten
seconds or more.

Of course the SA instance is idle for most of those ten seconds, but it
*does* have a ~45Mb memory hit which isn't likely to get swapped out in
that short a time. (If the site is loaded only intermittently, in SA
3.1+ SA will kill most of those instances during non-loaded periods, but
if it's perssistently loaded that won't help either.)

So turning off net tests really *does* greatly affect the number of
instances you can run, and because a single SA instance can only process
one message at once, that means it affects the number of messages you
can process.

I can see ways to fix this (mostly by allowing SA jobs to remember that
they are partway through processing a message and busy waiting for
something external and asking for another, and checking the list of
pending jobs whenever they have a moment to see if any earlier message
can be processed again), but this requires major changes to parts of
the guts of SA, so it's not going to be happening soon unless someone
else does it.

-- 
`The main high-level difference between Emacs and (say) UNIX, Windows,
 or BeOS... is that Emacs boots quicker.' --- PdS
-------------- next part --------------
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug