[Catalyst] Using Catalyst with mod_per or FastCGI on heavy traffic web application

[Carl Johnstone]

Another mod_perl user here! I've looked at FastCGI, but ongoing management 
has always looked to be more complecated than just altering the apache 
config.

We have a multi-server setup with hardware load-balancers. They balance 
between two threaded apache servers which serve all static files and then 
using mod_proxy, pass other requests via the load balancer again to the 
three apache/mod_perl/catalyst servers. (We did it with mod_proxy_balancer 
for a while, but found the hardware load-balancer did a better job).

The front apache servers are configured for maximum threads, with Keep-Alive 
on and very high connection timeouts. The mod_perl servers are configured 
with the usual pre-fork and Keep-Alive off. We maximise buffers in the 
front-end to release the mod_perl process as soon as they can to handle 
another request.

We generally restart the mod_perl servers in turn, so although there's a 
slow restart that's invisible to end users. Should there be a problem with 
all the servers the front apache gets a 502 proxy error, so we replace that 
with a pretty page.

One of the advantages of mod_perl is that you only have a single instance of 
perl running, so you only have code in common to your apps loaded once (so 
that covers everything from core and CPAN as well as bespoke common code.)

If we want to run a slightly different version of a codebase we use 
"PerlOptions +Parent" which gives you a separate perl interpreter for a 
particular vhost.

Now I wouldn't say this is a good setup for *everybody* to use, but 
certainly there's a class of users where this type of setup will work very 
well.

Carl