Make it possible for faulthandler to terminate the pytest process on timeout (#13679)

* Make it possible for faulthandler to terminate the pytest process on timeout

* Add myself to AUTHORS and add changelog entry.

* Update the documentation

* Skip flakky CI tests

* Revert "Skip flakky CI tests"

This reverts commit be4f0ba.

* Pass the CI env in tox.ini and use it to conditionally skip tests on CI

* Revert "Pass the CI env in tox.ini and use it to conditionally skip tests on CI"

This reverts commit 000876e.

* Revert "Revert "Skip flakky CI tests""

This reverts commit 5066d98.

* Apply suggestions from code review

Co-authored-by: 🇺🇦 Sviatoslav Sydorenko (Святослав Сидоренко) <[email protected]>

* Better way to tell mypy that faulthandler_exit_on_timeout is always a bool instance

* skip test only on CI

* Trigger CI

* Skip the new test on ubuntu-py314

* Actually skip the new test on ubuntu-py314

* Trigger CI

* Trigger CI

---------

Co-authored-by: 🇺🇦 Sviatoslav Sydorenko (Святослав Сидоренко) <[email protected]>

Author: ogrisel

AUTHORS

@@ -342,6 +342,7 @@ Oleg Sushchenko
 Oleksandr Zavertniev
 Olga Matoula
 Oliver Bestwalter
+Olivier Grisel
 Omar Kohl
 Omer Hadari
 Ondřej Súkup

changelog/13678.feature.rst

@@ -0,0 +1,3 @@
+Added a new `faulthandler_exit_on_timeout` ini option set to "false" by default to let `faulthandler` interrupt the `pytest` process after a timeout in case of deadlock -- by :user:`ogrisel`.
+
+Previously, a `faulthandler` timeout would only dump the traceback of all threads to stderr, but would not interrupt the `pytest` process.

doc/en/reference/reference.rst

@@ -1484,6 +1484,25 @@ passed multiple times. The expected format is ``name=value``. For example::
 
    For more information please refer to :ref:`faulthandler`.
 
+
+.. confval:: faulthandler_exit_on_timeout
+
+   Exit the pytest process after the per-test timeout is reached by passing
+   `exit=True` to the :func:`faulthandler.dump_traceback_later` function. This
+   is particularly useful to avoid wasting CI resources for test suites that
+   are prone to putting the main Python interpreter into a deadlock state.
+
+   This option is set to 'false' by default.
+
+   .. code-block:: ini
+
+        # content of pytest.ini
+        [pytest]
+        faulthandler_timeout=5
+        faulthandler_exit_on_timeout=true
+
+   For more information please refer to :ref:`faulthandler`.
+
 .. confval:: filterwarnings
 
 
@@ -2401,6 +2420,9 @@ All the command-line flags can be obtained by running ``pytest --help``::
       faulthandler_timeout (string):
                             Dump the traceback of all threads if a test takes
                             more than TIMEOUT seconds to finish
+      faulthandler_exit_on_timeout (bool):
+                            Exit the test process if a test takes more than
+                            faulthandler_timeout seconds to finish
       addopts (args):       Extra command line options
       minversion (string):  Minimally required pytest version
       pythonpath (paths):   Add paths to sys.path

src/_pytest/faulthandler.py

@@ -16,11 +16,18 @@
 
 
 def pytest_addoption(parser: Parser) -> None:
-    help = (
+    help_timeout = (
         "Dump the traceback of all threads if a test takes "
         "more than TIMEOUT seconds to finish"
     )
-    parser.addini("faulthandler_timeout", help, default=0.0)
+    help_exit_on_timeout = (
+        "Exit the test process if a test takes more than "
+        "faulthandler_timeout seconds to finish"
+    )
+    parser.addini("faulthandler_timeout", help_timeout, default=0.0)
+    parser.addini(
+        "faulthandler_exit_on_timeout", help_exit_on_timeout, type="bool", default=False
+    )
 
 
 def pytest_configure(config: Config) -> None:
@@ -72,14 +79,21 @@ def get_timeout_config_value(config: Config) -> float:
     return float(config.getini("faulthandler_timeout") or 0.0)
 
 
+def get_exit_on_timeout_config_value(config: Config) -> bool:
+    exit_on_timeout = config.getini("faulthandler_exit_on_timeout")
+    assert isinstance(exit_on_timeout, bool)
+    return exit_on_timeout
+
+
 @pytest.hookimpl(wrapper=True, trylast=True)
 def pytest_runtest_protocol(item: Item) -> Generator[None, object, object]:
     timeout = get_timeout_config_value(item.config)
+    exit_on_timeout = get_exit_on_timeout_config_value(item.config)
     if timeout > 0:
         import faulthandler
 
         stderr = item.config.stash[fault_handler_stderr_fd_key]
-        faulthandler.dump_traceback_later(timeout, file=stderr)
+        faulthandler.dump_traceback_later(timeout, file=stderr, exit=exit_on_timeout)
         try:
             return (yield)
         finally:

testing/test_faulthandler.py

@@ -118,6 +118,43 @@ def test_timeout():
     assert result.ret == 0
 
 
+@pytest.mark.keep_ci_var
+@pytest.mark.skipif(
+    "CI" in os.environ and sys.platform == "linux" and sys.version_info >= (3, 14),
+    reason="sometimes crashes on CI because of truncated outputs (#7022)",
+)
+@pytest.mark.parametrize("exit_on_timeout", [True, False])
+def test_timeout_and_exit(pytester: Pytester, exit_on_timeout: bool) -> None:
+    """Test option to force exit pytest process after a certain timeout."""
+    pytester.makepyfile(
+        """
+    import os, time
+    def test_long_sleep_and_raise():
+        time.sleep(1 if "CI" in os.environ else 0.1)
+        raise AssertionError(
+            "This test should have been interrupted before reaching this point."
+        )
+    """
+    )
+    pytester.makeini(
+        f"""
+        [pytest]
+        faulthandler_timeout = 0.01
+        faulthandler_exit_on_timeout = {"true" if exit_on_timeout else "false"}
+        """
+    )
+    result = pytester.runpytest_subprocess()
+    tb_output = "most recent call first"
+    result.stderr.fnmatch_lines([f"*{tb_output}*"])
+    if exit_on_timeout:
+        result.stdout.no_fnmatch_line("*1 failed*")
+        result.stdout.no_fnmatch_line("*AssertionError*")
+    else:
+        result.stdout.fnmatch_lines(["*1 failed*"])
+        result.stdout.fnmatch_lines(["*AssertionError*"])
+    assert result.ret == 1
+
+
 @pytest.mark.parametrize("hook_name", ["pytest_enter_pdb", "pytest_exception_interact"])
 def test_cancel_timeout_on_hook(monkeypatch, hook_name) -> None:
     """Make sure that we are cancelling any scheduled traceback dumping due