RE: [exim] Exim 4.52 and SA 3.0.4 memory problem

Página Principal
Apagar esta mensagem
Responder a esta mensagem
Autor: Herb Martin
Data:  
Para: srunschke, exim-users
CC: 
Assunto: RE: [exim] Exim 4.52 and SA 3.0.4 memory problem
> following system is behaving very strange:
>
> RH Enterprise Linux 4
> Exim 4.52
> SA 3.0.4
>
> After upgrading exim from 4.43 to 4.52 today, the memory
> usage from _SpamAssassin_ skyrocketed. And it skyrocketed
> that much and so fast that I had to do a hardware-reboot
> after like 1 hour.
> Mind you, that machine has 2 Gig physical and 4 Gig swap ram.


I am no true expert and my experience may be totally unrelated
to yours (different OS), but if it helps you then consider this:

Strictly FYI...

    Windows Server 2003, 1 GigRam, 4 Gig page file
    Exim 4.52-(and 4.51 earlier), SpamAssassin 3.04
    SA in 3 process with -m 3 (3 children each)


My processes would go up to, and even sometimes exceed, 100 Meg
each, and the machine would slow to a crawl even though CPU
usage was not technically high (presumedly excessive paging.)

This was not always immediate or predictable, and restarting
all SA processess tended to clear the issue (at least for a
while). Using large SARE rules (e.g., either/both the two 1
meg+ URL lists) would exacerbate the problem -- or perhaps the
effect was just the same or similar from a different problem.

I upgraded to the pre-release 3.10pre3 then to 3.10pre4, and
not to nightly snapshots through 7/17 (none are being posted
now) and the problem has largely disappeared except perhaps
when using the very large URL lists which I have disabled for
now. I also just dropped the "-m 3" since supposedly 3.10
handles such children more automatically.

This may seem very agressive but 3.10 is about to release
and those builds were only experiencing bug fixes as the
final release stabilizes and most work has moved to scoring.

Disadvantage MIGHT be due to the scores being less than
optimal, since optimization is the current activity for the
3.10 release (i.e., scoring runs), but I have seen no major
problems so far. My suspicion is that the estimated scores
for 3.10 are approximately as good as 3.04 with calculated
scores -- and I am hoping for additional improvements when
3.10 releases with fully optimized score calculations.

Of course the above is somewhat subjective -- sort of the
oppositive of many reports we see from users who say
things like "My system runs slow" -- well, "My system now
(seems) to run better."

--
Herb Martin