As it turns out, I was having this problem until your post made me
realize that it was solaris (and not my program, as I had assumed).
Some research turn this up:

http://developers.sun.com/solaris/articles/stdio_256.html

which says that 32-bit applications are limited to 256 file
descriptors when running under Solaris, unless you do the following:

export LD_PRELOAD_32=/usr/lib/extendedFILE.so.1

before running your program. Ruby on Solaris is a 32 bit app (we've
never had it compile cleanly as a 64 bit application), so this makes
sense. Except for the part about the limit of 256 file descriptors.
But oh well!

cheers,

b.

fkat... / gmail.com wrote:
> Hi,
>
> I have a ruby / DRb process that is quiting with Errno::EMFILE every
> time it reaches 256 connections. This happens regardless of what I set
> in ulimit -n and only on Solaris.
>
> /opt/csw/lib/ruby/1.8/drb/drb.rb:944:in `accept': Too many open files
> (Errno::EMFILE)
>         from /opt/csw/lib/ruby/1.8/drb/drb.rb:944:in `accept'
>         from /opt/csw/lib/ruby/1.8/drb/drb.rb:1581:in `main_loop'
>         from /opt/csw/lib/ruby/1.8/drb/drb.rb:1430:in `run'
>         from /opt/csw/lib/ruby/1.8/drb/drb.rb:1427:in `start'
>         from /opt/csw/lib/ruby/1.8/drb/drb.rb:1427:in `run'
>         from /opt/csw/lib/ruby/1.8/drb/drb.rb:1347:in `initialize'
>         from /opt/csw/lib/ruby/1.8/drb/drb.rb:1627:in `new'
>         from /opt/csw/lib/ruby/1.8/drb/drb.rb:1627:in `start_service'
>         from ./tuple_space.sh:10
>
>
> I also tried this test both in OSX and Solaris:
>
>  255.times {a<<File.open("/etc/hosts")}
>
> and I got it working on OSX with a ulimit -n 2048 but it seems that
> Ruby on Solaris does not change its behavior based on ulimit -n and it
> still throws:
>
> Errno::EMFILE: Too many open files - /etc/hosts
>         from (irb):17:in `initialize'
>         from (irb):17:in `open'
>         from (irb):17
>         from (irb):17:in `times'
>         from (irb):17
>
>
> Anyone has any insight on how to get Solaris to open more than 256
> files?
>
> Thanks in advance.