A PyTest Hot Reloading Plugin

A hot reloading pytest daemon, implemented as a plugin.

Notice - Busted VS Code behavior - 5/28/2024

It was noted that VS Code does not properly detect tests finishing. This is due to buggy experimental behavior that you opt-in to. To opt-out, add to your settings:

"python.experiments.optOutFrom": [
    "pythonTestAdapter"
],

Features

• Uses the jurigged library to watch and hot reload files
• Caches test discovery in many situations
• Improved performance by not having to import libraries again and again and skipping initialization logic
• System for registering workarounds in case something doesn't work out of the box

Trade-offs

• First time imports are slower (measured < 10% to > 100% slower depending on the repo)
• May not work with some libraries
• Sometimes gets in a bad state and needs to be restarted
• Requires starting the daemon separately or adding a command line option to automatically start it

If it takes less than 5 seconds to do all of the imports necessary to run a unit test, then you probably don't need this.

If you're using Django, recommended to use --keep-db to preserve the test database.

The minimum Python version is 3.10

Demo

With the hot reloading daemon

Faster subsequent runs

Without the hot reloading daemon

Installation

Do not install in production code. This is exclusively for the developer environment.

pip: Add pytest-hot-reloading to your dev-requirements.txt file and pip install -r dev-requirements.txt

poetry: poetry add --group=dev pytest-hot-reloading

Usage

Add the plugin to the pytest arguments.

Example using pyproject.toml:

[tool.pytest.ini_options]
addopts = "-p pytest_hot_reloading.plugin"

When running pytest, the plugin will detect whether the daemon is running. If the daemon is not running, it will error unless the --daemon-start-if-needed argument is passed.

The current version of the VS Code Python extension is not, by default, compatible with automatically starting the daemon. The test runner will hang. However, you can revert to legacy behavior which will allow for using the automatic starting. See the VS Code section below for more information.

The recommended way to run the daemon is to give it its own run run profile so you can easily use the debugger in tests. As a convenience you can also, if you're using a dev container, add this to your postStartCommand: pytest --daemon &. If the daemon doesn't start from your postStartCommand, see: microsoft/vscode-remote-release#8536

Note that a pid file is created to track the pid.

Imports and in many cases initialization logic are not reran on subsequent runs, which can be a huge time saver.

Currently, if you want to debug, you will want to run the daemon manually with debugging.

JetBrains (IDEA, PyCharm, etc)

Create a REGULAR Python run configuration, with pytest as the module. For parameters, add --daemon. Strongly consider storing in the project so it is shared with other developers. Note that you most likely also need to set the working directory to the project root where the pytest configuration is located so that it knows to use the plugin you configured earlier.

For more information on parameters, see the VS Code section below.

VS Code

This can easily be done in VS Code with the following launch profile:

        {
            "name": "Pytest Daemon",
            "type": "python",
            "request": "launch",
            "module": "pytest",
            "justMyCode": false,
            "args": [
                "--daemon",
                //
                // everything below this is optional
                //
                "--daemon-port",
                "4852", // the default value
                "--daemon-watch-globs",
                "./*.py" // the default value
                // "./my-project/*.py:./some-thing-else/*.py",  // example of colon separated globs
                "--daemon-ignore-watch-globs",
                "./.venv/*" // this is the default value, also colon separated globs
            ]
        },

The daemon can be configured to use either file system polling or OS-based file system events. The polling behavior is used by default and has higher compatibility. For example, if you're using Docker for Windows with WSL2, you're going to have a bad time with inotify.

If the daemon is already running and you run pytest with --daemon, then the old one will be stopped and a new one will be started. Note that pytest --daemon is NOT how you run tests. It is only used to start the daemon.

The daemon can be stopped with pytest --stop-daemon. This can be used if it gets into a bad state.

To enable automatically starting the server, you have to, currently, disable the new Python Test Adapter:

In your devcontainer.json or user settings:

"python.experiments.optOutFrom": [
    "pythonTestAdapter"
],

Then enable automatically starting the daemon in your settings:

"python.testing.pytestArgs": [
    "--daemon-start-if-needed",
    "tests"
],

Arguments and Env Variables

• PYTEST_DAEMON_USE_OS_EVENTS
  • Instead of polling the file system, use OS events such as inotify to check for file changes (recommended if your system supports it)
  • Default: False
  • Command line: --daemon-use-os-events
• PYTEST_DAEMON_POLL_THROTTLE
  • A multipler for how aggressive the daemon does file system polling. This is not used if OS events are used.
  • 2.0 = twice as slow, less CPU usage
  • Default: 1.0
  • Command line: --daemon-poll-throttle
• PYTEST_DAEMON_PORT
  • The port the daemon listens on.
  • Default: 4852.
  • Command line: --daemon-port
• PYTEST_DAEMON_PYTEST_NAME
  • The name of the pytest executable. Used for spawning the daemon.
  • Default: pytest.
  • Command line: --pytest-name
• PYTEST_DAEMON_WATCH_GLOBS
  • The colon separated globs to watch.
  • Default: ./**/*.py.
  • Command line: --daemon-watch-globs
• PYTEST_DAEMON_IGNORE_WATCH_GLOBS
  • The colon separated globs to ignore.
  • Default: ./.venv/*.
  • Command line: --daemon-ignore-watch-globs
• PYTEST_DAEMON_START_IF_NEEDED
  • Start the pytest daemon if it is not running.
  • Default: False
  • Command line: --daemon-start-if-needed
• PYTEST_DAEMON_DISABLE
  • Disable the pytest plugin for this test run.
  • Default: False
  • Command line: --daemon-disable
• PYTEST_DAEMON_DO_NOT_AUTOWATCH_FIXTURES
  • Disable automatically watching files containing fixtures
  • Default: False
  • Command line: --daemon-do-not-autowatch-fixtures

Workarounds

Libraries that use mutated globals may need a workaround to work with this plugin. The preferred route is to have the library update its code to not mutate globals in a test environment, or to restore them after a test suite has ran. In some cases, that isn't possible, usually because the person with the problem doesn't own the library and can't wait around for a fix.

To register a workaround, create a function that is decorated by the pytest_hot_reloading.workaround.register_workaround decorator. It may optionally yield. If it does, then code after the yield is executed after the test suite has ran.

Example:

from pytest_hot_reloading.workaround import register_workaround

@register_workaround("my_library")
def my_library_workaround():
    import my_library

    yield

    my_library.some_global = BackToOriginalValue()

If you are a library author, you can disable any workarounds for your library by creating an empty module _clear_hot_reload_workarounds.py. If this is successfully imported, then workarounds for the given module will not be executed.

Need More Speed?

• If using docker, run everything out of a named volume. Do not use bind mounts (sharing with the host file system)
• Set PYTEST_DISABLE_PLUGIN_AUTOLOAD=1 in the environment to disable automatic search and loading of plugins
  • Note: You will need to specify explicitly specify the plugins to use if you do this.
  • Example:

        [tool.pytest.ini_options]
        asyncio_mode = "auto"
        addopts = "-p pytest_asyncio.plugin -p megamock.plugins.pytest -p pytest_hot_reloading.plugin"

• Run out of a Github Codespace or similar dedicated external environment
• Prefer using OS events, if your system works well with it. It uses less CPU and can pick up changes faster. Enable it with the environment variable PYTEST_DAEMON_USE_OS_EVENTS=1. It is only disabled by default for maximum compatibility.

Known Issues

• This is alpha, although it's getting closer to where it can be called beta
• If you run out of docker, prefer named volumes. Bind mounds and cached file systems can create random issues where changes aren't reflected.
• The jurigged library is not perfect and sometimes it gets in a bad state
• Some libraries were not written with hot reloading in mind, and will not work without some changes.
• Many systems have conservative inotify limits. Consider bumping this up if you see errors about hitting the inotify limit.
  • Possible command to see current limit: cat /proc/sys/fs/inotify/max_user_instances
  • sudo sysctl fs.inotify.max_user_instances=4096 in the postStartCommand will for example, help with dev containers. Increase as needed. Consult with ChatGPT if you need assistance with your OS.
  • Use the PYTEST_DAEMON_WATCH_GLOBS env variable when there are simply too many files.

Notes

• pytest-xdist will have its logic disabled, even if args are passed in to enable it
• pytest-django will not create test database suffixes for multiworker runs such as tox.
• Extreme example using import perf test from megamock:
  • Without hot reloading: 0.71s
  • With hot reloading (1st run): 1.79s
  • With hot reloading (2nd run): 0.00s