Memory leak in Core

[alexsanderp]

Memory leak in Core

Describe the bug

When running CompreFace Core 1.2.0-mobilenet-gpu with uwsgi workers, the RAM usage of the workers continuously grows until it reaches the pod memory limit and the worker crashes.

To Reproduce

Steps to reproduce the behavior:

1. Deploy CompreFace Core 1.2.0-mobilenet-gpu in Kubernetes with uwsgi workers.
2. Send a batch of images for face recognition (various sizes tested).
3. Monitor RAM usage of the worker processes (ps aux, kubectl top pod).
4. Observe that RAM usage continuously increases over time until the pod reaches its memory limit and the worker crashes.

Expected behavior

The RAM usage of the workers should remain stable after processing images, and workers should not crash due to growing RAM usage.

Screenshots

(Optional) Attach Grafana graphs or logs showing RAM usage growth of uwsgi workers.

Desktop (please complete the following information)

• OS: Linux (Docker/Kubernetes environment)
• Browser: N/A
• Version: CompreFace Core 1.2.0-mobilenet-gpu

Additional context

• Pod memory limits: minMemory: 5Gi, maxMemory: 6Gi.
• No code modification was done.