From: "Iñaki Baz Castillo" <ibc@...>
Date: 2011-01-12T07:44:29+09:00
Subject: Ruby 1.9.2-p0: segmenfault in multithreading when using a custom C extension

Hi, I've coded a SIP protocol parser (very close to the Ragel HTTP
parser) as a Ruby C extension. It works nice (even with very high
traffic)... until now when I've tryed to add multithread logic to
handle the parser SIP message. It seems complex but I've been able to
create a very simple test script that generates the segmentfault.


To summarize, the following code DOES NOT crash:

-----------------------------------------------------------------------------
data = "INVITE sip:alice@example.org SIP/2.0\r\n\r\n"
parser = SIP::MessageParser.new

100000.times do
  msg = SIP::Message.new
  puts "msg.object_id = #{msg.object_id}"
  parser.execute(msg, data.to_str, 0)
  parser.reset

  t = Thread.new { }
end
-----------------------------------------------------------------------------


but the following code DOES crash:

-----------------------------------------------------------------------------
threads = [ ]
data = "INVITE sip:alice@example.org SIP/2.0\r\n\r\n"
parser = SIP::MessageParser.new

100000.times do
  msg = SIP::Message.new
  puts "msg.object_id = #{msg.object_id}"
  parser.execute(msg, data.to_str, 0)
  parser.reset

  threads << Thread.new { }
end
-----------------------------------------------------------------------------


The segmentfault occurs randomly in any iteration of the loop (usually
after 4900 iterations).
As you can see I do nothing within the thread. The only difference
between both cases is the fact that in the second case (the crashing
one) the thread is stored in an array.

And worse: I get segmentfault for MANY reasons (randomly too):

a) In line:  threads << Thread.new { }
-- control frame ----------
c:0008 p:---- s:0032 b:0032 l:000031 d:000031 CFUNC  :initialize
c:0007 p:---- s:0030 b:0030 l:000029 d:000029 CFUNC  :new

b) In line:  parser.execute(msg, data.to_str, 0)

c) When Message#initialize() executes "@timestamp = Time.now":
-- control frame ----------
c:0011 p:---- s:0039 b:0039 l:000038 d:000038 CFUNC  :(null)
c:0010 p:---- s:0037 b:0037 l:000036 d:000036 CFUNC  :now
c:0009 p:0017 s:0034 b:0034 l:000033 d:000033 METHOD /xxxx/message.rb:66

d) In line:   puts "msg.object_id = #{msg.object_id}"



In other server with ruby 1.9.1p376 I've got, just once, a different
error (not a segmentfault in fact):

  in `execute': method `hash' called on terminated
  object (0x00000017c48ac8) (NotImplementedError)



Any help please? I cannot imagine the reason of this issue, neither
I'm sure that it's caused by my C extension.

Thanks a lot.


-- 
Iñaki Baz Castillo
<ibc@aliax.net>