Search Issue Tracker
Won't Fix
Votes
1
Found in
2019.4
2020.3
2021.2
2021.2.8f1
2022.1
2022.2
Issue ID
1408974
Regression
No
Memory leak when instantiating and destroying GameObjects with a collider
Reproduction steps:
1. Open the user's attached project "ColliderMemLeak.zip"
2. Open "SampleScene"
3. Enter Play mode
4. Click "Start" in the Game view
5. Observe memory usage in the Profiler window
Expected result: Memory is released
Actual result: Memory is never released, the Editor becomes slow and unresponsive after a while
Reproducible with: 2019.4.36f1, 2020.3.31f1, 2021.2.15f1, 2022.1.0b11, 2022.2.0a7
Note:
1. Tested with Mesh and Box colliders
2. "Total Used Memory" under the Memory module decreases slightly after clicking "Stop" but "Physics Used Memory" under the Physics module does not
Add comment
All about bugs
View bugs we have successfully reproduced, and vote for the bugs you want to see fixed most urgently.
Latest issues
- [Android] Stage information is not logged when Log Shader Compilation is enabled
- [Vulkan] The memory allocation increases rapidly when there are multiple (three or more) Real-Time Reflection Probes in the Scene
- [macOS] Library folder of the opened project can be deleted which leads to the crash
- “Default Scene” dropdown field contains a spelling mistake “Default Builtin”
- Editor crashes on PPtr<Mesh> after adding Text Mesh and Cloth Components to the same GameObject
Resolution Note:
The issue here is that element IDs for PhysX objects are never flushed and dynamic arrays that are indexed by them grow to infinity over time as the objects are spammed. Unused IDs are flushed in the PxScene::simulate() call of PhysX but since there are 0 active physics objects in the scene, the call is always skipped. Removing the 0 object check would be a performance regression.
Adding at least a single physics object to the scene circumvents the issue.