Thread deadlock problem

[wimxa]

I made the following test:

#-------------------------------------------------------------------
require 'thread'

mutex = Mutex.new
c_produce = ConditionVariable.new
c_consume = ConditionVariable.new
full = false
t1 = Thread.new do
while true
mutex.synchronize do
c_produce.wait mutex while full
print 'F'
full = true
c_consume.signal
end
#sleep 0.001
end
end
t2 = Thread.new do
while true
mutex.synchronize do
c_consume.wait mutex while !full
print '.'
full = false
c_produce.signal
end
end
end
t1.join
t2.join
#-------------------------------------------------------------------

This should be a classic producer/consumer problem solution. The
output:

F.F.F.F.F.F.F.F.F.F.F.F.F.F.F.F.F.F.F.F.F.F.F.F.F.F.F.F.F.F.F.deadlock
0x2b68300: sleep:- - c:/ruby/projects/threads2.rb:9
deadlock 0x27bc74c: sleep:J(0x2b68300) (main) - c:/ruby/projects/
threads2.rb:27
deadlock 0x2b68274: sleep:- - c:/ruby/projects/threads2.rb:20
c:/ruby/projects/threads2.rb:20: Thread(0x2b68274): deadlock (fatal)

What is wrong with the code?

ruby -v

ruby 1.8.6 (2007-03-13 patchlevel 0) [i386-mswin32]

Also, uncommenting a sleep from above makes the deadlocks disappear
and everything works rather fine - is this just a matter of tweaking
the probability? It ran for quite a long time (several minutes),
that's why I doubt this, but hey - I just might be completely wrong.
Any ideas?

 

[ara.t.howard]

require 'thread'

mutex = Mutex.new
c_produce = ConditionVariable.new
c_consume = ConditionVariable.new
full = false
t1 = Thread.new do
while true
mutex.synchronize do
c_produce.wait mutex while full
print 'F'
full = true
c_consume.signal
end
#sleep 0.001
end
end
t2 = Thread.new do
while true
mutex.synchronize do
c_consume.wait mutex while !full
print '.'
full = false
c_produce.signal
end
end
end
t1.join
t2.join

it seems like you are making if very hard on yourself. why not
something like this:

cfp:~ > cat a.rb
require 'thread'

q = SizedQueue.new 42

producer = Thread.new do
loop do
q.push Time.now.to_f
end
end

consumer = Thread.new do
loop do
p(( data = q.pop ))
end
end

producer.join
consumer.join

??

if you really want to roll your own stuff check out the code in
thread.rb - the Queue and SizedQueue classes are basically the code
you are writing above.


cheers.

a @ http://drawohara.com/

 

[Sylvain Joyeux]

What version of Ruby are you using ? The thread implementation has been
broken in many ways since 1.8.6, so it can be the problem

(I have a patched ruby 1.8.6 here with some fixes and it works fine)

 

[wimxa]

@ara

it seems like you are making if very hard on yourself.

No, this is just a test. I have another problem requiring the
synchronization. It is not working, so I made this test to see what is
I am doing wrong. It seems that it is not a problem with the other
code, but the part of the code that is similar to what I have made in
the above test. However, thanks for the pointers - I will take a look
at them and I might just find an answer.

@Sylvain

What version of Ruby are you using ?

I have posted the output of

ruby -v

which is:

ruby 1.8.6 (2007-03-13 patchlevel 0) [i386-mswin32]

Do you need some more info (and how can I obtain it)?

Also, is there some good documentation about multi-threading in ruby,
especially the synchronization part? I have looked at some tutorials
(e.g. Programming Ruby: The Pragmatic Programmer's Guide), but they
are sparse on this topic (just a few examples, not even near to what I
need). Any others?

 

[MenTaLguY]

ruby 1.8.6 (2007-03-13 patchlevel 0) [i386-mswin32]

Also, uncommenting a sleep from above makes the deadlocks disappear
and everything works rather fine - is this just a matter of tweaking
the probability? It ran for quite a long time (several minutes),
that's why I doubt this, but hey

I think you at least need a newer patchlevel of 1.8.6; IIRC, there are
bugs in the mutex implementation that shipped with 1.8.6p0.

-mental

 

[wimxa]

ruby 1.8.6 (2007-03-13 patchlevel 0) [i386-mswin32]

Also, uncommenting a sleep from above makes the deadlocks disappear
and everything works rather fine - is this just a matter of tweaking
the probability? It ran for quite a long time (several minutes),
that's why I doubt this, but hey

I think you at least need a newer patchlevel of 1.8.6; IIRC, there are
bugs in the mutex implementation that shipped with 1.8.6p0.

-mental

Thanks, that worked. I just downloaded patch 36 and it worked fine.
*sigh*