[Gllug] Spamassassin (again)

[Adrian McMenamin]

On Mon, 2008-11-17 at 23:20 +0000, Nix wrote:
> On 14 Nov 2008, Adrian McMenamin uttered the following:
> 
> > Thanks for the help with the previous issue. In the end I reinstalled it
> > and it looked like it was running fine - my previous problems were because
> > I was trying to maintain a site wide set of Bayesian tokens etc.
> >
> > Once I realised that it was an unequal struggle to maintain that and just
> > copied the files into various users' folders it worked (or so I thought).
> 
> You could alternatively maintain the Bayes DB in a database, which is
> far, *far* faster as well.
> 
> > Now, though I am noticing it (spamd) eats memory like it is going out of
> > fashion and consequently - it seems when the box recycles its logs etc -
> > will crash (having sent the usuage number up to as high as 40 and then the
> > process bails out with an oom).
> 
> How much memory is that? Numbers, please...
> 
> ... but the load average >40 implies that you have dozens of children
> running. On a box with 1.5Gb RAM that will cause insane swapping: reduce
> the child count drastically, preferably right back down to the default
> of 5, and then psuh it up only when it asks you to.
> 
> How are you starting spamd?
> 

Apologies for not replying earlier - connection problems (not
spamassassin related!).

I added another 3G of swap space and it has run fine since. The box
originally had 500MB of memory and 1.5 G of swap. Memory was upgraded to
1.5G but swap wasn't increased. But it still ran fine.

On the upgrade that didn't seem to be enough. But it seems to run fine
now:

root at dragoneye:~# free
             total       used       free     shared    buffers
cached
Mem:       1553192    1511392      41800          0      12232
158888
-/+ buffers/cache:    1340272     212920
Swap:      4616176    3100056    1516120


-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug