Best way to run a Plack PSGI Perl application outside of Apache using FastCGI?

Question

I'm currently running Apache/mod_perl with a PSGI-application invoked by Plack::Handler::Apache2. The problem we have is each Apache process consumes a connection to the PostgreSQL database, which is expensive. To solve this, we plan to run the PSGI-application separate from Apache, and let all Apache processes communicate with it over a UNIX domain socket. What setup would you recommend in my case?

My plan is to run it using plackup:

plackup -s FCGI -E production --nproc 100 --daemonize --listen /tmp/myapp.sock \
/usr/local/bin/myapp.psgi

I asked the author of Plack, Tatsuhiko Miyagawa, if plackup -s FCGI is recommended for production purposes. He provided the following answer:

"while it could be used for production, i usually recommend using other specialized servers such as Starman, Starlet or uwsgi. FCGI is an exception because we don't have a specific FCGI daemon other than the default Plack::Handler. We have a plan to split out FCGI out of Plack core and make that a separate installation."

Now, until FCGI is splitted out of Plack, the question is therefore, what is the best possible way to run a PSGI application using FastCGI outside of Apache?

Answer 1

If you insist on using the FastCGI protocol, you can make Apache connect to an external FastCGI server using mod_fastcgi and the FastCgiExternalServer directive (note that mod_fcgid does not support this mode of operation). This should work with the FCGI handler built into Plack.

But as Tatsuhiko said, it's recommended to use a server like Starman or Starlet that works as reverse proxy over HTTP. So you would use mod_proxy on the Apache side. With Apache 2.4.7, it's also possible to communicate over Unix domain sockets.