From: Bill Kelly <billk@...>
Date: 2004-08-27T11:11:57+09:00
Subject: Re: [BUG?] erroneous deadlock?  (was: apparent hang...)

Hello once again,

> Hopefully have this figured out soon.

Er.  Well I was hoping all this wouldn't turn out to 
be ENTIRELY in my code...  <rueful grin>

It appears the problem is with my wait/timeout code:

def global_wait(timeout_secs)
  begin
    alarm_th = Thread.start {
      sleep timeout_secs
      $global_signal.broadcast
    }
    $global_mutex.synchronize {
############## if alarm_th's broadcast happens
############## before we get to the wait here,
############## then, obviously, we may never wake up
      $global_signal.wait($global_mutex)
    }
  ensure
    alarm_th.kill if alarm_th and alarm_th.alive?
  end
end


As a workaround, I've changed it to:

def global_wait(timeout_secs)
  begin
    awake = false
    alarm_th = Thread.start {
      sleep timeout_secs
      while not awake
        $global_signal.broadcast
        sleep(0.001)   # WAS: Thread.pass
      end
    }
    $global_mutex.synchronize {
      $global_signal.wait($global_mutex)
      awake = true
    }
  ensure
    alarm_th.kill if alarm_th and alarm_th.alive?
  end
end

...So the alarm thread sticks around until it's
*sure* the primary thread has awakened.

I guess it's pretty cheesy, with the sleep(0.001)
and all... But as long as it's foolproof, it will
work for my purposes.  

I'm now wondering if I should add any begin/rescue
in the alarm thread, in case #broadcast could ever
raise an exception....  This code will be in a 
server that I want to run indefinitely, with no
problems... :)


Thanks everyone who followed the thread; sorry if
I've wasted your time . . . 



Regards,

Bill