Retry port binding in start.sh for dev mode restarts

backend/start.sh

@@ -1,15 +1,31 @@
 #!/bin/bash
 # Entrypoint for HF Spaces dev mode compatibility.
-# Dev mode spawns CMD multiple times simultaneously on restart.
-# Only the first instance can bind port 7860 — the rest must exit
-# with code 0 so the dev mode daemon doesn't mark the app as crashed.
-
-# Run uvicorn; if it fails due to port conflict, exit cleanly.
-uvicorn main:app --host 0.0.0.0 --port 7860
-EXIT_CODE=$?
-
-if [ $EXIT_CODE -ne 0 ]; then
-    # Check if this was a port-in-use failure (another instance already running)
-    echo "uvicorn exited with code $EXIT_CODE, exiting gracefully."
-    exit 0
-fi
+# Dev mode spawns multiple CMD instances simultaneously on restart.
+# The old process may still hold port 7860 briefly, so we retry
+# with backoff until the port is free.
+
+MAX_RETRIES=5
+RETRY_DELAY=2
+
+for i in $(seq 1 $MAX_RETRIES); do
+    uvicorn main:app --host 0.0.0.0 --port 7860
+    EXIT_CODE=$?
+
+    if [ $EXIT_CODE -eq 0 ]; then
+        exit 0
+    fi
+
+    # Check if another instance from this restart batch is already running
+    if ss -tlnp 2>/dev/null | grep -q ":7860 "; then
+        echo "Port 7860 already bound by another instance, exiting."
+        exit 0
+    fi
+
+    if [ $i -lt $MAX_RETRIES ]; then
+        echo "uvicorn exited ($EXIT_CODE), retrying in ${RETRY_DELAY}s (attempt $i/$MAX_RETRIES)..."
+        sleep $RETRY_DELAY
+    fi
+done
+
+echo "Failed to bind port 7860 after $MAX_RETRIES attempts."
+exit 1