Upload gunicorn_config.py

gunicorn_config.py

@@ -1,6 +1,6 @@
 """
 Gunicorn configuration for AEGIS Economics AI
-Optimized for Hugging Face Spaces deployment
+Optimized for Hugging Face Spaces deployment with memory constraints
 """
 
 import multiprocessing
@@ -10,16 +10,16 @@ import os
 bind = "0.0.0.0:7860"
 backlog = 2048
 
-# Worker processes
-workers = 1  # Single worker for model loading efficiency
+# Worker processes - optimized for memory constraints
+workers = 1  # Single worker to avoid memory issues
 worker_class = "sync"
 worker_connections = 1000
-timeout = 300  # 5 minutes for model loading
+timeout = 600  # 10 minutes timeout for model inference
 keepalive = 2
 
-# Restart workers
-max_requests = 1000
-max_requests_jitter = 50
+# Restart workers to prevent memory leaks
+max_requests = 50  # Restart more frequently to prevent memory buildup
+max_requests_jitter = 10
 preload_app = True
 
 # Logging
@@ -37,4 +37,19 @@ worker_tmp_dir = "/dev/shm"  # Use shared memory for better performance
 # Security
 limit_request_line = 4094
 limit_request_fields = 100
-limit_request_field_size = 8190
+limit_request_field_size = 8190
+
+# Graceful shutdown
+graceful_timeout = 60
+
+def when_ready(server):
+    server.log.info("AEGIS Economics AI server is ready. Listening on %s", server.address)
+
+def worker_int(worker):
+    worker.log.info("Worker received INT or QUIT signal")
+
+def pre_fork(server, worker):
+    server.log.info("Worker spawned (pid: %s)", worker.pid)
+
+def post_fork(server, worker):
+    server.log.info("Worker spawned (pid: %s)", worker.pid)