Bug #844

Interpreter wide IO deadlock

Added by coderrr . about 3 years ago. Updated 9 months ago.

[ruby-core:20446]
Status:Closed Start date:12/10/2008
Priority:Low Due date:
Assignee:- % Done:

100%
Category:-
Target version:-
ruby -v:

Description

two threads reading from the same socket at same time produces race condition whichs locks interpreter

info at: http://coderrr.wordpress.com/2008/12/10/mri-io-deadlock/

code to repo:

require 'socket'
# if this ever stops printing we're screwed
Thread.new{loop{p 1;sleep 0.5}}

s = Thread.new do
  cli = TCPServer.new(3020).accept
  loop { cli.write 'hi' }
end

sock = TCPSocket.new('localhost', 3020)
2.times do
  Thread.new do
    loop { sock.readpartial 1024 }
  end
end

s.join

Associated revisions

Revision 21165
Added by akr about 3 years ago

* eval.c (rb_thread_schedule): Don't change status of threads which don't run next even if select notify readability/writability. [ruby-core:20446]

History

#1
Updated by Yukihiro Matsumoto about 3 years ago

Hi,

In message "Re: [ruby-core:20907] Re: [Bug #844] Interpreter wide IO deadlock"
    on Sat, 27 Dec 2008 12:51:46 +0900, Tanaka Akira <akr@fsij.org> writes:

|> I investigated the problem.
|
|There is similar problem with write.

|Updated patch:
|
|% svn diff --diff-cmd diff -x '-u -p'
|Index: eval.c

Can you check in?

							matz.

#2
Updated by Akira Tanaka about 3 years ago

  • Status changed from Open to Closed
  • % Done changed from 0 to 100
Applied in changeset r21165.

Also available in: Atom PDF