Warning
Please read this README carefully and only use this plugin if you understand the consequences. This plugin is designed to catch excessively long test durations like deadlocked or hanging tests, it is not designed for precise timings or performance regressions. Remember your test suite should aim to be fast, with timeouts being a last resort, not an expected failure mode.
This plugin will time each test and terminate it when it takes too long. Termination may or may not be graceful, please see below, but when aborting it will show a stack dump of all thread running at the time. This is useful when running tests under a continuous integration server or simply if you don't know why the test suite hangs.
Note
While by default on POSIX systems pytest will continue to execute the tests after a test has timed out this is not always possible. Often the only sure way to interrupt a hanging test is by terminating the entire process. As this is a hard termination (os._exit()
) it will result in no teardown, JUnit XML output etc. But the plugin will ensure you will have the debugging output on stderr nevertheless, which is the most important part at this stage. See below for detailed information on the timeout methods and their side-effects.
The pytest-timeout plugin has been tested on Python 3.6 and higher, including PyPy3. See tox.ini for currently tested versions.
Install is as simple as e.g.:
pip install pytest-timeout
Now you can run the test suite while setting a timeout in seconds, any individual test which takes longer than the given duration will be terminated:
pytest --timeout=300
Furthermore you can also use a decorator to set the timeout for an individual test. If combined with the --timeout
flag this will override the timeout for this individual test:
@pytest.mark.timeout(60)
def test_foo():
pass
By default the plugin will not time out any tests, you must specify a valid timeout for the plugin to interrupt long-running tests. A timeout is always specified as a number of seconds, and can be defined in a number of ways, from low to high priority:
You can set a global timeout in the pytest configuration file using the
timeout
option. E.g.:[pytest] timeout = 300
- The
PYTEST_TIMEOUT
environment variable sets a global timeout overriding a possible value in the configuration file. - The
--timeout
command line option sets a global timeout overriding both the environment variable and configuration option. Using the
timeout
marker on test items you can specify timeouts on a per-item basis:@pytest.mark.timeout(300) def test_foo(): pass
Setting a timeout to 0 seconds disables the timeout, so if you have a global timeout set you can still disable the timeout by using the mark.
Interrupting tests which hang is not always as simple and can be platform dependent. Furthermore some methods of terminating a test might conflict with the code under test itself. The pytest-timeout plugin tries to pick the most suitable method based on your platform, but occasionally you may need to specify a specific timeout method explicitly.
If a timeout method does not work your safest bet is to use the thread method.
This is the surest and most portable method. It is also the default on systems not supporting the signal method. For each test item the pytest-timeout plugin starts a timer thread which will terminate the whole process after the specified timeout. When a test item finishes this timer thread is cancelled and the test run continues.
The downsides of this method are that there is a relatively large overhead for running each test and that test runs are not completed. This means that other pytest features, like e.g. JUnit XML output or fixture teardown, will not function normally. The second issue might be alleviated by using the --boxed
option of the pytest-xdist plugin.
The benefit of this method is that it will always work. Furthermore it will still provide you debugging information by printing the stacks of all the threads in the application to stderr.
If the system supports the SIGALRM signal the signal method will be used by default. This method schedules an alarm when the test item starts and cancels the alarm when the test finishes. If the alarm expires during the test the signal handler will dump the stack of any other threads running to stderr and use pytest.fail()
to interrupt the test.
The benefit of this method is that the pytest process is not terminated and the test run can complete normally.
The main issue to look out for with this method is that it may interfere with the code under test. If the code under test uses SIGALRM itself things will go wrong and you will have to choose the thread method.
The timeout method can be specified by using the timeout_method
option in the pytest configuration file, the --timeout_method
command line parameter or the timeout
marker. Simply set their value to the string thread
or signal
to override the default method. On a marker this is done using the method
keyword:
@pytest.mark.timeout(method="thread")
def test_foo():
pass
The full signature of the timeout marker is:
pytest.mark.timeout(timeout=0, method=DEFAULT_METHOD)
You can use either positional or keyword arguments for both the timeout and the method. Neither needs to be present.
See the marker api documentation and examples for the various ways markers can be applied to test items.
The plugin will happily terminate timeouts in the finalisers of fixtures. The timeout specified applies to the entire process of setting up fixtures, running the tests and finalising the fixtures. However when a timeout occurs in a fixture finaliser and the test suite continues, i.e. the signal method is used, it must be realised that subsequent fixtures which need to be finalised might not have been executed, which could result in a broken test-suite anyway. In case of doubt the thread method which terminates the entire process might result in clearer output.
The timeout applies to the entire test including any fixtures which may need to be setup or torn down for the test (the exact affected fixtures depends on which scope they are and whether other tests will still use the same fixture). If the timeouts really are too short to include fixture durations, firstly make the timeouts larger ;). If this really isn't an option a timeout_func_only
boolean setting exists which can be set in the pytest ini configuration file, as documented in pytest --help
.
This plugin tries to avoid triggering the timeout when a debugger is detected. This is mostly a convenience so you do not need to remember to disable the timeout when interactively debugging.
The way this plugin detects whether or not a debugging session is active is by checking if a trace function is set and if one is, it check to see if the module it belongs to is present in a set of known debugging frameworks modules OR if pytest itself drops you into a pdb session using --pdb
or similar.
pytest-timeout
provides two hooks that can be used for extending the tool. These hooks are used for setting the timeout timer and cancelling it if the timeout is not reached.
For example, pytest-asyncio
can provide asyncio-specific code that generates better traceback and points on timed out await
instead of the running loop iteration.
See pytest hooks documentation for more info regarding to use custom hooks.
@pytest.hookspec(firstresult=True)
def pytest_timeout_set_timer(item, settings):
"""Called at timeout setup.
'item' is a pytest node to setup timeout for.
'settings' is Settings namedtuple (described below).
Can be overridden by plugins for alternative timeout implementation strategies.
"""
When pytest_timeout_set_timer
is called, settings
argument is passed.
The argument has Settings
namedtuple type with the following fields:
Attribute | Index | Value |
---|---|---|
timeout | 0 | timeout in seconds or None for no timeout |
method | 1 | Method mechanism, 'signal' and 'thread' are supported by default |
func_only | 2 |
|
@pytest.hookspec(firstresult=True)
def pytest_timeout_cancel_timer(item):
"""Called at timeout teardown.
'item' is a pytest node which was used for timeout setup.
Can be overridden by plugins for alternative timeout implementation strategies.
"""
When the timeout occurs, user can open the debugger session. In this case, the timeout should be discarded. A custom hook can check this case by calling is_debugging()
function:
import pytest
import pytest_timeout
def on_timeout():
if pytest_timeout.is_debugging():
return
pytest.fail("+++ Timeout +++")
- Get terminal width from shutil instead of deprecated py, thanks Andrew Svetlov.
- Add an API for extending
pytest-timeout
functionality with third-party plugins, thanks Andrew Svetlov.
- Fix debugger detection on OSX, thanks Alexander Pacha.
- Fix Python 2 removal, thanks Nicusor Picatureanu.
- Increase pytest requirement to >=5.0.0. Thanks Dominic Davis-Foster.
- Use thread timeout method when plugin is not called from main thread to avoid crash.
- Fix pycharm debugger detection so timeouts are not triggered during debugger usage.
- Dropped support for Python 2, minimum pytest version supported is 5.0.0.
- Fix compatibility when run with pytest pre-releases, thanks Bruno Oliveira,
- Fix detection of third-party debuggers, thanks Bruno Oliveira.
- Fix coverage compatibility which was broken by 1.4.0.
- Better detection of when we are debugging, thanks Mattwmaster58.
- Give the threads a name to help debugging, thanks Thomas Grainger.
- Changed location to https://github.com/pytest-dev/pytest-timeout because bitbucket is dropping mercurial support. Thanks Thomas Grainger and Bruno Oliveira.
- Fix support for pytest >= 3.10.
- This changelog was omitted for the 1.3.2 release and was added afterwards. Apologies for the confusion.
- Fix pytest 3.7.3 compatibility. The capture API had changed slightly and this needed fixing. Thanks Bruno Oliveira for the contribution.
- Fix deprecation warning on Python 3.6. Thanks Mickaël Schoentgen
- Create a valid tag for the release. Somehow this didn't happen for 1.3.0, that tag points to a non-existing commit.
- Make it possible to only run the timeout timer on the test function and not the whole fixture setup + test + teardown duration. Thanks Pedro Algarvio for the work!
- Use the new pytest marker API, Thanks Pedro Algarvio for the work!
- Fix for pytest 3.3, thanks Bruno Oliveira.
- Update supported python versions:
- Add CPython 3.6.
- Drop CPyhon 2.6 (as did pytest 3.3)
- Drop CPyhon 3.3
- Drop CPyhon 3.4
- Allow using floats as timeout instead of only integers, thanks Tom Myers.
- Report (default) timeout duration in header, thanks Holger Krekel.
- Bump version to 1.0 to commit to semantic versioning.
- Fix issue #12: Now compatible with pytest 2.8, thanks Holger Krekel.
- No longer test with pexpect on py26 as it is no longer supported
- Require pytest 2.8 and use new hookimpl decorator
- Timeouts will no longer be triggered when inside an interactive pdb session started by
pytest.set_trace()
/pdb.set_trace()
. - Add pypy3 environment to tox.ini.
- Transfer repository to pytest-dev team account.
- Support timeouts happening in (session scoped) finalizers.
- Change command line option --timeout_method into --timeout-method for consistency with pytest
- Added the PYTEST_TIMEOUT environment variable as a way of specifying the timeout (closes issue #2).
- More flexible marker argument parsing: you can now specify the method using a positional argument.
- The plugin is now enabled by default. There is no longer a need to specify
timeout=0
in the configuration file or on the command line simply so that a marker would work.
- Add a marker to modify the timeout delay using a @pytest.timeout(N) syntax, thanks to Laurant Brack for the initial code.
- Allow the timeout marker to select the timeout method using the
method
keyword argument. - Rename the --nosigalrm option to --method=thread to future proof support for eventlet and gevent. Thanks to Ronny Pfannschmidt for the hint.
- Add
timeout
andtimeout_method
items to the configuration file so you can enable and configure the plugin using the ini file. Thanks to Holger Krekel and Ronny Pfannschmidt for the hints. - Tested (and fixed) for python 2.6, 2.7 and 3.2.