From: Francis Cianfrocca <garbagecat10@...>
Date: 2007-07-07T03:57:00+09:00
Subject: Re: ruby real time web server multi threaded probs

------=_Part_132301_5873422.1183748221285
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
Content-Disposition: inline

On 7/6/07, Roger Pack <rogerpack2005@gmail.com> wrote:
>
> So...I seem to have run into a few problems in a ruby multi-threaded
> socket based program that, after quite awhile, I am still scratching my
> head at these.  Has any body encountered/overcome...[sorry if these are
> old, or not real bugs]
>
> 1) Exceptions that are 'jump their bounds'  Sometimes a 'normal'
> exception, typically caught, will just bypass its begin rescue blocks!
> Brings the whole system to a crash.
>
> 2) Crossed streams.  You open up a socket and start receiving data meant
> for another socket.
>   Related to that sometimes when doing "server.accept" when under duress
> it throws exception like 'not a socket' or 'bad file descriptor' or
> 'argument invalid' when it is a GOOD file descriptor, and is a socket.
> Weirder than weird.
> Also, I've had the following happen:
> r,e,w = select([socket],nil,nil,5) # succeeds
> then r is set to something other than [socket]! [very rare--once or
> twice has that ever happened, but it has!]
> or
> r is set to [socket], then you when you run "socket.recv" it comes back
> with "" -- in other words I think it read from some other socket that
> was closed.
>
> 2) An app just freezes(at least on win32).  Sometimes (too many open
> sockets?) Ruby will just freeze.  There it is.  0% cpu use. ~100 threads
> exist. Ruby is frozen [1.8.6]
> Or, related, a read from a socket 'freezes' all threads in the whole
> program, waiting for a "recv" it come back [and if it doesn't....)
> [pretty rare but happened once].
>
>
> Thus far the 'answer' to these problems has been to...not run a multi
> threaded app.  Mongrel runs several different apps, for example.  Any
> ideas or experiences like this?
> If nobody recognizes these then I will try to recreate them and submit
> them as bugs.  This is nuts.
> Thanks!
> -Roger



For a nonthreaded approach that is fast and scalable, look at EventMachine.

------=_Part_132301_5873422.1183748221285--