Make upload limit and Caddy settings configurable via env vars

- API: read the upload size limit from MAX_UPLOAD_MEGABYTES (default 15)
  and keep MAX_UPLOAD_BYTES as the byte value used by /predict
- Caddy: read the ACME email and request body limit from CADDY_EMAIL and
  CADDY_MAX_BODY (default 16MB, ~1 MB above the API limit)
- compose: pass the new variables to caddy_proxy and cnn_api

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

api/src/production/server.py

@@ -45,7 +45,7 @@ QUEUED_INFERENCES = max(0, int(os.getenv("QUEUED_INFERENCES", "1")))
 # PyTorch threads used by EACH running prediction. Total CPU use is roughly
 # PARALLEL_INFERENCES * TORCH_THREADS, so keep the product <= your CPU cores.
 TORCH_THREADS = max(1, int(os.getenv("TORCH_THREADS", "1")))
-MAX_UPLOAD_BYTES = 5 * 1024 * 1024
+MAX_UPLOAD_BYTES = int(max(1, float(os.getenv("MAX_UPLOAD_MEGABYTES", "15"))) * 1024 * 1024)
 MAX_FRAME_BYTES = 1 * 1024 * 1024
 MAX_IMAGE_PIXELS = 20_000_000
 MAX_WEBSOCKETS = 3

caddy/Caddyfile

@@ -1,5 +1,5 @@
 {
-	email marvin.krausser277@gmail.com
+	email {$CADDY_EMAIL:marvin.krausser277@gmail.com}
 }
 
 (common) {
@@ -25,7 +25,7 @@ marvinkrausser.com {
 api.marvinkrausser.com {
 	import common
 	request_body {
-		max_size 10MB
+		max_size {$CADDY_MAX_BODY:16MB}
 	}
 	reverse_proxy cnn_api:8000
 }

compose.yaml

@@ -7,6 +7,10 @@ services:
       - "80:80"
       - "443:443"
     restart: unless-stopped
+    environment:
+      - CADDY_EMAIL=marvin.krausser277@gmail.com
+      # Keep about 1 MB above the API's MAX_UPLOAD_MEGABYTES (multipart overhead).
+      - CADDY_MAX_BODY=16MB
     volumes:
       - ./caddy/data:/data/caddy
       - ./caddy/config:/config/caddy
@@ -41,6 +45,7 @@ services:
       - PREDICT_RATE_WINDOW=60
       - REVIEW_RATE_LIMIT=5
       - REVIEW_RATE_WINDOW=60
+      - MAX_UPLOAD_MEGABYTES=15
     cpus: "5"
     mem_limit: 3g