Coverage in parallel failing

[rowleya]

Describe the bug
When we run pytest using multiple processes with coverage on our tests, we get an error message that seems to be related to the integration of SQLite. The error we get is available from our integration test suite here:
http://apollo.cs.man.ac.uk:8080/blue/organizations/jenkins/sPyNNaker8%20Integration%20Tests/detail/neuron_recorder_agdr/10/pipeline/102

Replicated a sample of the output here in case this disappears!

[gw7] linux -- Python 3.6.7 /usr/bin/python3
self = <SqliteDb @0x7f9efba0c1d0 debug=None filename='/var/lib/jenkins/workspace/ation_Tests_neuron_recorder_agdr/.coverage' nest=0 con=None>
sql = 'insert into meta (key, value) values (?, ?)'
parameters = ('has_arcs', '1')

    def execute(self, sql, parameters=()):
        """Same as :meth:`python:sqlite3.Connection.execute`."""
        if self.debug:
            tail = " with {!r}".format(parameters) if parameters else ""
            self.debug.write("Executing {!r}{}".format(sql, tail))
        try:
>           return self.con.execute(sql, parameters)
E           sqlite3.IntegrityError: UNIQUE constraint failed: meta.key

To Reproduce
How can we reproduce the problem? Please be specific.

1. What version of Python are you using?
   3.6
2. What version of coverage.py are you using? The output of coverage debug sys is helpful.
   5.0
3. What versions of what packages do you have installed? The output of pip freeze is helpful.
   See above link for details.
4. What code are you running? Give us a specific commit of a specific repo that we can check out.
   Integration tests in Jenkins; see above for details.
5. What commands did you run?

py.test SpiNNUtils/unittests -rs -n auto --forked --show-progress --cov-branch --cov spynnaker8 --cov spynnaker --cov spinn_front_end_common --cov pacman --cov data_specification --cov spinnman --cov spinn_machine --cov spinn_storage_handlers --cov spalloc --cov spinn_utilities --junitxml junit/SpiNNUtils.xml --cov-report xml:coverage.xml --cov-append --timeout 1200

Expected behavior
A clear and concise description of what you expected to happen.
No error message; protection against writing from mulitple processes through checking for key in database before attempted insert.

Additional context
Add any other context about the problem here.

[kstaniek]

Facing the same issue. Is there any workaround, pls?

[nedbat]

The simplest way to avoid this issue for now is to pin coverage.py to 4.5.4. Another thing that would help me find a solution for you is to provide me with specific, complete instructions that would let me reproduce the problem.

[nedbat]

@rowleya Can you give me specific instructions I can follow to reproduce the problem? What repo, what commit, what commands to run?

[AlexandreYang]

I couldn't reproduce reliably the issue we have on this CI Pipeline https://dev.azure.com/datadoghq/integrations-core/_build/results?buildId=4968&view=logs&jobId=6856723d-645b-5a7b-98a0-8af874afe33f&j=82f604ee-2edd-5639-6463-dbdc7756af1f&t=1d9be0b0-41b5-5871-19d1-c01bf70ecbd1

But here are few things that might help:

• The bug happen when using pytest with both --cov-append --cov=X options. Without --cov-append, it works fine.
• When the error sqlite3.IntegrityError: UNIQUE constraint failed: meta.key is raised (while inserting a new entry with has_arcs key in meta table), there is already an entry in the meta table with ('has_arcs', '0')

[rowleya]

The specific instructions that seem to fail for me are:

python3.6 -m virtualenv test
cd test
source bin/activate
git clone https://github.com/SpiNNakerManchester/SpiNNUtils.git
cd SpiNNUtils && python setup.py develop && cd ..
pip install python-coveralls "coverage>=4.4"
pip install pytest-instafail pytest-xdist pytest-cov pytest-progress pytest-timeout pytest-forked
py.test SpiNNUtils/unittests -rs -n auto --forked --show-progress --cov-branch --cov spynnaker8 --cov spynnaker --cov spinn_front_end_common --cov pacman --cov data_specification --cov spinnman --cov spinn_machine --cov spinn_storage_handlers --cov spalloc --cov spinn_utilities --junitxml junit/SpiNNUtils.xml --cov-report xml:coverage.xml --cov-append --timeout 1200

[rowleya]

It makes sense that --cov-append is the issue; I am trying to achieve something specific with this, which is to run lots of tests from different repositories, including integration tests, and get the overall coverage, so this is essential.

A quick test shows that without the -n auto --forked but with the other arguments (including --cov-append) this also works. It fails some of the tests due to the above not installing all the requirements, but the coverage module doesn't fail; the extra dependencies are:

pip install numpy testfixtures httpretty

[Urth]

I found a reproducible configuration which was plaguing one of our projects. The coverage database/internals will enter a broken state when sub-process coverage is used.
A minimal example can be found at https://github.com/Urth/coveragepy_883

[sco1]

I believe I might be seeing a similar issue in one of my latest builds, though in our case it seems specific to having a test that invokes subprocess.

Here is the specific commit that triggers the CI failure (3.6 log, azure build link, tox config). While this fails in CI for 3.6, 3.7, and 3.8, on my MBP it only fails for 3.6. Removing the test_flake8_actually_runs_checker.py test (the one with subprocess) eliminates the failure.

I will work on capturing more details, but might not be able to get them up until tomorrow.

[nedbat]

This works to reproduce the problem:

git clone https://github.com/python-discord/flake8-annotations
cd flake8-annotations
git checkout 1dfaab8100a686e2171b0e27d0c21cc195be4db6
mktmpenv -p python3.6 -n
pip install tox
rm .coverage; tox -e py36 -- testing/test_flake8_actually_runs_checker.py

[nedbat]

BTW,

While this fails in CI for 3.6, 3.7, and 3.8, on my MBP it only fails for 3.6.

The test fails if there is no .coverage file, which is why CI fails always, but only the first tox env fails on your laptop. With "rm .coverage", it fails on all versions of Python locally.

[nedbat]

One solution is to add this to a .coveragerc file:

[run]
parallel = True

Removing --cov-append also prevents the error. I'm not sure which of these gives you the proper measurements, if either.
I don't know if there's a better solution that can be built in to coverage.py or to pytest-cov.

[sco1]

Thanks Ned, away from a dev machine until this evening to check this out, but I don't think I'm following why the missing coverage file is ultimate source of the problem.

I can't speak for the OP, but for my project if the malfunctioning test is removed then the test suite runs fine, both locally and in CI. For example, the latest release commit (CI) runs without issue.

edit: I've added the suggested .coveragerc file and it seems to resolve the issue, yay!

[xavfernandez]

Is there any drawback to using:

[run]
parallel = True

?

And if there isn't, why is it not the default ? (honest question)

[nedbat]

The drawback to parallel = True is that you need an explicit coverage combine step before you can use coverage report or coverage html (or any other reporting).

[MartinThoma]

Just to highlight: Adding the following section to the setup.cfg fixes the issue:

[coverage:run]
parallel=true

edit: Strange ... I think this worked a while ago. Now it doesn't anymore.

[KingDarBoja]

Having the same issue since the update to 5.0.4. On 5.0.3 is working fine on GitHub Actions 😢

Does the parallel option is required if I am trying to do something like below?

pytest --cov-report= --cov=test_file1.py -vv &
pytest --cov-report= --cov=test_file2.py --cov-append -vv &
pytest --cov-report= --cov=test_file3.py --cov-append -vv &
coverage report -m

[JulienPalard]

TL;DR

The issue happen when running multiple coverage combine in parallel from multiple distincts coverage runs, as there is no way for coverage to tell which .coverage.* files are from which set of coverage run. It happen typically when using pytest-cov in parallel, as pytest-cov always uses a combine step, even when a single .coverage file is needed.

The way to tell coverage which set of .coverage.* file are from which set of runs (with or without tox), is to specify the .coverage filename. For tox, one can use:

setenv =
  COVERAGE_FILE=.coverage.{envname}

I think this issue can be closed, but pytest-cov may have to document this better.

Long version

I'm trying to reproduce it today, I'm able to reproduce it with either tox -p all and:

$ /tmp/3.6/bin/pytest --cov-report term-missing --cov-fail-under=100 --cov src/mediaserver/ &
/tmp/3.7/bin/pytest --cov-report term-missing --cov-fail-under=100 --cov src/mediaserver/ &
/tmp/3.8/bin/pytest --cov-report term-missing --cov-fail-under=100 --cov src/mediaserver/

It looks like it's always an issue near .coverage, as I got those errors:

coverage.misc.CoverageException: Couldn't use data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.40820.348297': no such table: file

or

INTERNALERROR> coverage.misc.CoverageException: Data file '/home/mdk/clones/coverage-reproducer/.coverage' doesn't seem to be a coverage data file: Couldn't use data file '/home/mdk/clones/coverage-reproducer/.coverage': no such table: coverage_schema

Interestingly I got a no such table: file in .coverage.seraph.41964.820593 while not using --parallel-mode. Happen that pytest-cov is adding it.

stacktrace

INTERNALERROR> Traceback (most recent call last):
INTERNALERROR>   File "/tmp/3.6/lib/python3.6/site-packages/coverage/sqldata.py", line 1048, in execute
INTERNALERROR>     return self.con.execute(sql, parameters)
INTERNALERROR> sqlite3.OperationalError: no such table: file
INTERNALERROR> 
INTERNALERROR> During handling of the above exception, another exception occurred:
INTERNALERROR> 
INTERNALERROR> Traceback (most recent call last):
INTERNALERROR>   File "/tmp/3.6/lib/python3.6/site-packages/_pytest/main.py", line 191, in wrap_session
INTERNALERROR>     session.exitstatus = doit(config, session) or 0
INTERNALERROR>   File "/tmp/3.6/lib/python3.6/site-packages/_pytest/main.py", line 247, in _main
INTERNALERROR>     config.hook.pytest_runtestloop(session=session)
INTERNALERROR>   File "/tmp/3.6/lib/python3.6/site-packages/pluggy/hooks.py", line 286, in __call__
INTERNALERROR>     return self._hookexec(self, self.get_hookimpls(), kwargs)
INTERNALERROR>   File "/tmp/3.6/lib/python3.6/site-packages/pluggy/manager.py", line 93, in _hookexec
INTERNALERROR>     return self._inner_hookexec(hook, methods, kwargs)
INTERNALERROR>   File "/tmp/3.6/lib/python3.6/site-packages/pluggy/manager.py", line 87, in 
INTERNALERROR>     firstresult=hook.spec.opts.get("firstresult") if hook.spec else False,
INTERNALERROR>   File "/tmp/3.6/lib/python3.6/site-packages/pluggy/callers.py", line 203, in _multicall
INTERNALERROR>     gen.send(outcome)
INTERNALERROR>   File "/tmp/3.6/lib/python3.6/site-packages/pytest_cov/plugin.py", line 271, in pytest_runtestloop
INTERNALERROR>     self.cov_controller.finish()
INTERNALERROR>   File "/tmp/3.6/lib/python3.6/site-packages/pytest_cov/engine.py", line 46, in ensure_topdir_wrapper
INTERNALERROR>     return meth(self, *args, **kwargs)
INTERNALERROR>   File "/tmp/3.6/lib/python3.6/site-packages/pytest_cov/engine.py", line 231, in finish
INTERNALERROR>     self.cov.stop()
INTERNALERROR>   File "/tmp/3.6/lib/python3.6/site-packages/coverage/control.py", line 687, in combine
INTERNALERROR>     combine_parallel_data(self._data, aliases=aliases, data_paths=data_paths, strict=strict)
INTERNALERROR>   File "/tmp/3.6/lib/python3.6/site-packages/coverage/data.py", line 117, in combine_parallel_data
INTERNALERROR>     data.update(new_data, aliases=aliases)
INTERNALERROR>   File "/tmp/3.6/lib/python3.6/site-packages/coverage/sqldata.py", line 575, in update
INTERNALERROR>     cur = conn.execute('select path from file')
INTERNALERROR>   File "/tmp/3.6/lib/python3.6/site-packages/coverage/sqldata.py", line 1065, in execute
INTERNALERROR>     raise CoverageException("Couldn't use data file {!r}: {}".format(self.filename, msg))
INTERNALERROR> coverage.misc.CoverageException: Couldn't use data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.41964.820593': no such table: file

I see that pytest-cov is instanciating two Coverage, which may lead to the two sqlite file being overwritten.

I'm trying with 3 pytest-cov in parallel, with export COVERAGE_DEBUG=dataio, what I'm getting is surprising:

First process (everything OK)

Erasing data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.46134.640334'
Creating data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.46134.640334'
Combining data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.46134.640334'
Opening data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.46134.640334'
Erasing data file '/home/mdk/clones/coverage-reproducer/.coverage'
Creating data file '/home/mdk/clones/coverage-reproducer/.coverage'
Opening data file '/home/mdk/clones/coverage-reproducer/.coverage'
Deleting combined data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.46134.640334'

2nd process (get mixed up with 3st one)

Erasing data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.46133.288847'
Creating data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.46133.288847'
Opening data file '/home/mdk/clones/coverage-reproducer/.coverage'
Combining data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.46132.615372'
Opening data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.46132.615372'
Opening data file '/home/mdk/clones/coverage-reproducer/.coverage'
Deleting combined data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.46132.615372'
Combining data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.46133.288847'
Opening data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.46133.288847'

3rd process (get mixed up with 2st one)

Erasing data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.46132.615372'
Creating data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.46132.615372'
Opening data file '/home/mdk/clones/coverage-reproducer/.coverage'
Combining data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.46132.615372'
Opening data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.46132.615372'
Opening data file '/home/mdk/clones/coverage-reproducer/.coverage'
Deleting combined data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.46132.615372'
Combining data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.46133.288847'
Opening data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.46133.288847'
Opening data file '/home/mdk/clones/coverage-reproducer/.coverage'
Deleting combined data file '/home/mdk/clones/coverage-reproducer/.coverage.seraph.46133.288847'

my test script

export COVERAGE_DEBUG=dataio
/tmp/3.6/bin/pytest --cov-report term-missing --cov-fail-under=100 --cov src/ > 1.log 2>1.err &
/tmp/3.7/bin/pytest --cov-report term-missing --cov-fail-under=100 --cov src/ > 2.log 2>2.err &
/tmp/3.8/bin/pytest --cov-report term-missing --cov-fail-under=100 --cov src/ > 3.log 2>3.err
wait

Looks like the problem tighten up around pytest-cov using coverage combine, which can't really work when multiple process are spawning files in parallel.

Reproducer https://mdk.fr/x/pytest-cov-reproducer.tar.bz2, which I run using while tox -p all; do :; done and wait (it can take from a few seconds to ... I don't know sorry).

[JulienPalard]

Following up on this one, I successfully use coverage in parallel since enough time to tell it's stable, but without pytest-cov which uses an intermediate combine step (see previous message).

An example using tox to run pytest and coverage for multiple Python versions in parallel and then combining the result can be found here:

https://github.com/JulienPalard/oeis/blob/14bea829d556bd617426c8a85235ba5da1b9f157/tox.ini#L33

I think this issue can be closed.

[nedbat]

I believe this is now fixed in commit b41be3f

[nedbat]

This is now released as part of coverage 6.3.

[bulat15g]

coverage==6.3.1 no errors.
Thank you