From: Norman Makoto Su <normsu@...>
Date: 2002-03-27T11:18:58+09:00
Subject: FastCGI and threading

Hello,

I'm currently developing a web application server with fastcgi.  I want to be
able to handle multiple logins simultaneously.  I noticed that in an approx year
old post [ruby-talk:11922] matz mentions that fastcgi is not thread safe and
that a thread-safe version will be released (by him or someone else).  I'm
wondering, is the current version of fcgi (C or pure ruby version) thread safe?

From the official FastCGI web site, I notice that there are two versions-- a
pure ruby version and a C version.  I've played around with Eli's excellent ruby
version but it seems that it will block all requests until the current request
has been fully handled.  I've written a simple toy server, client:

---server side:
I added this line to my httpd.conf (among other fcgi configuration options)
FastCgiExternalServer /usr/local/apache/fcgi-bin/test.rb -host localhost:9000

ruby script:
z = 0
server = FastCGI::TCP.new('localhost', 9000)
begin
  arr = []
  server.each_request do |request|
    arr[arr.length] = Thread.new {
      puts "Connection received!"
      request.out << "Content-type: text/plain\r\n\r\n"
      request.out << "Z = #{z}"
      sleep 5 if (arr.length % 2) == 0
      z = z + 1
      request.finish
    }
  end
  arr.each { |aThread| aThread.join } # not needed?
ensure
  server.close
end

---and then this on the client:

h = Net::HTTP.new('some web server')
resp, data = h.get('/fcgi-bin/test.rb', nil)
puts data

I expect that if I run two instances of the client at once, the server app would
print out "Connection received!" followed immediately by another "Connection
received!". Instead, it always waits for one of the requests to fully handle
before going on to the next one.  This is exactly what I don't want-- rather, I
want the threads to immediately go off and do their stuff.  Right now, its just
_fully_ executing each request sequentially, one after another.  Also, I
expected both instances of the client to perhaps print out Z=0 (but it is always
Z=0, Z=1).  I've tried various things like moving request.finish:

out = request.out
request.finish
out << "Content-type: text/plain\r\n\r\n"
out << "Z = #{z}"

But that seems to mess things up and the client returns:

/usr/local/lib/ruby/1.6/net/protocol.rb:221:in `error!': 500 "Internal Server
Error" (Net::ProtoFatalError)
        from /usr/local/lib/ruby/1.6/net/http.rb:1217:in `value'
        from /usr/local/lib/ruby/1.6/net/http.rb:605:in `get'
        from test_client2.rb:5

My environment setup is linux Apache 1.3.12 using mod_fastcgi 2.2.12 and ruby
1.6.7 (client-cygwin).  On a side note, I have noticed several multi-threading
versions of fastcgi for python (sz_fcgi.py (a wrapper for a nonthreaded
version), thfcgi.py) on the fastcgi official page.  I'm wondering if an explicit
threading version/wrapper is needed for ruby's fastcgi?  Or perhaps, I'm just
using threads incorrectly?

Any help would be greatly appreciated, fastcgi will just make things a lot
easier for my development.  I've corresponded with Eli regarding this matter,
but, understandably, he seems quite busy and I'd like any other opinions or
suggestions as to how to proceed.

Thanks,

--
Norman Makoto Su