Maybe I sound little extreme, but I keep bayes and some other SA-related db
files on ramdisk.
Is much more faster than reading from harddisk.
~a
On Sun, Mar 14, 2010 at 8:54 PM, Andy Smith <andy@???> wrote:
> Hello,
>
> In the last month I've been looking at improving the performance of
> the spamassassin spamd service.
>
> I added two more hosts to the load balancer and that has helped
> some, but the bottleneck has now moved to the bayes database.
>
> The servers are connecting to a MySQL bayes database so that they're
> all in sync, but the problem comes when the database slows down.
> Queries stack up and spamd doesn't seem to have config for giving up
> on bayes after (say) 10 seconds, so then those connections back up
> until all the servers are full.
>
> Normally I'd look at scaling it by doing multi-master replication,
> but the spamassassin docs specifically say this isn't supported and
> won't work.
>
> Has anyone scaled a spamassassin bayes DB before?
>
> Is my only option to replicate it to slaves with a MySQL proxy in
> front to send writes to the master?
>
> Maybe I could revert to using DB_File databases, but over NFS.
> Except it sounds horrible.
>
> Cheers,
> Andy
>
> --
> http://bitfolk.com/ -- No-nonsense VPS hosting
>
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.9 (GNU/Linux)
>
> iEYEAREDAAYFAkudMNsACgkQIJm2TL8VSQtycACg1LR6J1+ADZsNMHRUIUOcUgLH
> 7m8AoM3b51pYl46CJv9HcF+Wl53OLmFp
> =bege
> -----END PGP SIGNATURE-----
>
> _______________________________________________
> users mailing list
> users@???
> https://lists.bitfolk.com/mailman/listinfo/users
>
>
