Bugs item #1183780, was opened at 2005-04-15 16:27
Message generated for change (Comment added) made by loewis
You can respond by visiting:
https://sourceforge.net/tracker/?func=detail&atid=105470&aid=1183780&group_id=5470
Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: Python Library
Group: Python 2.4
Status: Closed
Resolution: Accepted
Priority: 5
Private: No
Submitted By: Taale Skogan (tskogan)
Assigned to: Neal Norwitz (nnorwitz)
Summary: Popen4 wait() fails sporadically with threads
Initial Comment:
Calling wait() on a popen2.Popen4 object fails
intermittently with the error
Traceback (most recent call last):
...
File "/usr/local/lib/python2.3/popen2.py", line 90, in wait
pid, sts = os.waitpid(self.pid, 0)
OSError: [Errno 10] No child processes
when using threads.
The problem seems to be a race condition when a thread
calls wait() on a popen2.Popen4 object. This also apllies
to Popen3 objects.
The constructor of Popen4. calls _cleanup() which calls
poll() which calls the system call waitpid() for all acitve
child processes. If another thread calls poll() before the
current thread calls wait() on it's child process and the
child process has terminated, the child process is no
longer waitable and the second call to wait() fails.
Code to replicate this behavoir is attached in popen_bug.
py.
Solution: Popen4 and Popen3 should be threadsafe.
Related modules: A seemingly related error occurs with
Popen from the new subprocess module. Use the -s
option in the popen_bug.py script to test this.
Tested on Linux RedHat Enterprise 3 for Python 2.3.3,
Python 2.3.5 and Python 2.4.1 and Solaris for Python 2.
4.1. The error did not occur on a RedHat 7.3 machine
with Python 2.3.5. See the attached file popen_bug.py for
details on the platforms.
----------------------------------------------------------------------
>Comment By: Martin v. Löwis (loewis)
Date: 2007-07-10 23:50
Message:
Logged In: YES
user_id=21627
Originator: NO
If you are seeing a bug in subprocess, please report it separately. This
one has been fixed.
Please don't assume that it is the "same" problem as the one reported
here, unless you have a working patch that proves that it is indeed a
similar problem.
----------------------------------------------------------------------
Comment By: Geoffrey Bache (gjb1002)
Date: 2007-07-10 18:00
Message:
Logged In: YES
user_id=769182
Originator: NO
As an additional note, I have also reproduced the popen problem using
Python 2.4.4, though only once.
It seems to occur more frequently in subprocess.py.
----------------------------------------------------------------------
Comment By: Geoffrey Bache (gjb1002)
Date: 2007-07-10 17:38
Message:
Logged In: YES
user_id=769182
Originator: NO
Did this get fixed in subprocess.py? The patches all seem to be for
popen2.
I have been observing similar problems in subprocess.py, so I downloaded
the test script and ran it with the -s option. It didn't work out of the
box, I had to pass "shell=True" to the subprocess.Popen before it did
anything at all. Which also made me wonder if the subprocess variant of
this problem got forgotten.
Having done that, I have observed failures using Python 2.4.4 on Red Hat
EL3 Linux, and also using Python 2.4.3 on Red Hat EL4 linux. Most of the
time it works, sometimes it hangs forever, and sometimes we get something
that look like this:
Started 20 threads
Exception in thread Thread-19:
Traceback (most recent call last):
File "/usr/lib/python2.4/threading.py", line 442, in __bootstrap
self.run()
File "popen_bug.py", line 53, in run
pipe.wait()
File "/usr/lib/python2.4/subprocess.py", line 1007, in wait
pid, sts = os.waitpid(self.pid, 0)
OSError: [Errno 10] No child processes
P.S. Googling for "[Errno 10] No child processes" suggests others have
this problem. There have been long discussions on the Zope list as to why
some people on Linux get exceptions that look like this, for example.
----------------------------------------------------------------------
Comment By: cheops (atila-cheops)
Date: 2006-04-10 16:55
Message:
Logged In: YES
user_id=1276121
see patch # 1467770 for subprocess.py library module
----------------------------------------------------------------------
Comment By: Martin v. Löwis (loewis)
Date: 2006-03-24 09:16
Message:
Logged In: YES
user_id=21627
Committed as 43286. I also added .cmd to Popen4.
----------------------------------------------------------------------
Comment By: Neal Norwitz (nnorwitz)
Date: 2006-03-24 08:56
Message:
Logged In: YES
user_id=33168
It makes sense to remove from _active on ECHILD.
I wondered the same thing about waitpid(), but left it as it
was. I don't believe it's possible for waitpid to return
any pid other than what you ask for unless the O/S is very,
very broken.
This patch is fine with me, feel free to check it in. BTW,
nice comments and precondition checks in the test.
----------------------------------------------------------------------
Comment By: Martin v. Löwis (loewis)
Date: 2006-03-24 08:40
Message:
Logged In: YES
user_id=21627
This looks all fine.
As a further issue, I think _cleanup should also clean
processes which already have been waited on. So if waitpid
gives ECHILD (in _cleanup), I think the object should get
removed from _active - otherwise, it would stay there
forever. Of course, care is then need to avoid __del__
adding it back to _active.
Putting these all together, I propose v3 of the patch.
Another aspect that puzzles me is the repeated test that
waitpid() really returns the pid you asked for. How could it
not? If it fails, you get an os.error.
----------------------------------------------------------------------
Comment By: Neal Norwitz (nnorwitz)
Date: 2006-03-24 06:17
Message:
Logged In: YES
user_id=33168
I agree with your comment about setting self.sts to 0. That
was the problem I alluded to on python-dev.
Although I dislike __del__, this does seem like an
appropriate place to do the modification of _active.
Note that currently all os.error's are swallowed in poll().
I'm not sure if that was the best idea, but that's the
current interface. wait() does *not* catch any exceptions.
I wasn't really sure what to do about threads. The threads
could always manage their calls into a popen object like you
propose (don't try to handle simultaneous calls to poll and
wait). Another question I had was if popen should be
deprecated in favor of subprocess?
I've attached a patch which I think implements your
suggestion. It also seems to fix all the problems.
----------------------------------------------------------------------
Comment By: Martin v. Löwis (loewis)
Date: 2006-03-24 01:37
Message:
Logged In: YES
user_id=21627
I don't understand why you are setting self.sts to 0 if wait
fails: most likely, there was a simultaneous call to .poll,
which should have set self.sts to the real return value. So
we should return that instead.
I think the whole issue can be avoid if we use resurrection:
If __del__ would put unwaited objects into _active, rather
than __init__, it would not happen that _cleanup polls a pid
which a thread still intends to wait for. In fact, it would
be sufficient to only put the pid into _active (avoiding the
need for resurrection).
If then a thread calls poll explicitly, and another calls
wait, they deserve to lose (with ECHILD). I would claim the
same error exists if part of the application calls
os.wait[3,4](), and another calls .wait later - they also
deserve the exception.
With that approach, I don't think further thread
synchronization would be needed.
----------------------------------------------------------------------
Comment By: Neal Norwitz (nnorwitz)
Date: 2006-03-23 09:41
Message:
Logged In: YES
user_id=33168
The attached patch fixes the problem for me. It also
addresses another issue where wait could be called from
outside the popen2 module. I'm not sure this is the best
solution. I'm not sure there really is a good solution.
Perhaps it's best to allow an exception to be raised?
----------------------------------------------------------------------
You can respond by visiting:
https://sourceforge.net/tracker/?func=detail&atid=105470&aid=1183780&group_id=5470
_______________________________________________
Python-bugs-list mailing list
Unsubscribe:
http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
