Going to production soon, need server advice

I’ve got a RoR app that will be going to production soon (in the next
month or so) and would like to get some feedback on what kind of web
server setup people have had good luck with. I’m looking for something
that is:

  1. Stable
  2. Easy to setup
  3. Secure (sorry but Windows is ruled out here)

My deployment will either be on Fedora Core 5 or possible Mac OS X
Server. Currently I’ve been looking at Litespeed and their LSAPI
integration and am tempted by that. I also have looked at Mongrel with
Pound/Pens/Balance. I could do Apache 2.2 and Mongrel but want to
avoid Apache if possible as I’d like to keep my memory contraints as low
as possible and I’ve found Apache can be a bit of a memory hog (perhaps
its my configuration - but I just don’t care to use Apache if need be).

Thanks for all the feedback.

On 7/7/06, Jacob B. [email protected] wrote:

integration and am tempted by that. I also have looked at Mongrel with
Pound/Pens/Balance. I could do Apache 2.2 and Mongrel but want to
avoid Apache if possible as I’d like to keep my memory contraints as low
as possible and I’ve found Apache can be a bit of a memory hog (perhaps
its my configuration - but I just don’t care to use Apache if need be).

I think most people will agree that avoiding Apache for Rails work is
a good idea unless you can get fcgi running instead of FastCGI.
Regardless, Lighttpd seems to be the way to go nowadays but there are
a lot of people, myself included, who’re still using Apache2 +
FastCGI and doing just fine.

My two bits,
Chuck V.

Regardless, Lighttpd seems to be the way to go nowadays

if you also need php from the same server process, yeah. if youre just
load-balancing to mongrel there are lighter options.

i would explain why i gave up on fcgi but zed has a more concise answer:

08:04 zedas#rubyonrails c, fastcgi has other issues besides install.
it frequently zombies processes, has problems on restarts, leaks memory
unless you have a per request GC call, and starts to get weird with the
spawner/reaper stuff.

Apache 2.2 + mod_proxy balancer + Mongrel cluster - Works like a charm;
easy to scale.

I know lot of people are afraid or sick of using apache; but it is a
solid and a proven product. I dont want to try any of the new web
servers just from a security point of view.

Never user Apache with mod_fastcgi - Had too many issues; not worth
risking yur career !!

-Thila

once you decide on a configuration, be sure to smoke test everything…
A good basic test is to pound the shit out of with a a couple of shell
scripts that spawn a coulple thousand curl or wget process-es.

On Jul 8, 2006, at 12:57 PM, Chuck V. wrote:

My deployment will either be on Fedora Core 5 or possible Mac OS X

I think most people will agree that avoiding Apache for Rails work is
a good idea unless you can get fcgi running instead of FastCGI.

I think you’re saying most people would agree avoiding Apache FCGI
configuration that don’t work is a good idea. :slight_smile:

Apache 2.2.2/mod_rewrite/mod_proxy_balancer/mod_proxy + mongrel
rock.

What’s the advantage of FCGI, other than that Lighty works better with
FCGI than HTTP?


– Tom M.

Im running FC5 / Lighttpd / Rails / PHP / fcgi, no problem, works like a
charm and is quick as hell. Others make a very valid point about
testing… Look up a handy app called ‘seige’, its for seige testing,
will tell you how many concurrent hits it can take before it falls over

  • amongst other handy info!

Cheers

Tim

How about FreeBSD, Apache2 and mod_fcgid?

Isak

On 7/8/06, Jon [email protected] wrote:

once you decide on a configuration, be sure to smoke test everything…
A good basic test is to pound the shit out of with a a couple of shell
scripts that spawn a coulple thousand curl or wget process-es.

I can’t second this enough. Tests have saved my marriage (or they
would have if I were married.) Seriously, the time spent fixing bugs
was overwhelming my time spent coding until I started building tests
religiously. They take time up front which is rarely a happy thing
when it comes to the MBA types, but the difference between a bug fixed
when you’re knee deep in the code and a bug found after release is
unbelieveable.

I haven’t been called into work afterhours once since starting
testing…

Cheers,
Chuck V.