Application Killed when Processing Large Audio Files via Uvicorn (Systemd)

[rajat-rafay]

Issue Description:

I'm experiencing an issue where my application is being killed when processing large audio files. The application works correctly for smaller audio files, but consistently gets killed when processing larger files.

Steps to Reproduce:

1. Run the application using the following command: uvicorn --factory --host 0.0.0.0 speaches.main:create_app
2. Submit a large audio file (e.g., one with a duration of ~2 hours 23 minutes as indicated in the logs) for transcription.
3. Observe the application being killed.

Expected Behavior:

The application should process the large audio file without being killed.

Actual Behavior:

The application is killed during processing of the large audio file. The logs show that the audio processing starts, but the process is terminated before completion.

Logs:

2025-02-26 08:41:47,456:DEBUG:asyncio:__init__:64:Using selector: EpollSelector
2025-02-26 08:41:47,457:DEBUG:speaches.model_manager:load_model:134:Loading model Systran/faster-distil-whisper-large-v3
2025-02-26 08:41:47,457:DEBUG:speaches.model_manager:load_model:136:Acquired lock
2025-02-26 08:41:47,457:DEBUG:speaches.model_manager:load_model:138:Systran/faster-distil-whisper-large-v3 model already loaded
2025-02-26 08:41:47,457:DEBUG:speaches.model_manager:_increment_ref:73:Model was set to expire in 300s, cancelling
2025-02-26 08:41:47,457:DEBUG:speaches.model_manager:_increment_ref:75:Incremented ref count for Systran/faster-distil-whisper-large-v3, self.ref_count=1
2025-02-26 08:41:47,457:INFO:faster_whisper:transcribe:839:Processing audio with duration 02:23:56.372
Killed

Additional Information:

• Operating System: (Ubuntu 24)
• GPU: 2 GPU running on t4
• Model Used: Systran/faster-distil-whisper-large-v3
• Kubernetes Comparison: Interestingly, when running the same application as a container image in Kubernetes on the same node, the issue does not occur, and large files are processed successfully.

Please help me out here 🙏

[rajat-rafay]

Found out the issue. The process got killed because of the less memory I was having on my machine.
Closing the issue.