From: rdlugosz.1044583@...
Date: 2004-08-31T11:52:16+09:00
Subject: Two issues with webrick - both threading & OS related

All - 

I've recently been playing with a simple webrick servlet that will
allow me to build playlists and stream mp3s off of my home system.  I've ran
into some very peculiar issues that appear to be thread & OS related.  It
seems that on Linux, my program will only handle *one* connection at a time;
i.e. - if your media player is streaming a file, the browser cannot load the
pages to look at other music directories.  However, on WinXP I can have several
simultaneous connections to the system & there are no blocking issues.

The media program aside, working with Jim Weirich today we discovered some
other interesting things.  A absolutely basic webrick servlet that has a Thread.sleep
in it's do_get will block all access to the servlet on any OS - the blocked
request will execute immediately upon the blocker completing.  Here's the
code:
class SleepServer < HTTPServlet::AbstractServlet
  def do_GET(req,
res)
    sleep(30)
    res.body = '<h1>DONE</h1>'
  end

  def do_POST(req,
res)
    do_GET(req, res)
  end
end

This seems pretty strange; it's
as if webrick is operating in a "Single thread model" servlet mode.  I'm sure
that's not the correct behavior.


Finally, another issue we ran into is
one somehow related to threads and IO and is OS specific.  The code below
executes properly on Linux, but appears to deadlock on XP.  
class Streamlet
< WEBrick::HTTPServlet::AbstractServlet
    def do_GET(req, res)
      res["content-type"]
= "text/plain"
      reader, writer = IO.pipe
      Thread.start{
    
   1000.times{|i|
          puts i
          writer << "a" * 1000
    
     writer << "\n"
        }
        writer.close
      }
      res.body
= reader
    end
end

Personally, I think this problem is even more strange,
since Ruby is handling the threads internally it doesn't seem like it should
deadlock on one OS and not the other.

The issue I'm having with my media
player is strange too, since it has problems only on Linux yet our test code
(the sleep(30) servlet) is broken on both platforms...

Sorry for the length
of the email, but I felt the detail was required.  I appreciate any suggestions
you may have.

-Ryan
ryan@dlugosz.net