Re^2: Performance issues

Thanks.
I tried already to start it nice (e.g. reniced after i noticed) it, it wont help much it is very "consuming" anyway.
As a practice i wanted to convert the bash script and insert the "sleep things" combined with watching CPU load factor, that i cant convert sa-learn i know, but maybe something could be won by making the whole thing a bit more "intelligent" . I forgot to mention that i tried nice earlier.
Sorry if i didnt describe my goals accurately, didnt give all infos.
So i give "breathe beyond nice" a try.

Comment on Re^2: Performance issues

Replies are listed 'Best First'.
Re^3: Performance issues by tirwhan (Abbot) on Nov 28, 2008 at 16:42 UTC
I tried already to start it nice (e.g. reniced after i noticed) it, it wont help much it is very "consuming" anyway. Then you're either a.) on a crap operating system or b.) not CPU-bound. I'd guess b.), especially with sa-learn and a huge bayesian database it's quite likely that you're either running out of RAM (and thus starting to swap) or just using up all the systems IO. Try doing a system call trace on the process (`strace` on Linux), also see what your general resource usage ist (`vmstat`). Stopping and restarting the process is an extremely crude tool and likely to do more harm and good IMO (especially since AFAICR sa-learn actually locks the bayesian database while inserting, which means spamassassin has to wait for the process to restart if you stop and start when the lock is being held). All dogma is stupid.	[reply] [d/l] [select]

Replies are listed 'Best First'.

Re^3: Performance issues
by tirwhan (Abbot) on Nov 28, 2008 at 16:42 UTC

I tried already to start it nice (e.g. reniced after i noticed) it, it wont help much it is very "consuming" anyway.

Then you're either a.) on a crap operating system or b.) not CPU-bound. I'd guess b.), especially with sa-learn and a huge bayesian database it's quite likely that you're either running out of RAM (and thus starting to swap) or just using up all the systems IO. Try doing a system call trace on the process (strace on Linux), also see what your general resource usage ist (vmstat). Stopping and restarting the process is an extremely crude tool and likely to do more harm and good IMO (especially since AFAICR sa-learn actually locks the bayesian database while inserting, which means spamassassin has to wait for the process to restart if you stop and start when the lock is being held).

All dogma is stupid.

[reply]
[d/l]
[select]