[Gllug] Spamassassin (again)

[Nix]

On 19 Nov 2008, Adrian McMenamin outgrape:

> On Wed, 2008-11-19 at 22:29 +0000, Nix wrote:
>
>> 
>> Holy crap you're using a lot of swap. You might want to run the
>> wonderful ps_mem script on that box and see what's using so much
>> space up (it's *much* more accurate than ps or top).
>
> So does this indicate I have a memory leak?

Maybe.

> root at dragoneye:~# free
>              total       used       free     shared    buffers
> cached
> Mem:       1553192    1446384     106808          0      24108
> 377308
> -/+ buffers/cache:    1044968     508224
> Swap:      4616176    3904348     711828
>
>
> root at dragoneye:~# ./ps_mem.py
>  Private  +   Shared  =  RAM used	Program 
>  ....
>
> 810.8 MiB +  30.8 MiB = 841.6 MiB	spamd (3)
> ---------------------------------
>                         970.0 MiB
> =================================
>
> So what has happened to the other approx 2GB of memory that is being
> used but is not in buffers/cache?

free(1) throws too much information away for us to know. cat /proc/meminfo
and perhaps slabtop might give you more clues.


However, ps_mem shows *RAM* used, not RAM+swap. It's a physical-memory-
consumption reporting tool. The spamd(3) output is instructive, though:
finally I know how many children you've got, and 841Mb is *far far*
larger than I'd expect from a mere three children: here, I see:

hades:/usr/lib/gcj# ps -o rss,vsz -C spamd
  RSS    VSZ
 3392  47900
53048 181580
 2548  55112

 51.9 MiB +   1.3 MiB =  53.2 MiB       spamd (3)

Are you running some enormous additional rulesets? (There are a couple of
long-unmaintained ones which can gigantically bloat spamd past anything
sane, quite like you're seeing.)

-- 
`Not even vi uses vi key bindings for its command line.' --- PdS
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug