wip

start_monitoring.py

@@ -0,0 +1,160 @@
+#!/usr/bin/env python3
+"""
+Helper script to start monitoring services for RL training
+"""
+
+import subprocess
+import sys
+import time
+import requests
+import os
+import json
+from pathlib import Path
+
+# Available ports to try for TensorBoard
+TENSORBOARD_PORTS = [6006, 6007, 6008, 6009, 6010, 6011, 6012]
+
+def check_port(port, service_name):
+    """Check if a service is running on the specified port"""
+    try:
+        response = requests.get(f"http://localhost:{port}", timeout=3)
+        print(f"✅ {service_name} is running on port {port}")
+        return True
+    except requests.exceptions.RequestException:
+        return False
+
+def is_port_in_use(port):
+    """Check if a port is already in use"""
+    import socket
+    with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
+        try:
+            s.bind(('localhost', port))
+            return False
+        except OSError:
+            return True
+
+def find_available_port(ports_list, service_name):
+    """Find an available port from the list"""
+    for port in ports_list:
+        if not is_port_in_use(port):
+            print(f"🔍 Found available port {port} for {service_name}")
+            return port
+        else:
+            print(f"⚠️  Port {port} is already in use")
+    return None
+
+def save_port_config(tensorboard_port):
+    """Save the port configuration to a file"""
+    config = {
+        "tensorboard_port": tensorboard_port,
+        "web_dashboard_port": 8051
+    }
+    with open("monitoring_ports.json", "w") as f:
+        json.dump(config, f, indent=2)
+    print(f"💾 Port configuration saved to monitoring_ports.json")
+
+def start_tensorboard():
+    """Start TensorBoard in background on an available port"""
+    try:
+        # First check if TensorBoard is already running on any of our ports
+        for port in TENSORBOARD_PORTS:
+            if check_port(port, "TensorBoard"):
+                print(f"✅ TensorBoard already running on port {port}")
+                save_port_config(port)
+                return port
+                
+        # Find an available port
+        port = find_available_port(TENSORBOARD_PORTS, "TensorBoard")
+        if port is None:
+            print(f"❌ No available ports found in range {TENSORBOARD_PORTS}")
+            return None
+            
+        print(f"🚀 Starting TensorBoard on port {port}...")
+        
+        # Create runs directory if it doesn't exist
+        Path("runs").mkdir(exist_ok=True)
+        
+        # Start TensorBoard
+        if os.name == 'nt':  # Windows
+            subprocess.Popen([
+                sys.executable, "-m", "tensorboard", 
+                "--logdir=runs", f"--port={port}", "--reload_interval=1"
+            ], creationflags=subprocess.CREATE_NEW_CONSOLE)
+        else:  # Linux/Mac
+            subprocess.Popen([
+                sys.executable, "-m", "tensorboard", 
+                "--logdir=runs", f"--port={port}", "--reload_interval=1"
+            ], stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
+        
+        # Wait for TensorBoard to start
+        print(f"⏳ Waiting for TensorBoard to start on port {port}...")
+        for i in range(15):
+            time.sleep(2)
+            if check_port(port, "TensorBoard"):
+                save_port_config(port)
+                return port
+                
+        print(f"⚠️  TensorBoard failed to start on port {port} within 30 seconds")
+        return None
+        
+    except Exception as e:
+        print(f"❌ Error starting TensorBoard: {e}")
+        return None
+
+def check_web_dashboard_port():
+    """Check if web dashboard port is available"""
+    port = 8051
+    if is_port_in_use(port):
+        print(f"⚠️  Web dashboard port {port} is in use")
+        # Try alternative ports
+        for alt_port in [8052, 8053, 8054, 8055]:
+            if not is_port_in_use(alt_port):
+                print(f"🔍 Alternative port {alt_port} available for web dashboard")
+                return alt_port
+        print("❌ No alternative ports found for web dashboard")
+        return port
+    else:
+        print(f"✅ Web dashboard port {port} is available")
+        return port
+
+def main():
+    """Main function"""
+    print("=" * 60)
+    print("🎯 RL TRAINING MONITORING SETUP")
+    print("=" * 60)
+    
+    # Check web dashboard port
+    web_port = check_web_dashboard_port()
+    
+    # Start TensorBoard
+    tensorboard_port = start_tensorboard()
+    
+    print("\n" + "=" * 60)
+    print("📊 MONITORING STATUS")
+    print("=" * 60)
+    
+    if tensorboard_port:
+        print(f"✅ TensorBoard: http://localhost:{tensorboard_port}")
+        # Update port config
+        save_port_config(tensorboard_port)
+    else:
+        print("❌ TensorBoard: Failed to start")
+        print("   Manual start: python -m tensorboard --logdir=runs --port=6007")
+    
+    if web_port:
+        print(f"✅ Web Dashboard: Ready on port {web_port}")
+    
+    print(f"\n🎯 Ready to start RL training!")
+    if tensorboard_port and web_port != 8051:
+        print(f"Run: python train_realtime_with_tensorboard.py --episodes 10 --web-port {web_port}")
+    else:
+        print("Run: python train_realtime_with_tensorboard.py --episodes 10")
+    
+    print(f"\n📋 Available URLs:")
+    if tensorboard_port:
+        print(f"   📊 TensorBoard: http://localhost:{tensorboard_port}")
+    if web_port:
+        print(f"   🌐 Web Dashboard: http://localhost:{web_port} (starts with training)")
+
+if __name__ == "__main__":
+    main()