Bug #10076

2nd thread can't get mutex even though 1st thread released it (race)

Added by Dmitry Maksyoma 8 months ago. Updated 8 months ago.

[ruby-core:63872]
Status:Feedback
Priority:Normal
Assignee:-
ruby -v:ruby 2.1.2p95 (2014-05-08 revision 45877) [i686-linux] Backport:2.0.0: UNKNOWN, 2.1: UNKNOWN

Description

require 'thread'

m = Mutex.new

Thread.abort_on_exception = true
Thread.new {
  loop {
    m.synchronize {
      puts 'got mutex in thread'
      sleep 0.1
    }
  }
}

loop {
  m.synchronize {
    puts 'got mutex in loop'
    sleep 1
  }
  # Without sleep, the thread above has no chance of getting the mutex.
  #sleep 0.1
}

monitor-test (344 Bytes) Dmitry Maksyoma, 07/19/2014 01:28 PM

History

#1 Updated by Dmitry Maksyoma 8 months ago

Attaching source code with reproduced problem.

#2 Updated by Nobuyoshi Nakada 8 months ago

  • Description updated (diff)

Thread fairness issue?
Checking interrupts seems to help it, but not sure.

diff --git a/thread.c b/thread.c
index 682e05f..6f67df6 100644
--- a/thread.c
+++ b/thread.c
@@ -4528,11 +4528,15 @@ rb_mutex_synchronize(VALUE mutex, VALUE (*func)(VALUE arg), VALUE arg)
 static VALUE
 rb_mutex_synchronize_m(VALUE self, VALUE args)
 {
+    VALUE ret;
+
     if (!rb_block_given_p()) {
    rb_raise(rb_eThreadError, "must be called with a block");
     }

-    return rb_mutex_synchronize(self, rb_yield, Qundef);
+    ret = rb_mutex_synchronize(self, rb_yield, Qundef);
+    RUBY_VM_CHECK_INTS(GET_THREAD());
+    return ret;
 }

 void rb_mutex_allow_trap(VALUE self, int val)

#3 Updated by Yui NARUSE 8 months ago

  • Status changed from Open to Feedback

#4 Updated by Motohiro KOSAKI 8 months ago

Your code is very artifactual and the patch of comment #2 reduce a performance of normal case. So, until someone provide a real usecase, I don't take it.

Also available in: Atom PDF