Debian bigmem buildbot hanging in test_bigmem

[nadeemvawda]

BPO	14107
Nosy	@loewis, @pitrou, @vstinner
Files	mem_watchdog_nonblock.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 2012-02-27.18:01:25.243>
created_at = <Date 2012-02-24.15:17:25.851>
labels = ['type-bug', 'tests']
title = 'Debian bigmem buildbot hanging in test_bigmem'
updated_at = <Date 2012-02-27.18:01:25.243>
user = 'https://bugs.python.org/nadeemvawda'

bugs.python.org fields:

activity = <Date 2012-02-27.18:01:25.243>
actor = 'neologix'
assignee = 'none'
closed = True
closed_date = <Date 2012-02-27.18:01:25.243>
closer = 'neologix'
components = ['Tests']
creation = <Date 2012-02-24.15:17:25.851>
creator = 'nadeem.vawda'
dependencies = []
files = ['24645']
hgrepos = []
issue_num = 14107
keywords = ['patch', 'buildbot']
message_count = 13.0
messages = ['154139', '154166', '154168', '154169', '154170', '154172', '154310', '154330', '154335', '154345', '154353', '154369', '154439']
nosy_count = 6.0
nosy_names = ['loewis', 'pitrou', 'vstinner', 'nadeem.vawda', 'neologix', 'python-dev']
pr_nums = []
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue14107'
versions = ['Python 3.3']

[nadeemvawda]

On the debian bigmem buildbot, test_bigmem hangs until it gets killed
by a timeout:

http://www.python.org/dev/buildbot/all/builders/AMD64%20debian%20bigmem%203.x/builds/134/steps/test/logs/stdio
http://www.python.org/dev/buildbot/all/builders/AMD64%20debian%20bigmem%203.x/builds/131/steps/test/logs/stdio
http://www.python.org/dev/buildbot/all/builders/AMD64%20debian%20bigmem%203.x/builds/129/steps/test/logs/stdio
http://www.python.org/dev/buildbot/all/builders/AMD64%20debian%20bigmem%203.x/builds/128/steps/test/logs/stdio
http://www.python.org/dev/buildbot/all/builders/AMD64%20debian%20bigmem%203.x/builds/123/steps/test/logs/stdio
http://www.python.org/dev/buildbot/all/builders/AMD64%20debian%20bigmem%203.x/builds/85/steps/test/logs/stdio

This has happened on every run I've checked that hasn't crashed before
reaching test_bigmem for some other reason (e.g. the recently-fixed zlib
bus error), as far back as the buildbot history goes.

bpo-5438 might be related.

[python-dev]

New changeset 4afcb25988c4 by Victor Stinner in branch 'default':
Issue bpo-14107: fix bigmem tests on str.capitalize(), str.swapcase() and
http://hg.python.org/cpython/rev/4afcb25988c4

[vstinner]

I ran test_bigmem on my computer. The test pass but it is really slow.
----

$ ./python -m test -v -M 8G test_bigmem
(...
Ran 164 tests in 3228.227s

OK (skipped=44)
1 test OK.
----

One hour, whereas 44 tests are skipped. Skipped tests are the tests using the most quantity of memory, so I expect that they are even slower.

Do you need these tests using so much memory (and are really slow)?
---
test_capitalize (test.test_bigmem.StrTest) ... skipped 'not enough memory: 24.0G minimum needed'
test_encode_utf32 (test.test_bigmem.StrTest) ... skipped 'not enough memory: 9.0G minimum needed'
test_format (test.test_bigmem.StrTest) ... skipped 'not enough memory: 12.0G minimum needed'
test_repr_large (test.test_bigmem.StrTest) ... skipped 'not enough memory: 10.0G minimum needed'
test_split_large (test.test_bigmem.StrTest) ... skipped 'not enough memory: 20.0G minimum needed'
test_swapcase (test.test_bigmem.StrTest) ... skipped 'not enough memory: 24.0G minimum needed'
test_title (test.test_bigmem.StrTest) ... skipped 'not enough memory: 24.0G minimum needed'
test_translate (test.test_bigmem.StrTest) ... skipped 'not enough memory: 12.0G minimum needed'
test_split_large (test.test_bigmem.BytesTest) ... skipped 'not enough memory: 20.0G minimum needed'
test_compare (test.test_bigmem.TupleTest) ... skipped 'not enough memory: 32.0G minimum needed'
test_concat_large (test.test_bigmem.TupleTest) ... skipped 'not enough memory: 48.0G minimum needed'
test_concat_small (test.test_bigmem.TupleTest) ... skipped 'not enough memory: 24.0G minimum needed'
test_contains (test.test_bigmem.TupleTest) ... skipped 'not enough memory: 16.0G minimum needed'
test_from_2G_generator (test.test_bigmem.TupleTest) ... skipped 'not enough memory: 9.0G minimum needed'
test_from_almost_2G_generator (test.test_bigmem.TupleTest) ... skipped 'not enough memory: 9.0G minimum needed'
test_hash (test.test_bigmem.TupleTest) ... skipped 'not enough memory: 16.0G minimum needed'
test_index_and_slice (test.test_bigmem.TupleTest) ... skipped 'not enough memory: 16.0G minimum needed'
test_repeat_large (test.test_bigmem.TupleTest) ... skipped 'not enough memory: 48.0G minimum needed'
test_repeat_large_2 (test.test_bigmem.TupleTest) ... skipped 'not enough memory: 12.0G minimum needed'
test_repeat_small (test.test_bigmem.TupleTest) ... skipped 'not enough memory: 24.0G minimum needed'
test_repr_large (test.test_bigmem.TupleTest) ... skipped 'not enough memory: 22.0G minimum needed'
test_append (test.test_bigmem.ListTest) ... skipped 'not enough memory: 18.0G minimum needed'
test_compare (test.test_bigmem.ListTest) ... skipped 'not enough memory: 32.0G minimum needed'
test_concat_large (test.test_bigmem.ListTest) ... skipped 'not enough memory: 48.0G minimum needed'
test_concat_small (test.test_bigmem.ListTest) ... skipped 'not enough memory: 24.0G minimum needed'
test_contains (test.test_bigmem.ListTest) ... skipped 'not enough memory: 16.0G minimum needed'
test_count (test.test_bigmem.ListTest) ... skipped 'not enough memory: 16.0G minimum needed'
test_extend_large (test.test_bigmem.ListTest) ... skipped 'not enough memory: 32.0G minimum needed'
test_extend_small (test.test_bigmem.ListTest) ... skipped 'not enough memory: 16.0G minimum needed'
test_hash (test.test_bigmem.ListTest) ... skipped 'not enough memory: 16.0G minimum needed'
test_index (test.test_bigmem.ListTest) ... skipped 'not enough memory: 16.0G minimum needed'
test_index_and_slice (test.test_bigmem.ListTest) ... skipped 'not enough memory: 16.0G minimum needed'
test_inplace_concat_large (test.test_bigmem.ListTest) ... skipped 'not enough memory: 48.0G minimum needed'
test_inplace_concat_small (test.test_bigmem.ListTest) ... skipped 'not enough memory: 24.0G minimum needed'
test_inplace_repeat_large (test.test_bigmem.ListTest) ... skipped 'not enough memory: 32.0G minimum needed'
test_inplace_repeat_small (test.test_bigmem.ListTest) ... skipped 'not enough memory: 16.0G minimum needed'
test_insert (test.test_bigmem.ListTest) ... skipped 'not enough memory: 18.0G minimum needed'
test_pop (test.test_bigmem.ListTest) ... skipped 'not enough memory: 16.0G minimum needed'
test_remove (test.test_bigmem.ListTest) ... skipped 'not enough memory: 16.0G minimum needed'
test_repeat_large (test.test_bigmem.ListTest) ... skipped 'not enough memory: 48.0G minimum needed'
test_repeat_small (test.test_bigmem.ListTest) ... skipped 'not enough memory: 24.0G minimum needed'
test_repr_large (test.test_bigmem.ListTest) ... skipped 'not enough memory: 22.0G minimum needed'
test_reverse (test.test_bigmem.ListTest) ... skipped 'not enough memory: 16.0G minimum needed'
test_sort (test.test_bigmem.ListTest) ... skipped 'not enough memory: 16.0G minimum needed'
---

[nadeemvawda]

I ran test_bigmem on my computer. The test pass but it is really slow.

Is this with or without the fix you just pushed? How much RAM does your
system have? I tried running with "-M 4.5G" on my 8GB machine earlier
today, and it ate through all of my RAM plus 4GB of swap before the OOM
killer terminated it.

Do you need these tests using so much memory (and are really slow)?

In general, I think it's important to exercise cases where 32-bit
truncation issues and the like might arise, because otherwise they'll lie
around undetected until they bite some unsuspecting user at an
inopportune moment. But I don't know much about those specific tests, so
I can't really offer much of an opinion here.

[Still, 48GB minimum? ೦_೦ ]

[vstinner]

> I ran test_bigmem on my computer. The test pass but it is really slow.

Is this with or without the fix you just pushed? How much RAM does your
system have? I tried running with "-M 4.5G" on my 8GB machine earlier
today, and it ate through all of my RAM plus 4GB of swap before the OOM
killer terminated it.

My PC has 12 GB of RAM and no swap. I ran the test after my commit.

I tried to enable swap but writing to the swap does completly crash my
kernel. My / uses btrfs and btrfs doesn't look to support swap files,
even with a loopback device.

[nadeemvawda]

My PC has 12 GB of RAM and no swap. I ran the test after my commit.

That explains. I ran it earlier (obviously), so it included a bunch of
those more-demanding tests, which must be where the OOM killer hit it.

[nadeemvawda]

The buildbot still isn't happy:

http://www.python.org/dev/buildbot/all/builders/AMD64%20debian%20bigmem%203.x/builds/144/steps/test/logs/stdio

Also, I ran the test on my own machine and noticed two tests that are
still exceeding their nominal peak memory usage:

• test_encode_utf7: expected 4.8G, actual 7.2G

• test_unicode_repr: expected 4.8G, actual 5.6G

[neologix]

"""
Thread 0x00002ba588709700:
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/test/support.py", line 1168 in consumer
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/threading.py", line 682 in run
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/threading.py", line 729 in _bootstrap_inner
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/threading.py", line 702 in _bootstrap

Current thread 0x00002ba57b2d4260:
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/test/support.py", line 1198 in stop
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/test/support.py", line 1240 in wrapper
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/unittest/case.py", line 385 in _executeTestPart
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/unittest/case.py", line 440 in run
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/unittest/case.py", line 492 in __call__
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/unittest/suite.py", line 105 in run
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/unittest/suite.py", line 67 in __call__
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/unittest/suite.py", line 105 in run
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/unittest/suite.py", line 67 in __call__
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/unittest/runner.py", line 168 in run
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/test/support.py", line 1369 in _run_suite
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/test/support.py", line 1403 in run_unittest
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/test/test_bigmem.py", line 1252 in test_main
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/test/regrtest.py", line 1221 in runtest_inner
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/test/regrtest.py", line 907 in runtest
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/test/regrtest.py", line 710 in main
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/test/main.py", line 13 in <module>
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/runpy.py", line 73 in _run_code
File "/var/tmpfs/martin.vonloewis/3.x.loewis-parallel2/build/Lib/runpy.py", line 160 in _run_module_as_main
"""

There's a problem with the _file_watchdog thread:
if the pipe gets full (because the consumer thread doesn't get to run
often enough), the watchdog thread will block on the write() to the
pipe.
Then, the main thread tries to stop the watchdog:

"""
static void
cancel_file_watchdog(void)
{
    /* Notify cancellation */
    PyThread_release_lock(watchdog.cancel_event);

    /* Wait for thread to join */
    PyThread_acquire_lock(watchdog.running, 1);
    PyThread_release_lock(watchdog.running);

    /* The main thread should always hold the cancel_event lock */
    PyThread_acquire_lock(watchdog.cancel_event, 1);
}
"""

The cancel_event lock is released, but the watchdog thread is stuck
on the write().
The only thing that could wake it up is a read() from the consumer
thread, but the main thread - the one calling cancel_file_watchdog()

• blocks when acquiring the running lock: since the GIL is not
  released, the consumer thread can't run, so it doesn't drain the pipe,
  and game over...

"""
/* We can't do anything if the consumer is too slow, just bail out */
if (write(watchdog.wfd, (void *) &x, sizeof(x)) < sizeof(x))
break;
if (write(watchdog.wfd, data, data_len) < data_len)
break;
"""

AFAICT, this can't happen, because the write end of the pipe is not in
non-blocking mode (which would solve this issue).

Otherwise, there are two things I don't understand:

1. IIUC, the goal of the watchdog thread is to collect memory
   consumption in a timely manner: that's now the case, but since the
   information is printed in a standard thread, it doesn't bring any improvement (because it can be delayed for arbitrarily long), or am I
   missing something?
2. instead of using a thread and the faulthandler infrastructure to run
   GIL-less, why not simply use a subprocess? It could then simply
   parse /proc/<PID>/statm at a regular interval, and print stats to
   stdout. It would also solve point 1.

[neologix]

1. IIUC, the goal of the watchdog thread is to collect memory
   consumption in a timely manner: that's now the case, but since the
   information is printed in a standard thread, it doesn't bring any improvement (because it can be delayed for arbitrarily long), or am I
   missing something?

Scratch that. It does bring an improvement for a post facto analysis
(I thought at first it was used in a pseudo-realtime way).

[pitrou]

"""
/* We can't do anything if the consumer is too slow, just bail out */
if (write(watchdog.wfd, (void *) &x, sizeof(x)) < sizeof(x))
break;
if (write(watchdog.wfd, data, data_len) < data_len)
break;
"""

AFAICT, this can't happen, because the write end of the pipe is not in
non-blocking mode (which would solve this issue).

I think my original plan was to put it in non-blocking mode, but I must
have forgotten in the end.

2. instead of using a thread and the faulthandler infrastructure to run
   GIL-less, why not simply use a subprocess? It could then simply
   parse /proc/<PID>/statm at a regular interval, and print stats to
   stdout. It would also solve point 1.

I can't think of any drawback off the top of my head, so that sounds
reasonable.

[neologix]

I think my original plan was to put it in non-blocking mode, but I
must have forgotten in the end.

Here's a patch.

I can't think of any drawback off the top of my head, so that sounds
reasonable.

I'll try to write up a patch (I have only 1.2GB RAM, so it won't be easy
to test :-).

[python-dev]

New changeset 52dc4fcd0d6f by Charles-François Natali in branch 'default':
Issue bpo-14107: test: Fix a deadlock involving the memory watchdog thread.
http://hg.python.org/cpython/rev/52dc4fcd0d6f

[neologix]

test_bigmem now passes on the buildbot:
http://python.org/dev/buildbot/all/builders/AMD64%20debian%20bigmem%203.x/builds/149/steps/test/logs/stdio