matt at sergeant.org
Tue Jul 10 17:50:11 UTC 2007
Our current gearman setup doesn't seem to be efficiently using all
We're using it inside qpsmtpd to scan emails (which is slow). Mostly
it works fine, but we wish it used more resources.
Our current architecture is as follows:
4 mail servers running qpsmtpd and scan workers (a simple forking
server for the workers).
Gearmand running on a "monitor" server and another "spare" server,
and also on every mail server. I argued against this setup but
anything else caused too many timeouts.
Workers also run on "monitor" and "spare".
Workers and clients on the mail servers connect to the same
gearmand's: localhost, monitor, spare.
Workers on "monitor" and "spare" connect to gearmand on localhost only.
Yes it's horrible, and I didn't design it. The ops guy just tried
running things a few different places until he got something that
kinda worked - I guess that's the beauty of gearman is that you can
just bring servers into action wherever and it just keeps running.
So suggestions of better architectures would be welcome.
What's the reason for having clients talk to multiple gearmand's? In
my original plan I was going to just have gearmand on "monitor" and
distribute work to all the workers, but that didn't seem to scale
very well (we're scanning about 90 mails/sec).
PS: First Post!!!
More information about the Gearman