GPU jobs getting stuck in Active state

[alikhanich]

Describe the bug
GPU version is unstable. Transcribe and frame-analysis jobs are getting stuck in Active state in bullmq forever.

To Reproduce
Steps to reproduce the behavior:

1. Import folder
2. Scan
3. start jobs
4. See jobs stuck, python threads idling with py-spy

Expected behavior
Jobs working 1by1

Screenshots
If applicable, add screenshots to help explain your problem.

Desktop (please complete the following information):

• Docker Service: [ml, background-jobs]
• Docker Version [e.g. 22] - docker desktop latest
• Windows

Additional context
`root@9be402551587:/app# py-spy dump --pid 27
Process 27: python /app/python/main.py --host 0.0.0.0 --port 8765
Python v3.10.12 (/usr/bin/python3.10)

Thread 27 (idle): "MainThread"
select (selectors.py:469)
_run_once (asyncio/base_events.py:1871)
run_forever (asyncio/base_events.py:603)
run_until_complete (asyncio/base_events.py:636)
run (asyncio/runners.py:44)
(main.py:173)
Thread 60 (idle): "Thread-1"
wait (threading.py:324)
wait (threading.py:607)
run (tqdm/_monitor.py:60)
_bootstrap_inner (threading.py:1016)
_bootstrap (threading.py:973)
Thread 67 (idle): "ThreadPoolExecutor-2_0"
find_alignment (faster_whisper/transcribe.py:1709)
add_word_timestamps (faster_whisper/transcribe.py:1590)
generate_segments (faster_whisper/transcribe.py:1279)
restore_speech_timestamps (faster_whisper/transcribe.py:1851)
_transcribe_video (service.py:123)
_process_sync (service.py:68)
run (concurrent/futures/thread.py:58)
_worker (concurrent/futures/thread.py:83)
run (threading.py:953)
_bootstrap_inner (threading.py:1016)
_bootstrap (threading.py:973)
Thread 68 (idle): "ThreadPoolExecutor-0_0"
_analyze_frames (service.py:147)
_process_sync (service.py:75)
run (concurrent/futures/thread.py:58)
_worker (concurrent/futures/thread.py:83)
run (threading.py:953)
_bootstrap_inner (threading.py:1016)
_bootstrap (threading.py:973)
Thread 434 (idle): "Thread-3"
wait (threading.py:324)
wait (threading.py:607)
run (tqdm/_monitor.py:60)
_bootstrap_inner (threading.py:1016)
_bootstrap (threading.py:973)
Thread 72 (idle)
`

[alikhanich]

ML image gets stuck after doing frame-analysis and face recognition for one video file. GPU mem graph goes to a flat line at 99% mem consumption, looks like frames are not offloaded between runs.

[alikhanich]

#152

[IliasHad]

Thank you for reporting @alikhanich and for the PR. I'll review this over the weekend and get back to you ASAP