Hacker News Reader

Comment by cocoflunchy | Hacker News Reader

Does anyone have tips on how to speed up test collection in pytest? In our django app (~5000 tests) it takes about 30s to run a single test with `-k testname` when the actual time spent in the test is on the order of a few hundred milliseconds.

It hangs forever on the `collecting ...` phase.

View on HN

Run the single test with "pytest tests/path/to/test_file.py::TestClass::test_method" instead of using pytest -k to parse all your test code looking for a method with the right name.

Alternatively, your IDE or editor probably already has an extension that allows you to right click - run single test. VS Code at least has some plugin that collects the tests in the background.

To add to this, the test ID pytest prints when a test has had a failure/warning is the same form you can call with pytest to run the test. For example, let's say pytest outputs this:

    =============================== warnings summary ===============================
    tests/parser/test_expressions.py::test_complex[+inf+1.23e7j-(inf+12300000j)]

You can run just that particular test (with that particular parametrized input) with

    pytest tests/parser/test_expressions.py::test_complex[+inf+1.23e7j-(inf+12300000j)]

To add to that, pytest's parametrized IDs can be fully controlled in the test code [0], so your test can be something like:

  pytest tests/parser/test_expressions.py::test_complex[special_case_7]

[0] https://docs.pytest.org/en/7.1.x/example/parametrize.html#di...

Also pytest has flags like —lf and —sw to rerun failed tests or tackle test suites with lots of failures.

Thanks, it is a lot better (11s instead of 30)

Times I've profiled large tests suites / codebases in the past, this has been entirely down to import overheads. Quite a difficult thing to fix ...

I'm hoping this PEP Proposal for lazy imports might solve things: https://peps.python.org/pep-0690/

Try profiling pytest with something like pyinstrument[1] and make sure you invoke it from python directly[2] with `--collect-only` as an argument. You'll essentially then just be profiling the import overheads before tests are ran.

[1] - https://pyinstrument.readthedocs.io/en/latest/guide.html#pro...

[2] - https://pyinstrument.readthedocs.io/en/latest/guide.html#pro...

Thanks, pyinstrument is really nice. You're right it looks like imports are taking a lot of time...

With a Django project, what it is actually doing is spinning up a new test database at this point which will be the cause of your slowness. If you haven’t made database schema changes you can run it with the flag “—reuse-db”.

If you’ve got a large project that’s been going for a while, squashing migrations helps too. It can be a bit tricky to do this though.

Thanks, I run it with --nomigrations so it's not the migrations. Reusing db shaves about 2 seconds (I'm down to ~9 seconds when running a single test and specifying the full path)

Reply on news.ycombinator.com