IQ.Lvbs/pytest-xdist
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Revamp README

Browse Source 
* Use a document title.
* Show a "short and sweet" section at the beginning highlighting the main usage of the plugin.
* Add a table of contents.
* Use a dedicated howto section at the end.
* Move the OVERVIEW section to the main README.
This commit is contained in:
Bruno Oliveira
2021-11-29 12:11:52 -03:00
parent 630c1eb6f2
commit 83bdbf4b95
2 changed files with 210 additions and 180 deletions
Download Patch File Download Diff File Expand all files Collapse all files

76
OVERVIEW.md
View File

@@ -1,76 +0,0 @@
# Overview #
`xdist` works by spawning one or more **workers**, which are controlled
by the **controller**. Each **worker** is responsible for performing
a full test collection and afterwards running tests as dictated by the **controller**.
The execution flow is:
1. **controller** spawns one or more **workers** at the beginning of
the test session. The communication between **controller** and **worker** nodes makes use of
[execnet](https://codespeak.net/execnet/) and its [gateways](https://codespeak.net/execnet/basics.html#gateways-bootstrapping-python-interpreters).
The actual interpreters executing the code for the **workers** might
be remote or local.
1. Each **worker** itself is a mini pytest runner. **workers** at this
point perform a full test collection, sending back the collected
test-ids back to the **controller** which does not
perform any collection itself.
1. The **controller** receives the result of the collection from all nodes.
At this point the **controller** performs some sanity check to ensure that
all **workers** collected the same tests (including order), bailing out otherwise.
If all is well, it converts the list of test-ids into a list of simple
indexes, where each index corresponds to the position of that test in the
original collection list. This works because all nodes have the same
collection list, and saves bandwidth because the **controller** can now tell
one of the workers to just *execute test index 3* index of passing the
full test id.
1. If **dist-mode** is **each**: the **controller** just sends the full list
of test indexes to each node at this moment.
1. If **dist-mode** is **load**: the **controller** takes around 25% of the
tests and sends them one by one to each **worker** in a round robin
fashion. The rest of the tests will be distributed later as **workers**
finish tests (see below).
1. Note that `pytest_xdist_make_scheduler` hook can be used to implement custom tests distribution logic.
1. **workers** re-implement `pytest_runtestloop`: pytest's default implementation
basically loops over all collected items in the `session` object and executes
the `pytest_runtest_protocol` for each test item, but in xdist **workers** sit idly
waiting for **controller** to send tests for execution. As tests are
received by **workers**, `pytest_runtest_protocol` is executed for each test.
Here it worth noting an implementation detail: **workers** always must keep at
least one test item on their queue due to how the `pytest_runtest_protocol(item, nextitem)`
hook is defined: in order to pass the `nextitem` to the hook, the worker must wait for more
instructions from controller before executing that remaining test. If it receives more tests,
then it can safely call `pytest_runtest_protocol` because it knows what the `nextitem` parameter will be.
If it receives a "shutdown" signal, then it can execute the hook passing `nextitem` as `None`.
1. As tests are started and completed at the **workers**, the results are sent
back to the **controller**, which then just forwards the results to
the appropriate pytest hooks: `pytest_runtest_logstart` and
`pytest_runtest_logreport`. This way other plugins (for example `junitxml`)
can work normally. The **controller** (when in dist-mode **load**)
decides to send more tests to a node when a test completes, using
some heuristics such as test durations and how many tests each **worker**
still has to run.
1. When the **controller** has no more pending tests it will
send a "shutdown" signal to all **workers**, which will then run their
remaining tests to completion and shut down. At this point the
**controller** will sit waiting for **workers** to shut down, still
processing events such as `pytest_runtest_logreport`.
## FAQ ##
> Why does each worker do its own collection, as opposed to having
the controller collect once and distribute from that collection to the workers?
If collection was performed by controller then it would have to
serialize collected items to send them through the wire, as workers live in another process.
The problem is that test items are not easily (impossible?) to serialize, as they contain references to
the test functions, fixture managers, config objects, etc. Even if one manages to serialize it,
it seems it would be very hard to get it right and easy to break by any small change in pytest.

314
README.rst
View File

@@ -1,4 +1,6 @@
============
pytest-xdist
============
.. image:: http://img.shields.io/pypi/v/pytest-xdist.svg .. image:: http://img.shields.io/pypi/v/pytest-xdist.svg
:alt: PyPI version :alt: PyPI version
@@ -17,36 +19,17 @@
.. image:: https://img.shields.io/badge/code%20style-black-000000.svg .. image:: https://img.shields.io/badge/code%20style-black-000000.svg
:target: https://github.com/ambv/black :target: https://github.com/ambv/black
xdist: pytest distributed testing plugin The `pytest-xdist`_ plugin extends pytest with new test execution modes, the most used being distributing
======================================== tests across multiple CPUs to speed up test execution::
The `pytest-xdist`_ plugin extends pytest with some unique pytest -n auto
test execution modes:
* test run parallelization_: if you have multiple CPUs or hosts you can use
those for a combined test run. This allows to speed up
development or to use special resources of `remote machines`_.
* ``--looponfail``: run your tests repeatedly in a subprocess. After each run
pytest waits until a file in your project changes and then re-runs
the previously failing tests. This is repeated until all tests pass
after which again a full run is performed.
* `Multi-Platform`_ coverage: you can specify different Python interpreters
or different platforms and run tests in parallel on all of them.
Before running tests remotely, ``pytest`` efficiently "rsyncs" your
program source code to the remote place. All test results
are reported back and displayed to your local terminal.
You may specify different Python versions and interpreters.
If you would like to know how pytest-xdist works under the covers, checkout
`OVERVIEW <https://github.com/pytest-dev/pytest-xdist/blob/master/OVERVIEW.md>`_.
With this call, pytest will spawn a number of workers processes equal to the number of available CPUs, and distribute
the tests randomly across them. There is also a number of `distribution modes`_ to choose from.
**NOTE**: due to how pytest-xdist is implemented, the ``-s/--capture=no`` option does not work. **NOTE**: due to how pytest-xdist is implemented, the ``-s/--capture=no`` option does not work.
.. contents:: **Table of Contents**
Installation Installation
------------ ------------
@@ -61,29 +44,47 @@ To use ``psutil`` for detection of the number of CPUs available, install the ``p
pip install pytest-xdist[psutil] pip install pytest-xdist[psutil]
Features
--------
* Test run parallelization_: tests can be executed across multiple CPUs or hosts.
This allows to speed up development or to use special resources of `remote machines`_.
* ``--looponfail``: run your tests repeatedly in a subprocess. After each run
pytest waits until a file in your project changes and then re-runs
the previously failing tests. This is repeated until all tests pass
after which again a full run is performed.
* `Multi-Platform`_ coverage: you can specify different Python interpreters
or different platforms and run tests in parallel on all of them.
Before running tests remotely, ``pytest`` efficiently "rsyncs" your
program source code to the remote place.
You may specify different Python versions and interpreters. It does not
installs/synchronize dependencies however.
**Note**: this mode exists mostly for backward compatibility, as modern development
relies on continuous integration for multi-platform testing.
.. _parallelization: .. _parallelization:
Speed up test runs by sending tests to multiple CPUs Running tests across multiple CPUs
---------------------------------------------------- ----------------------------------
To send tests to multiple CPUs, use the ``-n`` (or ``--numprocesses``) option:: To send tests to multiple CPUs, use the ``-n`` (or ``--numprocesses``) option::
pytest -n NUMCPUS pytest -n 8
Pass ``-n auto`` to use as many processes as your computer has CPU cores. This Pass ``-n auto`` to use as many processes as your computer has CPU cores. This
can lead to considerable speed ups, especially if your test suite takes a can lead to considerable speed ups, especially if your test suite takes a
noticeable amount of time. noticeable amount of time.
If a test crashes a worker, pytest-xdist will automatically restart that worker The test distribution algorithm is configured with the ``--dist`` command-line option:
and report the test’s failure. You can use the ``--max-worker-restart`` option
to limit the number of worker restarts that are allowed, or disable restarting
altogether using ``--max-worker-restart 0``.
By default, using ``--numprocesses`` will send pending tests to any worker that .. _distribution modes:
is available, without any guaranteed order. You can change the test
distribution algorithm this with the ``--dist`` option. It takes these values:
* ``--dist no``: The default algorithm, distributing one test at a time. * ``--dist load`` **(default)**: Sends pending tests to any worker that is
available, without any guaranteed order.
* ``--dist loadscope``: Tests are grouped by **module** for *test functions* * ``--dist loadscope``: Tests are grouped by **module** for *test functions*
and by **class** for *test methods*. Groups are distributed to available and by **class** for *test methods*. Groups are distributed to available
@@ -96,67 +97,31 @@ distribution algorithm this with the ``--dist`` option. It takes these values:
distributed to available workers as whole units. This guarantees that all distributed to available workers as whole units. This guarantees that all
tests in a file run in the same worker. tests in a file run in the same worker.
* ``--dist loadgroup``: Tests are grouped by xdist_group mark. Groups are * ``--dist loadgroup``: Tests are grouped by the ``xdist_group`` mark. Groups are
distributed to available workers as whole units. This guarantees that all distributed to available workers as whole units. This guarantees that all
tests with same xdist_group name run in the same worker. tests with same ``xdist_group`` name run in the same worker.
Making session-scoped fixtures execute only once .. code-block:: python
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
``pytest-xdist`` is designed so that each worker process will perform its own collection and execute @pytest.mark.xdist_group(name="group1")
a subset of all tests. This means that tests in different processes requesting a high-level def test1():
scoped fixture (for example ``session``) will execute the fixture code more than once, which pass
breaks expectations and might be undesired in certain situations.
While ``pytest-xdist`` does not have a builtin support for ensuring a session-scoped fixture is class TestA:
executed exactly once, this can be achieved by using a lock file for inter-process communication. @pytest.mark.xdist_group("group1")
def test2():
pass
The example below needs to execute the fixture ``session_data`` only once (because it is This will make sure ``test1`` and ``TestA::test2`` will run in the same worker.
resource intensive, or needs to execute only once to define configuration options, etc), so it makes Tests without the ``xdist_group`` mark are distributed normally as in the ``--dist=load`` mode.
use of a `FileLock <https://pypi.org/project/filelock/>`_ to produce the fixture data only once
when the first process requests the fixture, while the other processes will then read
the data from a file.
Here is the code: * ``--dist no``: The normal pytest execution mode, runs one test at a time (no distribution at all).
.. code-block:: python
import json
import pytest
from filelock import FileLock
@pytest.fixture(scope="session")
def session_data(tmp_path_factory, worker_id):
if worker_id == "master":
# not executing in with multiple workers, just produce the data and let
# pytest's fixture caching do its job
return produce_expensive_data()
# get the temp directory shared by all workers
root_tmp_dir = tmp_path_factory.getbasetemp().parent
fn = root_tmp_dir / "data.json"
with FileLock(str(fn) + ".lock"):
if fn.is_file():
data = json.loads(fn.read_text())
else:
data = produce_expensive_data()
fn.write_text(json.dumps(data))
return data
The example above can also be use in cases a fixture needs to execute exactly once per test session, like
initializing a database service and populating initial tables.
This technique might not work for every case, but should be a starting point for many situations
where executing a high-scope fixture exactly once is important.
Running tests in a Python subprocess Running tests in a Python subprocess
------------------------------------ ------------------------------------
To instantiate a python3.9 subprocess and send tests to it, you may type:: To instantiate a ``python3.9`` subprocess and send tests to it, you may type::
pytest -d --tx popen//python=python3.9 pytest -d --tx popen//python=python3.9
@@ -253,8 +218,21 @@ at once. The specifications strings use the `xspec syntax`_.
.. _`execnet`: https://codespeak.net/execnet .. _`execnet`: https://codespeak.net/execnet
When tests crash
----------------
If a test crashes a worker, pytest-xdist will automatically restart that worker
and report the test’s failure. You can use the ``--max-worker-restart`` option
to limit the number of worker restarts that are allowed, or disable restarting
altogether using ``--max-worker-restart 0``.
How-tos
-------
Identifying the worker process during a test Identifying the worker process during a test
-------------------------------------------- ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
*New in version 1.15.* *New in version 1.15.*
@@ -315,7 +293,7 @@ Since version 2.0, the following functions are also available in the ``xdist`` m
Identifying workers from the system environment Identifying workers from the system environment
----------------------------------------------- ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
*New in version 2.4* *New in version 2.4*
@@ -335,7 +313,7 @@ external scripts.
Uniquely identifying the current test run Uniquely identifying the current test run
----------------------------------------- ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
*New in version 1.32.* *New in version 1.32.*
@@ -369,14 +347,14 @@ Additionally, during a test run, the following environment variable is defined:
* ``PYTEST_XDIST_TESTRUNUID``: the unique id of the test run. * ``PYTEST_XDIST_TESTRUNUID``: the unique id of the test run.
Accessing ``sys.argv`` from the controller node in workers Accessing ``sys.argv`` from the controller node in workers
---------------------------------------------------------- ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
To access the ``sys.argv`` passed to the command-line of the controller node, use To access the ``sys.argv`` passed to the command-line of the controller node, use
``request.config.workerinput["mainargv"]``. ``request.config.workerinput["mainargv"]``.
Specifying test exec environments in an ini file Specifying test exec environments in an ini file
------------------------------------------------ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
You can use pytest's ini file configuration to avoid typing common options. You can use pytest's ini file configuration to avoid typing common options.
You can for example make running with three subprocesses your default like this: You can for example make running with three subprocesses your default like this:
@@ -401,7 +379,7 @@ to run tests in each of the environments.
Specifying "rsync" dirs in an ini-file Specifying "rsync" dirs in an ini-file
-------------------------------------- ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
In a ``tox.ini`` or ``setup.cfg`` file in your root project directory In a ``tox.ini`` or ``setup.cfg`` file in your root project directory
you may specify directories to include or to exclude in synchronisation: you may specify directories to include or to exclude in synchronisation:
@@ -419,20 +397,148 @@ where the configuration file was found.
.. _`pytest-xdist repository`: https://github.com/pytest-dev/pytest-xdist .. _`pytest-xdist repository`: https://github.com/pytest-dev/pytest-xdist
.. _`pytest`: http://pytest.org .. _`pytest`: http://pytest.org
Groups tests by xdist_group mark
---------------------------------
*New in version 2.4.* Making session-scoped fixtures execute only once
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Two or more tests belonging to different classes or modules can be executed in same worker through the xdist_group marker: ``pytest-xdist`` is designed so that each worker process will perform its own collection and execute
a subset of all tests. This means that tests in different processes requesting a high-level
scoped fixture (for example ``session``) will execute the fixture code more than once, which
breaks expectations and might be undesired in certain situations.
While ``pytest-xdist`` does not have a builtin support for ensuring a session-scoped fixture is
executed exactly once, this can be achieved by using a lock file for inter-process communication.
The example below needs to execute the fixture ``session_data`` only once (because it is
resource intensive, or needs to execute only once to define configuration options, etc), so it makes
use of a `FileLock <https://pypi.org/project/filelock/>`_ to produce the fixture data only once
when the first process requests the fixture, while the other processes will then read
the data from a file.
Here is the code:
.. code-block:: python .. code-block:: python
@pytest.mark.xdist_group(name="group1") import json
def test1():
pass
class TestA: import pytest
@pytest.mark.xdist_group("group1") from filelock import FileLock
def test2():
pass
@pytest.fixture(scope="session")
def session_data(tmp_path_factory, worker_id):
if worker_id == "master":
# not executing in with multiple workers, just produce the data and let
# pytest's fixture caching do its job
return produce_expensive_data()
# get the temp directory shared by all workers
root_tmp_dir = tmp_path_factory.getbasetemp().parent
fn = root_tmp_dir / "data.json"
with FileLock(str(fn) + ".lock"):
if fn.is_file():
data = json.loads(fn.read_text())
else:
data = produce_expensive_data()
fn.write_text(json.dumps(data))
return data
The example above can also be use in cases a fixture needs to execute exactly once per test session, like
initializing a database service and populating initial tables.
This technique might not work for every case, but should be a starting point for many situations
where executing a high-scope fixture exactly once is important.
How does xdist work?
--------------------
``xdist`` works by spawning one or more **workers**, which are
controlled by the **controller**. Each **worker** is responsible for
performing a full test collection and afterwards running tests as
dictated by the **controller**.
The execution flow is:
1. **controller** spawns one or more **workers** at the beginning of the
test session. The communication between **controller** and **worker**
nodes makes use of `execnet <https://codespeak.net/execnet/>`__ and
its
`gateways <https://codespeak.net/execnet/basics.html#gateways-bootstrapping-python-interpreters>`__.
The actual interpreters executing the code for the **workers** might
be remote or local.
2. Each **worker** itself is a mini pytest runner. **workers** at this
point perform a full test collection, sending back the collected
test-ids back to the **controller** which does not perform any
collection itself.
3. The **controller** receives the result of the collection from all
nodes. At this point the **controller** performs some sanity check to
ensure that all **workers** collected the same tests (including
order), bailing out otherwise. If all is well, it converts the list
of test-ids into a list of simple indexes, where each index
corresponds to the position of that test in the original collection
list. This works because all nodes have the same collection list, and
saves bandwidth because the **controller** can now tell one of the
workers to just *execute test index 3* index of passing the full test
id.
4. If **dist-mode** is **each**: the **controller** just sends the full
list of test indexes to each node at this moment.
5. If **dist-mode** is **load**: the **controller** takes around 25% of
the tests and sends them one by one to each **worker** in a round
robin fashion. The rest of the tests will be distributed later as
**workers** finish tests (see below).
6. Note that ``pytest_xdist_make_scheduler`` hook can be used to
implement custom tests distribution logic.
7. **workers** re-implement ``pytest_runtestloop``: pytest’s default
implementation basically loops over all collected items in the
``session`` object and executes the ``pytest_runtest_protocol`` for
each test item, but in xdist **workers** sit idly waiting for
**controller** to send tests for execution. As tests are received by
**workers**, ``pytest_runtest_protocol`` is executed for each test.
Here it worth noting an implementation detail: **workers** always
must keep at least one test item on their queue due to how the
``pytest_runtest_protocol(item, nextitem)`` hook is defined: in order
to pass the ``nextitem`` to the hook, the worker must wait for more
instructions from controller before executing that remaining test. If
it receives more tests, then it can safely call
``pytest_runtest_protocol`` because it knows what the ``nextitem``
parameter will be. If it receives a “shutdown” signal, then it can
execute the hook passing ``nextitem`` as ``None``.
8. As tests are started and completed at the **workers**, the results
are sent back to the **controller**, which then just forwards the
results to the appropriate pytest hooks: ``pytest_runtest_logstart``
and ``pytest_runtest_logreport``. This way other plugins (for example
``junitxml``) can work normally. The **controller** (when in
dist-mode **load**) decides to send more tests to a node when a test
completes, using some heuristics such as test durations and how many
tests each **worker** still has to run.
9. When the **controller** has no more pending tests it will send a
“shutdown” signal to all **workers**, which will then run their
remaining tests to completion and shut down. At this point the
**controller** will sit waiting for **workers** to shut down, still
processing events such as ``pytest_runtest_logreport``.
FAQ
---
**Question**: Why does each worker do its own collection, as opposed to having the
controller collect once and distribute from that collection to the
workers?
If collection was performed by controller then it would have to
serialize collected items to send them through the wire, as workers live
in another process. The problem is that test items are not easily
(impossible?) to serialize, as they contain references to the test
functions, fixture managers, config objects, etc. Even if one manages to
serialize it, it seems it would be very hard to get it right and easy to
break by any small change in pytest.