Update Dockerfile

Dockerfile

@@ -199,6 +199,74 @@ wait_for_port() {
     return 1
 }
 
+# Function to start the frontend server
+start_frontend() {
+    cd /app/web
+    # Kill any existing node processes
+    pkill -f "node server.js" || true
+    # Ensure the output log directory exists
+    touch output.log
+    echo "Starting Next.js server at $(date)" > output.log
+    # Configure Next.js to listen on all interfaces
+    export HOSTNAME="0.0.0.0"
+    export NODE_ENV=production
+    export PORT=3000
+    # Start the server with proper error handling
+    node server.js >> output.log 2>&1 &
+    local pid=$!
+    echo "Frontend server started with PID: $pid"
+    # Wait a bit to ensure process is stable
+    sleep 5
+    if ! kill -0 $pid 2>/dev/null; then
+        echo "Frontend server failed to start. Last 50 lines of logs:"
+        tail -n 50 output.log
+        return 1
+    fi
+    return 0
+}
+
+# Function to monitor and restart services if needed
+monitor_services() {
+    local restart_count=0
+    local max_restarts=3
+    
+    while true; do
+        if ! pgrep -f "gunicorn" > /dev/null; then
+            echo "API server died"
+            tail -n 100 /app/api/gunicorn.error.log || true
+            exit 1
+        fi
+        
+        if ! pgrep -f "node server.js" > /dev/null; then
+            echo "Frontend server died. Attempting restart..."
+            if [ $restart_count -lt $max_restarts ]; then
+                restart_count=$((restart_count + 1))
+                echo "Restart attempt $restart_count of $max_restarts"
+                if start_frontend; then
+                    echo "Frontend server restarted successfully"
+                    # Reset counter on successful restart
+                    restart_count=0
+                else
+                    echo "Failed to restart frontend server"
+                    tail -n 100 /app/web/output.log || true
+                    exit 1
+                fi
+            else
+                echo "Maximum frontend restart attempts reached"
+                tail -n 100 /app/web/output.log || true
+                exit 1
+            fi
+        fi
+        
+        if ! pgrep -f "nginx" > /dev/null; then
+            echo "Nginx died"
+            tail -n 100 /var/log/nginx/error.log || true
+            exit 1
+        fi
+        sleep 5
+    done
+}
+
 # Initialize PostgreSQL database if not already initialized
 if [ ! -f "$PGDATA/PG_VERSION" ]; then
     echo "Initializing PostgreSQL database..."
@@ -268,22 +336,15 @@ cd /app/api && poetry run python -m gunicorn app:app \
 echo "Waiting for API server to be ready..."
 wait_for_port 5001 "API" 30 2 "127.0.0.1"
 
-# Start frontend server with logging
+# Start frontend server
 echo "Starting frontend server..."
-cd /app/web
-# Ensure the output log directory exists
-touch output.log
-echo "Starting Next.js server at $(date)" > output.log
-
-# Configure Next.js to listen on all interfaces
-export HOSTNAME="0.0.0.0"
-PORT=3000 node server.js >> output.log 2>&1 &
+start_frontend
 
 # Get container IP
 CONTAINER_IP=$(hostname -i || ip route get 1 | awk '{print $7}' || echo "127.0.0.1")
 echo "Container IP: $CONTAINER_IP"
 
-# Wait for frontend to be ready with increased timeout
+# Wait for frontend to be ready
 echo "Waiting for frontend server to be ready..."
 if ! wait_for_port 3000 "Frontend" 60 5 "0.0.0.0"; then
     echo "Frontend server failed to start. Last 50 lines of frontend logs:"
@@ -300,25 +361,8 @@ wait_for_port 7860 "Nginx" 30 2 "0.0.0.0"
 
 echo "All services are running. Starting monitoring..."
 
-# Monitor all processes
-while true; do
-    if ! pgrep -f "gunicorn" > /dev/null; then
-        echo "API server died"
-        tail -n 100 /app/api/gunicorn.error.log || true
-        exit 1
-    fi
-    if ! pgrep -f "node server.js" > /dev/null; then
-        echo "Frontend server died"
-        tail -n 100 /app/web/output.log || true
-        exit 1
-    fi
-    if ! pgrep -f "nginx" > /dev/null; then
-        echo "Nginx died"
-        tail -n 100 /var/log/nginx/error.log || true
-        exit 1
-    fi
-    sleep 30
-done
+# Start monitoring services
+monitor_services
 EOT
 
 # Set permissions for entrypoint script