Re: FastCGI daemon

[Stefan <stefan@...>]

Jacob Jay wrote:
> On 23 Oct 2009, at 10:58 AM, Stefan wrote:
>
>
> [cc'd to Kepler list]
>
> FastCGI is definitely the way to go, but if you haven't, do check out 
> wsapi-fcgi which is part of Kepler (http://keplerproject.org/). I use 
> this with spawn-fcgi and lighttpd/nginx. An additional FCGI-Lua 
> interface may be unnecessary unless you want it to do something notably 
> different to that of the Kepler WSAPI one.

I was under the impression that wsapi-fcgi did NOT persist the scripts 
between invocations, unless you run each script as it's own fastcgi 
server. Running a couple dozen (at least) extra processes that MUST stay 
up at all times doesn't appeal to sysadmin in me.

My main criteria are:

1) lightweight
2) fast
3) reliable, even under extremely heavy loads
4) controllable and intelligent resource utilization
4) easy administration for critical installations

> For a basic "hello world" test wsapi-fcgi yields results similar to 
> yours (both in the region of 4,000+ requests per second on 2 cores) and 
> the best I've seen thus far is 7000rps on 4 cores (makes all the 
> difference, given that I can't seem to otherwise push CPU use to the max.).

Interesting. What is the memory load of wsapi-fcgi at identical rps 
rates compared to luafcgid?

> Incidentally luafcgid is eating CPU at ~6% when idle (on OS X, anyway).

Yes, the idle loop of the parent process is a placeholder 'usleep(100)' 
right now. Some configurable housekeeping is planned there.