Re: Forking for background processing with rails/unicorn

[Eric Wong <normalperson@yhbt.net>]

Brian Underwood <bunderwood@rbmtechnologies.com> wrote:
> In the Rails 3 application that I support, we sometimes fork during a
> web request in order to do background processing behind the scenes
> without the user (and the browser connection) to wait for it to
> finish.  I realize that a better method would be to have a queue of
> jobs to be processed (I even created a gem called job_boss which works
> well for our case), but it will take us a while to get there, so I'm
> trying to fix this for now.
> 
> The problem that I'm finding is that, when I fork, the HTTP connection
> doesn't close.  IE doesn't handle this as well and the browser is left
> sitting, waiting for the child process to finish rather than being
> redirected.  This functionality worked fine in our application when we
> were using Rails 2.3.5 and Unicorn 1.1.3, but broke when we moved to
> Rails 3 and Unicorn 3.0.0 (we've since moved to Unicorn 3.4.0 and the
> problem remains, I had some hope that Unicorn 3.3.1 might have
> addressed it).

I'm not able to reproduce any difference between 1.x and 3.x in this
regard in forking behavior.  I know that the client socket is never
closed on fork by default, so there's probably something else in your
app that changed...

> I've been able to solve this problem temporarily by closing two file
> descriptors manually (#3 and, depending on the version of Rails #6 or
> #7).  Obviously this is not ideal.  I found the UNICORN_FD environment
> variable when looking through the Unicorn source, but that only seems
> to indicate #3.  Where might #6/#7 be coming from?

Whatever is in UNICORN_FD is the listener socket (probably 3), so 6/7 is
probably the client socket, but you should normally close everything in
the forked child for background processing.

The IO.closefrom method in the io-extra[1] gem will get everything
including descriptors opened without a corresponding Ruby object.
If you know everything uses Ruby IO, then walking ObjectSpace works,
too:

  # close everything except stdin, stdout and stderr
  ObjectSpace.each_object(IO) { |s|
    begin
      s.close if s.fileno > 2
    rescue IOError, Errno::EBADF
      # EBADF can come from duplicate IOs due to FD inheritance
    end
  }

> Is there some way
> that I can determine these without hard-coding the FD #?

lsof or scanning /proc/$$/fd (Linux), but if you have any open socket
connections (memcached/activerecord/net-http-persistent/etc...), you
should close all of those and restart them if necessary in the child.

[1] http://rubyforge.org/projects/shards

-- 
Eric Wong
_______________________________________________
Unicorn mailing list - mongrel-unicorn@rubyforge.org
http://rubyforge.org/mailman/listinfo/mongrel-unicorn
Do not quote signatures (like this one) or top post when replying