semaphore errors on AIX 7.1

[reshmigeorge]

BPO	12372
Nosy	@pitrou, @vstinner
Files	aix_broken_sem.diff

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = <Date 2011-07-21.18:01:45.378>
created_at = <Date 2011-06-20.08:56:08.348>
labels = ['type-bug']
title = 'semaphore errors on AIX 7.1'
updated_at = <Date 2011-07-21.18:01:45.377>
user = 'https://bugs.python.org/reshmigeorge'

bugs.python.org fields:

activity = <Date 2011-07-21.18:01:45.377>
actor = 'neologix'
assignee = 'none'
closed = True
closed_date = <Date 2011-07-21.18:01:45.378>
closer = 'neologix'
components = []
creation = <Date 2011-06-20.08:56:08.348>
creator = 'reshmi.george'
dependencies = []
files = ['22428']
hgrepos = []
issue_num = 12372
keywords = ['patch', 'needs review']
message_count = 11.0
messages = ['138693', '138871', '138873', '138962', '140606', '140653', '140706', '140820', '140821', '140822', '140823']
nosy_count = 6.0
nosy_names = ['pitrou', 'vstinner', 'sable', 'neologix', 'python-dev', 'reshmi.george']
pr_nums = []
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue12372'
versions = ['Python 2.7', 'Python 3.2', 'Python 3.3']

[reshmigeorge]

The same problem that was reported in bpo-9700 is appearing on AIX 7.1.
The following message has been seen when running multi-process python program:
sem_trywait: Permission denied
sem_post: Permission denied
sem_wait: Permission denied
sem_post: Permission denied

It can be easily corrected by defining HAVE_BROKEN_POSIX_SEMAPHORES for AIX 7, like it is done for AIX 6.

[neologix]

Patch attached.
I've used a wildcard so that it won't break when AIX 8 goes out (and it's POSIX).

[vstinner]

It is the fourth issue for the same problem, other issues:

• bpo-1106262 (AIX 5, 2005)
• bpo-1234 (duplicate of bpo-1106262, 2007)
• bpo-9700 (AIX 6)

The initial issue was related to semaphores and fork() (msg60639). Pass pshared=1 to sem_init() gets around this issue, but introduce new issues (msg60640). In general, locks and fork don't work together: there are still such issues in Python, see for example the issue bpo-6721.

It looks like the behaviour of the semaphore depends on the underlying file system: see msg56228.

Nobody is able to get more information, or maybe nobody wants to dig this issue. It would be better to get the real cause of the issue, and maybe report the issue to the editor of AIX. I don't have access to this AIX, and only few Python developers have access to this OS.

At least, I would like to know if Sébastien Sablé (the author of the original patch) changed his opinion about this issue since 2007 ;-)

[neologix]

Nobody is able to get more information, or maybe nobody wants to dig this issue. It would be better to get the real cause of the issue,

Well, it looks like a libc or kernel bug, and we probably can't do
much about it. Just falling back to mutex + condition variable instead
of POSIX semaphores does the trick, so I'll say just avoid POSIX
semaphores altogether on AIX and we should be fine. People could
report this to their AIX vendor, but they should definitely consider
dropping this terminally broken platform and switch to, let's says,
Linux or BSD...

[neologix]

So, what do we do now?

[sable]

Hi,

sorry for the late reply.

> STINNER Victor added the comment:
>

At least, I would like to know if Sébastien Sablé (the author of the original patch) changed his opinion about this issue since 2007 ;-)

I haven't changed my mind concerning this patch and we still use it for
the Python that we deploy on AIX 5.3 and AIX 6.1 systems (I don't use
AIX 7 yet).

Not using this patch will result in performance degradation and
"sem_post: Permission denied" error messages from the system popping on
the console.

> Charles-François Natali added the comment:
>
> Well, it looks like a libc or kernel bug, and we probably can't do
> much about it. Just falling back to mutex + condition variable instead
> of POSIX semaphores does the trick, so I'll say just avoid POSIX
> semaphores altogether on AIX and we should be fine. People could
> report this to their AIX vendor, but they should definitely consider
> dropping this terminally broken platform and switch to, let's says,
> Linux or BSD...

I agree with Charles-François that it is an AIX bug that should be
reported to IBM so that they could fix it in a future release.
I already submitted some "tickets" to IBM concerning other problems and
it is not exactly an "agile" company so you have to be very patient.

I also agree that people should consider dropping this broken platform,
but unfortunately IBM is aggressively and successfully (AFAIK) pushing
this platform to many customers (in the finance industry in particular)
so I don't expect it to die soon.

regards

[pitrou]

The patch looks fine to me.

in the finance industry in particular

Ha, I've seen that. There are even proprietary desktop applications for AIX!

[python-dev]

New changeset f5a7d413638d by Charles-François Natali in branch '2.7':
Issue bpo-12372: POSIX semaphores are broken on AIX: don't use them.
http://hg.python.org/cpython/rev/f5a7d413638d

[python-dev]

New changeset f0475f78d45c by Charles-François Natali in branch '3.2':
Issue bpo-12372: POSIX semaphores are broken on AIX: don't use them.
http://hg.python.org/cpython/rev/f0475f78d45c

[python-dev]

New changeset 44a02d6b74e4 by Charles-François Natali in branch 'default':
Merge - Issue bpo-12372: POSIX semaphores are broken on AIX: don't use them.
http://hg.python.org/cpython/rev/44a02d6b74e4

[neologix]

Patch committed: this should fix POSIX semaphores failures on AIX.
Reshmi, thanks for reporting this!