update monitor

main.py

@@ -4,8 +4,9 @@ from fastapi.staticfiles import StaticFiles
 from fastapi.middleware.cors import CORSMiddleware
 import os
 from dotenv import load_dotenv
-from monitor_logic import get_ssh_data, check_web_service
+from monitor_logic import get_ssh_data, check_web_service, check_testarena_backend_status, recover_testarena_backend, send_email_notification
 import asyncio
+from datetime import datetime
 
 load_dotenv()
 
@@ -32,6 +33,9 @@ PI_PORT = 2222
 PI_USER = "asf_tb"
 PI_PASS = os.getenv("PI_PASS", "ASF_TB")
 
+SMTP_USER = "support@nabd-co.com"
+SMTP_PASS = "zwziglbpxyfogafc"
+
 WEB_SERVICES = [
     {"name": "Gitea", "url": "https://gitea.nabd-co.com/"},
     {"name": "OpenProject", "url": "https://openproject.nabd-co.com/"},
@@ -41,34 +45,76 @@ WEB_SERVICES = [
     {"name": "Board", "url": "https://board.nabd-co.com/"},
 ]
 
+# Global state to store latest status
+LATEST_STATUS = {
+    "pc": {"status": "offline"},
+    "pi": {"status": "offline"},
+    "services": [],
+    "last_update": None
+}
+
+async def monitor_services_task():
+    while True:
+        try:
+            # Run SSH checks
+            pc_res = await asyncio.to_thread(get_ssh_data, PC_HOST, PC_PORT, PC_USER, PC_PASS)
+            pi_res = await asyncio.to_thread(get_ssh_data, PI_HOST, PI_PORT, PI_USER, PI_PASS)
+            
+            # Run web checks
+            web_results = []
+            for s in WEB_SERVICES:
+                res = await asyncio.to_thread(check_web_service, s["url"])
+                web_results.append({**s, **res})
+                if res["status"] != "online":
+                    await asyncio.to_thread(send_email_notification, SMTP_USER, SMTP_PASS, s["name"], res["status"])
+
+            # Check TestArena Backend
+            ta_res = await asyncio.to_thread(check_testarena_backend_status)
+            
+            if ta_res["status"] != "online":
+                # Recovery loop
+                for attempt in range(1, 6):
+                    print(f"Attempting recovery for TestArena Backend (Attempt {attempt})...")
+                    success = await asyncio.to_thread(recover_testarena_backend, PC_HOST, PC_PORT, PC_USER, PC_PASS, attempt)
+                    if success:
+                        # Wait a bit for services to start
+                        await asyncio.sleep(10)
+                        ta_res = await asyncio.to_thread(check_testarena_backend_status)
+                        if ta_res["status"] == "online":
+                            print("TestArena Backend recovered successfully.")
+                            break
+                
+                if ta_res["status"] != "online":
+                    print("TestArena Backend recovery failed after 5 attempts.")
+                    await asyncio.to_thread(send_email_notification, SMTP_USER, SMTP_PASS, "TestArena Backend", ta_res["status"])
+
+            web_results.append({
+                "name": "TestArena Backend",
+                "url": "http://asf-server.duckdns.org:8080/api/system/status",
+                **ta_res
+            })
+
+            LATEST_STATUS["pc"] = pc_res
+            LATEST_STATUS["pi"] = pi_res
+            LATEST_STATUS["services"] = web_results
+            LATEST_STATUS["last_update"] = datetime.now().isoformat()
+            
+        except Exception as e:
+            print(f"Monitor task error: {e}")
+            
+        await asyncio.sleep(300) # Run every 5 minutes
+
+@app.on_event("startup")
+async def startup_event():
+    asyncio.create_task(monitor_services_task())
+
 @app.get("/")
 async def read_index():
     return FileResponse("monitor.html")
 
 @app.get("/api/status")
 async def get_status():
-    # Run SSH checks in parallel
-    pc_task = asyncio.to_thread(get_ssh_data, PC_HOST, PC_PORT, PC_USER, PC_PASS)
-    pi_task = asyncio.to_thread(get_ssh_data, PI_HOST, PI_PORT, PI_USER, PI_PASS)
-    
-    # Run web checks in parallel
-    web_tasks = [asyncio.to_thread(check_web_service, s["url"]) for s in WEB_SERVICES]
-    
-    pc_res, pi_res, *web_res = await asyncio.gather(pc_task, pi_task, *web_tasks)
-    
-    services = []
-    for i, res in enumerate(web_res):
-        services.append({
-            "name": WEB_SERVICES[i]["name"],
-            "url": WEB_SERVICES[i]["url"],
-            **res
-        })
-        
-    return {
-        "pc": pc_res,
-        "pi": pi_res,
-        "services": services
-    }
+    return LATEST_STATUS
 
 if __name__ == "__main__":
     import uvicorn

monitor.html

@@ -509,18 +509,34 @@
 
                 // Update Services
                 const serviceList = document.getElementById('service-list');
-                serviceList.innerHTML = data.services.map(s => `
+                serviceList.innerHTML = data.services.map(s => {
+                    let statusColor = 'var(--status-red)';
+                    let statusIcon = 'fa-times-circle';
+                    let statusText = s.status;
+
+                    if (s.status === 'online') {
+                        statusColor = 'var(--status-green)';
+                        statusIcon = 'fa-check-circle';
+                        statusText = s.latency;
+                    } else if (s.status === 'partial') {
+                        statusColor = 'var(--status-yellow)';
+                        statusIcon = 'fa-exclamation-circle';
+                        statusText = 'Partial';
+                    }
+
+                    return `
                         <li class="service-item">
                             <div class="service-name">
                                 <img src="${getServiceLogo(s.name)}" class="service-logo" onerror="this.src='https://via.placeholder.com/32?text=${s.name.charAt(0)}'">
                                 ${s.name}
                             </div>
-                        <div class="service-status" style="color: ${s.status === 'online' ? 'var(--status-green)' : 'var(--status-red)'}">
-                            <i class="fas ${s.status === 'online' ? 'fa-check-circle' : 'fa-times-circle'}"></i>
-                            ${s.status === 'online' ? s.latency : 'Offline'}
+                            <div class="service-status" style="color: ${statusColor}">
+                                <i class="fas ${statusIcon}"></i>
+                                ${statusText}
                             </div>
                         </li>
-                `).join('');
+                    `;
+                }).join('');
 
                 document.getElementById('loading').style.opacity = '0';
                 setTimeout(() => document.getElementById('loading').style.display = 'none', 500);
@@ -537,7 +553,8 @@
                 'Draw.io': 'https://app.diagrams.net/images/logo-flat.svg',
                 'TestArena': '/static/testarena.png',
                 'TBM': '/static/tbm.ico',
-                'Board': 'https://excalidraw.com/favicon-32x32.png'
+                'Board': 'https://excalidraw.com/favicon-32x32.png',
+                'TestArena Backend': '/static/testarena.png'
             };
             return logos[name] || '';
         }

monitor_logic.py

@@ -2,6 +2,9 @@ import paramiko
 import requests
 import time
 import re
+import smtplib
+from email.mime.text import MIMEText
+from email.mime.multipart import MIMEMultipart
 
 def get_ssh_data(host, port, user, password):
     try:
@@ -57,3 +60,64 @@ def check_web_service(url):
             return {"status": "error", "code": response.status_code}
     except Exception:
         return {"status": "offline"}
+
+def check_testarena_backend_status():
+    url = "http://asf-server.duckdns.org:8080/api/system/status"
+    try:
+        start_time = time.time()
+        response = requests.get(url, timeout=5)
+        latency = int((time.time() - start_time) * 1000)
+        if response.status_code == 200:
+            data = response.json()
+            # {"testarena-app":"online","testarena-worker":"online","nginx":"online"}
+            is_online = all(v == "online" for v in data.values())
+            if is_online:
+                return {"status": "online", "latency": f"{latency}ms", "details": data}
+            else:
+                return {"status": "partial", "details": data}
+        else:
+            return {"status": "error", "code": response.status_code}
+    except Exception:
+        return {"status": "offline"}
+
+def recover_testarena_backend(host, port, user, password, attempt):
+    try:
+        ssh = paramiko.SSHClient()
+        ssh.set_missing_host_key_policy(paramiko.AutoAddPolicy())
+        ssh.connect(host, port=port, username=user, password=password, timeout=10)
+        
+        if attempt == 1:
+            # system restart
+            cmd = "echo 'ASF' | sudo -S /home/asf/testarena_backend/restart_services.sh"
+        else:
+            # system redeploy
+            cmd = "echo 'ASF' | sudo -S /home/asf/testarena_backend/deploy.sh"
+            
+        stdin, stdout, stderr = ssh.exec_command(cmd)
+        exit_status = stdout.channel.recv_exit_status()
+        ssh.close()
+        return exit_status == 0
+    except Exception as e:
+        print(f"Recovery failed: {e}")
+        return False
+
+def send_email_notification(smtp_user, smtp_pass, service_name, status):
+    try:
+        msg = MIMEMultipart()
+        msg['From'] = smtp_user
+        msg['To'] = smtp_user # Send to self as requested
+        msg['Subject'] = f"ALERT: Service {service_name} is {status}"
+        
+        body = f"The service {service_name} is currently {status}. Please check the system."
+        msg.attach(MIMEText(body, 'plain'))
+        
+        server = smtplib.SMTP('smtp.gmail.com', 587) # Assuming Gmail based on the password format
+        server.starttls()
+        server.login(smtp_user, smtp_pass)
+        text = msg.as_string()
+        server.sendmail(smtp_user, smtp_user, text)
+        server.quit()
+        return True
+    except Exception as e:
+        print(f"Email failed: {e}")
+        return False