Release FLUX CUDA context immediately

platform/docker/flux-worker/flux_worker.py

@@ -6,6 +6,7 @@ from __future__ import annotations
 import gc
 import json
 import os
+import signal
 import time
 from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
 from pathlib import Path
@@ -21,6 +22,11 @@ LOAD_SECONDS = 0.0
 if len(TOKEN) < 32:
     raise RuntimeError("WORKER_TOKEN is missing or too short")
 
+# The container is intentionally disposable. After the router has received a
+# completed response and saved image, an immediate process exit releases the
+# CUDA context much faster than Python/PyTorch interpreter teardown.
+signal.signal(signal.SIGTERM, lambda *_: os._exit(0))
+
 
 def load_pipeline() -> None:
     global PIPE, LOAD_SECONDS