[monarch] Introduce pytest marker to isolate tests in subprocess #2076
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
NB: this is mostly vibe-coded since I'm not really familiar with pytest's internals.
This diff adds a new pytest marker `pytest.mark.spawn_isolate(timeout=..., env=...)` that provides similar functionality as the `pytest-isolate` package, except that it launches subprocesses by spawning them rather than forking (since the latter is incompatible with torch). Output from the child process is continuously piped back to the parent process so that if the child exits unexpectedly, we can still observe what happened. Child output is marked clearly, surrounded by orange headers/footers that look like:
```
vvvvvvv spawn_isolate FAIL for {test name} (child output) vvvvvvv
{child output}
^^^^^ end spawn_isolate FAIL for {test name} (child output) ^^^^^
```
The contents of `{child output}` are colored tan where it otherwise would have been white to easily distinguish where the child output starts and stops, but it maintains any explicit coloration set by pytest.
There are some tests where this marker is necessary -- for example, `test_debugger.py` requires each test to be isolated because each test needs to independently reconfigure/make assertions about the global debug controller actor.
However, I would further argue that this marker should be used for most or all of our tests. For better or worse, monarch makes substantial use of global state, and it is too easy for one test to interfere with another test in the same process. For example, a test failure could put a global actor into a bad/failed state in a way that is consistent with the intended behavior of the system; however, other tests in the same process that try to use this global actor would then fail, which just adds noise to an already difficult debugging experience.
Moreover, running each test in its own subprocess will make it much easier to ensure that resources aren't leaked across tests.
Differential Revision: [D88395512](https://our.internmc.facebook.com/intern/diff/D88395512/)
[ghstack-poisoned]
samlurye
added a commit
that referenced
this pull request
Dec 5, 2025
NB: this is mostly vibe-coded since I'm not really familiar with pytest's internals.
This diff adds a new pytest marker `pytest.mark.spawn_isolate(timeout=..., env=...)` that provides similar functionality as the `pytest-isolate` package, except that it launches subprocesses by spawning them rather than forking (since the latter is incompatible with torch). Output from the child process is continuously piped back to the parent process so that if the child exits unexpectedly, we can still observe what happened. Child output is marked clearly, surrounded by orange headers/footers that look like:
```
vvvvvvv spawn_isolate FAIL for {test name} (child output) vvvvvvv
{child output}
^^^^^ end spawn_isolate FAIL for {test name} (child output) ^^^^^
```
The contents of `{child output}` are colored tan where it otherwise would have been white to easily distinguish where the child output starts and stops, but it maintains any explicit coloration set by pytest.
There are some tests where this marker is necessary -- for example, `test_debugger.py` requires each test to be isolated because each test needs to independently reconfigure/make assertions about the global debug controller actor.
However, I would further argue that this marker should be used for most or all of our tests. For better or worse, monarch makes substantial use of global state, and it is too easy for one test to interfere with another test in the same process. For example, a test failure could put a global actor into a bad/failed state in a way that is consistent with the intended behavior of the system; however, other tests in the same process that try to use this global actor would then fail, which just adds noise to an already difficult debugging experience.
Moreover, running each test in its own subprocess will make it much easier to ensure that resources aren't leaked across tests.
Differential Revision: [D88395512](https://our.internmc.facebook.com/intern/diff/D88395512/)
ghstack-source-id: 327474510
Pull Request resolved: #2076
meta-cla
bot
added
the
CLA Signed
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Stack from ghstack (oldest at bottom):
NB: this is mostly vibe-coded since I'm not really familiar with pytest's internals.
This diff adds a new pytest marker
pytest.mark.spawn_isolate(timeout=..., env=...)that provides similar functionality as thepytest-isolatepackage, except that it launches subprocesses by spawning them rather than forking (since the latter is incompatible with torch). Output from the child process is continuously piped back to the parent process so that if the child exits unexpectedly, we can still observe what happened. Child output is marked clearly, surrounded by orange headers/footers that look like:The contents of
{child output}are colored tan where it otherwise would have been white to easily distinguish where the child output starts and stops, but it maintains any explicit coloration set by pytest.There are some tests where this marker is necessary -- for example,
test_debugger.pyrequires each test to be isolated because each test needs to independently reconfigure/make assertions about the global debug controller actor.However, I would further argue that this marker should be used for most or all of our tests. For better or worse, monarch makes substantial use of global state, and it is too easy for one test to interfere with another test in the same process. For example, a test failure could put a global actor into a bad/failed state in a way that is consistent with the intended behavior of the system; however, other tests in the same process that try to use this global actor would then fail, which just adds noise to an already difficult debugging experience.
Moreover, running each test in its own subprocess will make it much easier to ensure that resources aren't leaked across tests.
Differential Revision: D88395512