BackgroundTask binding : Release GIL when destroying BackgroundTask #5579
Conversation
This file contains 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
|
Looks good. A minor stylistic thing: I'm not sure I like having both functions named gilManagingPtr just with different signatures - this code is technical enough that anyone looking at it needs to keep track of exactly what's happening, and while both "lock the gil before deleting this" and "acquire the gil before deleting this" are related to the gil, they are different enough in function that I think different names would be clearer - but these names aren't publicly visible, so it's not particularly important. |
I've taken a stab at improving this in 3469422 - let me know if you think any other names would be better. |
|
Looks good to me to squash and merge. |
The BackgroundTask constructor calls `cancelAndWait()`, and a Python function can't be cancelled unless it is able to aquire the GIL. Thus if we don't release the GIL before calling `wait()`, we can deadlock. We were already releasing the GIL appropriately for tasks returned by `ParallelAlgo::callOnBackgroundThread()` but not for those constructed with `BackgroundTask()` directly. Fixing this revealed another problem that was dealt with correctly by `callOnBackgroundThread()` but not by the `BackgroundTask()` constructor. We need to hold the GIL when destroying the Python function owned by the task, and we were doing this by manually deleting it after it ran, at a point when we knew we had the GIL. But this wasn't sufficient if the BackgroundTask was cancelled before the function even started. In this case, the function would be destroyed from C++ and we would be deleting Python objects without holding the GIL. The solution is again to adopt the GIL management used by `callOnBackgroundThread()`.
johnhaddon
force-pushed
the
backgroundTaskDeadlockFix
branch
from
3469422
to
ca37126
Compare
johnhaddon
added a commit
to johnhaddon/gaffer
that referenced
this pull request
Dec 8, 2023
This was made necessary by changes in GafferHQ#5579 which meant that the functor wasn't destroyed immediately after running. Using the WeakMethod is better though, because the old method wouldn't have broken the circular reference if the background task was cancelled before it was started.
johnhaddon
added a commit
to johnhaddon/gaffer
that referenced
this pull request
Dec 12, 2023
This was made necessary by changes in GafferHQ#5579 which meant that the functor wasn't destroyed immediately after running. Using the WeakMethod is better though, because the old method wouldn't have broken the circular reference if the background task was cancelled before it was started.
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.
The BackgroundTask constructor calls
cancelAndWait(), and a Python function can't be cancelled unless it is able to aquire the GIL. Thus if we don't release the GIL before callingwait(), we can deadlock. We were already releasing the GIL appropriately for tasks returned byParallelAlgo::callOnBackgroundThread()but not for those constructed withBackgroundTask()directly.Fixing this revealed another problem that was dealt with correctly by
callOnBackgroundThread()but not by theBackgroundTask()constructor. We need to hold the GIL when destroying the Python function owned by the task, and we were doing this by manually deleting it after it ran, at a point when we knew we had the GIL. But this wasn't sufficient if the BackgroundTask was cancelled before the function even started. In this case, the function would be destroyed from C++ and we would be deleting Python objects without holding the GIL. The solution is again to adopt the GIL management used bycallOnBackgroundThread().I'm just making this PR for
mainas the problem only showed up in the LocalDispatcher work I'm doing for 1.4. All current usage of BackgroundTasks goes throughcallOnBackgroundThread(), so shouldn't suffer from this issue.