fix: use weakref in EventGenerator to prevent memory leak by Krishsharma179 · Pull Request #3360 · mesa/mesa

Krishsharma179 · 2026-02-21T18:40:13Z

Summary

This PR fixes the memory leak by:

Using weak references instead of hard references (like Event class)
Adding error handling for invalid callables
Implementing silent no-op when references become invalid
Adding comprehensive tests

Bug / Issue

Pr for #3332
EventGenerator.function is a hard reference rather than a weak ref as used in Event. This can lead to memory leaks if a generator is used in combination with, e.g., an Agent. If the agent is removed from the model but remains part of an EventGenerator, it cannot be garbage-collected.

Implementation

As per the coversation in #3320
The behavior of both Event and EventGenerator should be the same
so,

if the function is not callable -> TypeError
if the function is not weakrefable -> TypeError
create weakref, drop local strong ref, then check wr() is None -> ValueError
so we are failing fast during the initialization but if the weakref resolve to None we are just stoping the generator(no-op)
made a __getstate__ and __setstate__ for the pickling issue

Testing

Error cases (non-callable, non-weakly-referenceable)
It is work as expected Event(5, lambda: 5) fail, while a = lambada:10; Event(5, a) still works.
Weak reference behavior and garbage collection
Pickling/unpickling support
No-op behavior when weakref dies during execution

Additional Notes

Matches Event class weak reference pattern for consistency
Maintains backward compatibility
Comprehensive tests for weakrefs, garbage collection, and pickling
Proper error handling for invalid callables

github-actions · 2026-02-21T18:57:36Z

Performance benchmarks:

Model	Size	Init time [95% CI]	Run time [95% CI]
BoltzmannWealth	small	🔵 +1.4% [+0.6%, +2.1%]	🔵 +0.2% [-0.0%, +0.4%]
BoltzmannWealth	large	🔵 -0.3% [-0.9%, +0.3%]	🔵 -2.5% [-4.5%, -0.4%]
Schelling	small	🔵 +0.4% [-0.2%, +0.9%]	🔵 -1.1% [-1.4%, -0.9%]
Schelling	large	🔵 -0.8% [-1.5%, -0.2%]	🔵 -2.0% [-3.6%, -0.4%]
WolfSheep	small	🔵 -0.2% [-0.5%, +0.2%]	🔵 +0.9% [+0.6%, +1.1%]
WolfSheep	large	🔵 -0.8% [-1.6%, -0.0%]	🔵 -0.8% [-2.2%, +0.4%]
BoidFlockers	small	🔵 -3.3% [-3.6%, -3.0%]	🔵 -1.6% [-1.8%, -1.3%]
BoidFlockers	large	🔵 -1.7% [-2.3%, -1.1%]	🔵 -1.2% [-1.5%, -0.8%]

Krishsharma179 · 2026-02-22T06:50:03Z

@quaquel This is the updated PR for the issue #3332 It passes all the checks just required your review please let me know if there is any changes required this PR follows the conversation #3320

quaquel · 2026-02-23T07:47:01Z

Thanks for this cleaning and PR. I just merged #3320, which adds a new function for creating the weakref and do some tests on it. Can you update this PR by merging with upstream main and evaluate whether the new function is useful for this PR?

quaquel · 2026-02-23T07:47:29Z

mesa/time/events.py

+                f"function {function} is not weakly referenceable. "
+                f"Use a user-defined function or method instead."
+            ) from e
+        # Here it will raise a value error if the function returns none


all of this is replaceable by the new function.

quaquel · 2026-02-23T07:48:19Z

mesa/time/events.py

+        if weak_fun() is None:
+            raise ValueError(
+                "Cannot create EventGenerator with a function that is already garbage collected"
+            )


we left this out of #3320, because it is very OS specific whether this adds value. I am still considering adding a check like this, but I need to check its behavior across different operating systems.

quaquel · 2026-02-23T07:49:34Z

mesa/time/events.py

+    @property
+    def function(self) -> Callable:
+        """Get the function. If garbage collected, returns no-op."""
+        func = self._function()
+
+        return func if func is not None else lambda: None


why did you add this property?

This property checks if the weak reference is dead (garbage collected) and then it will return none

I know what the code does, I am asking why you deemed it necessary to put this into a property.

Upon closer look, I realize the @property might be redundant.

We already check the weakref safely in _execute_and_reschedule:

func = self._function() if func is None: self.stop() return

Since the safety check happens at execution time anyway, I'm okay to remove this @Property completely and rely on the check in _execute_and_reschedule.

quaquel · 2026-02-23T07:50:06Z

mesa/time/events.py

+            return  # Silent no-op (no error raised)
+
+        # Execute the function
+        func()


I suggest not calling this variable func. This is a protected term in Python.

ok i will change it

quaquel · 2026-02-23T07:50:48Z

tests/test_lifespan.py

        assert lifetimes.Step.max() == 2


+class TestEventGeneratorMemoryLeak(unittest.TestCase):


I suggest moving these tests into test_events.py, rather than here.

ok I will move it

I meant all tests you added to test_lifespan.py

Krishsharma179 · 2026-02-24T07:29:20Z

Thanks for this cleaning and PR. I just merged #3320, which adds a new function for creating the weakref and do some tests on it. Can you update this PR by merging with upstream main and evaluate whether the new function is useful for this PR?

Hello @quaquel , I was going through the merged #3320 code and in the function

if isinstance(function, types.FunctionType) and function.__name__ == "<lambda>":
        raise ValueError("function must be alive at Event creation.")

this line will raise the value error on both the cases

my_func = lambda x: x + 10
event = Event(my_func)  

# This is unsafe (inline), also blocked:
event = Event(lambda x: x + 10)

Question: Is blocking Case 1 intentional? Since my_func holds a reference, it shouldn't be garbage collected. Should we allow stored lambdas, or is the strict blocking deliberate to avoid edge cases?

On the new function: Yes, function is definitely useful for this PR - it centralizes the weak reference logic and makes the code cleaner. I'll update my implementation to use it.
Thanks!

quaquel · 2026-02-24T07:37:00Z

Assigning a a lambda function to a variable defeats the purpose of a lambda as an anonymous function. Most linters will also flag it as a problem and it's considered an anti pattern. So, I am fine with just blocking lambda's in general.

for more information, see https://pre-commit.ci

Krishsharma179 · 2026-02-25T07:09:44Z

@quaquel i have made the changes please check

mesa/time/events.py

quaquel · 2026-02-25T07:54:58Z

mesa/time/events.py

@@ -270,7 +273,16 @@ def _execute_and_reschedule(self) -> None:
        if not self._active:
            return

-        self.function()
+        # Check weakref HERE (execution time), not in property getter
+        # This matches Event class behavior - weakref check during execution
+        fn = self._function()


Suggested change

# Check weakref HERE (execution time), not in property getter

# This matches Event class behavior - weakref check during execution

fn = self.function()

Krishsharma179 force-pushed the feature-event-generator branch from 1afe307 to b939c4f Compare February 21, 2026 18:53

Krishsharma179 force-pushed the feature-event-generator branch 2 times, most recently from 7e79eb8 to 7ce0b03 Compare February 21, 2026 19:11

quaquel reviewed Feb 23, 2026

View reviewed changes

quaquel added the enhancement Release notes label label Feb 23, 2026

Krishsharma179 force-pushed the feature-event-generator branch from 06c637d to ed51889 Compare February 24, 2026 12:12

Krishsharma179 added 5 commits February 24, 2026 19:15

fix eventgenerator memory leak

43d5f05

Changed tests code

352133e

Fixed typo

7d099f1

Add coverage for EventGenerator __setstate__ None branch

8e758e8

made changes

0ed3c79

Krishsharma179 force-pushed the feature-event-generator branch from a605785 to 0ed3c79 Compare February 24, 2026 13:48

Move EventGenerator tests to event_tests.py

8b1a4b2

Krishsharma179 force-pushed the feature-event-generator branch from 51c46bb to 8b1a4b2 Compare February 25, 2026 06:00

altered test cases

685b1cc

Krishsharma179 force-pushed the feature-event-generator branch from 4fc42d7 to 685b1cc Compare February 25, 2026 06:15

altered test cases

8d5fb89

Krishsharma179 force-pushed the feature-event-generator branch from 4f67f5f to 8d5fb89 Compare February 25, 2026 06:29

[pre-commit.ci] auto fixes from pre-commit.com hooks

44bf8b5

for more information, see https://pre-commit.ci

quaquel reviewed Feb 25, 2026

View reviewed changes

mesa/time/events.py Outdated Show resolved Hide resolved

quaquel reviewed Feb 25, 2026

View reviewed changes

quaquel added 3 commits February 25, 2026 08:56

Apply suggestion from @quaquel

66d7c23

rename

3081f2b

test fixes

ef9889d

quaquel approved these changes Feb 25, 2026

View reviewed changes

quaquel merged commit 5219e82 into mesa:main Feb 25, 2026
13 of 14 checks passed

EwoutH pushed a commit that referenced this pull request Mar 15, 2026

fix: use weakref in EventGenerator to prevent memory leak (#3360)

a017d5d

		assert lifetimes.Step.max() == 2


		class TestEventGeneratorMemoryLeak(unittest.TestCase):

Uh oh!

Conversation

Krishsharma179 commented Feb 21, 2026

Summary

Bug / Issue

Implementation

Testing

Additional Notes

Uh oh!

github-actions bot commented Feb 21, 2026

Uh oh!

Krishsharma179 commented Feb 22, 2026

Uh oh!

quaquel commented Feb 23, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Krishsharma179 commented Feb 24, 2026

Uh oh!

quaquel commented Feb 24, 2026

Uh oh!

Krishsharma179 commented Feb 25, 2026

Uh oh!

Uh oh!

quaquel Feb 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

quaquel Feb 25, 2026 •

edited

Loading