Skip to contentSkip to conversation

Signal timing problems in python testsuite

DISCUSSION Rolf Eike Beer 2 participants 3 messages Jan 23, 2021 – Jan 28, 2021

lore
Share & tools
lore
#0 lore raw

The Python testsuite in all versions of Python3 I remember fails the eintr tests with output like this:

FAIL: test_poll (__main__.SelectEINTRTest)
----------------------------------------------------------------------
Traceback (most recent call last):
File "/usr/lib/python3.9/test/eintrdata/eintr_tester.py", line 451, in
test_poll
self.assertGreaterEqual(dt, self.sleep_time)
AssertionError: 0.19999999995343387 not greater than or equal to 0.2

There are 2 more failing tests with likewise output. This happens on my 4-core C8000. I have attached the testcase, when running it with "python3 test_eintr.py" I also get that results.

I suspect this could be some cross-cpu clock scew, but of course it can be something entirely different. Could someone have a look what is up there? Kernel is currently 5.10.4, but this has been there for a while.

Eike

import os
import signal
import subprocess
import sys
import unittest

from test import support
from test.support import script_helper

@unittest.skipUnless(os.name == "posix", "only supported on Unix") class EINTRTests(unittest.TestCase):

@unittest.skipUnless(hasattr(signal, "setitimer"), "requires setitimer()") def test_all(self):
# Run the tester in a sub-process, to make sure there is only one # thread (for reliable signal delivery).
tester = support.findfile("eintr_tester.py", subdir="eintrdata") # use -u to try to get the full output if the test hangs or crash args = ["-u", tester, "-v"]
if support.verbose:
print()
print("--- run eintr_tester.py ---", flush=True)
# In verbose mode, the child process inherit stdout and stdout, # to see output in realtime and reduce the risk of losing output. args = [sys.executable, "-E", "-X", "faulthandler", *args] proc = subprocess.run(args)
print(f"--- eintr_tester.py completed: "
f"exit code {proc.returncode} ---", flush=True)
if proc.returncode:
self.fail("eintr_tester.py failed")
else:
script_helper.assert_python_ok("-u", tester, "-v")

if __name__ == "__main__":
unittest.main()

#1 lore raw
Quoted 10 lines · Rolf Eike Beer

The Python testsuite in all versions of Python3 I remember fails the eintr tests with output like this:

FAIL: test_poll (__main__.SelectEINTRTest)
----------------------------------------------------------------------
Traceback (most recent call last):
File "/usr/lib/python3.9/test/eintrdata/eintr_tester.py", line 451, in
test_poll
self.assertGreaterEqual(dt, self.sleep_time)
AssertionError: 0.19999999995343387 not greater than or equal to 0.2

There have been some changes in the generic timekeeping functions. I see timing issues on a few of my machines too, e.g. in the xor time measurement functions as reported at lkml.
Sadly I don't have time right now to look deeper into this.

Helge

Quoted 10 lines

There are 2 more failing tests with likewise output. This happens on my 4-core C8000. I have attached the testcase, when running it with "python3 test_eintr.py" I also get that results.

I suspect this could be some cross-cpu clock scew, but of course it can be something entirely different. Could someone have a look what is up there? Kernel is currently 5.10.4, but this has been there for a while.

Eike

#2 lore raw
Quoted 12 lines · Helge Deller
Quoted 10 lines · Rolf Eike Beer

The Python testsuite in all versions of Python3 I remember fails the eintr tests with output like this:

FAIL: test_poll (__main__.SelectEINTRTest)
----------------------------------------------------------------------
Traceback (most recent call last):
File "/usr/lib/python3.9/test/eintrdata/eintr_tester.py", line 451, in
test_poll
self.assertGreaterEqual(dt, self.sleep_time)
AssertionError: 0.19999999995343387 not greater than or equal to 0.2

Quoted 3 lines

There have been some changes in the generic timekeeping functions. I see timing issues on a few of my machines too, e.g. in the xor time measurement functions as reported at lkml.

One more idea:
You could try to enable CONFIG_DEBUG_TIMEKEEPING.

Quoted 13 lines
Quoted 11 lines

There are 2 more failing tests with likewise output. This happens on my 4-core C8000. I have attached the testcase, when running it with "python3 test_eintr.py" I also get that results.

I suspect this could be some cross-cpu clock scew, but of course it can be something entirely different. Could someone have a look what is up there? Kernel is currently 5.10.4, but this has been there for a while.

Eike