new three layer deployment

ctrl/collector/collector.py

@@ -0,0 +1,136 @@
+#!/usr/bin/env python3
+"""Lightweight WebSocket metrics collector for sysmonstm standalone deployment."""
+
+import asyncio
+import json
+import logging
+import os
+import socket
+import time
+
+import psutil
+
+# Configuration from environment
+HUB_URL = os.environ.get("HUB_URL", "ws://localhost:8080/ws")
+MACHINE_ID = os.environ.get("MACHINE_ID", socket.gethostname())
+API_KEY = os.environ.get("API_KEY", "")
+INTERVAL = int(os.environ.get("INTERVAL", "5"))
+LOG_LEVEL = os.environ.get("LOG_LEVEL", "INFO")
+
+# Logging setup
+logging.basicConfig(
+    level=getattr(logging, LOG_LEVEL.upper(), logging.INFO),
+    format="%(asctime)s [%(levelname)s] %(message)s",
+    datefmt="%Y-%m-%d %H:%M:%S",
+)
+log = logging.getLogger("collector")
+
+
+def collect_metrics() -> dict:
+    """Collect system metrics using psutil."""
+    metrics = {
+        "type": "metrics",
+        "machine_id": MACHINE_ID,
+        "hostname": socket.gethostname(),
+        "timestamp": time.time(),
+    }
+
+    # CPU
+    try:
+        metrics["cpu"] = psutil.cpu_percent(interval=None)
+    except Exception:
+        pass
+
+    # Memory
+    try:
+        mem = psutil.virtual_memory()
+        metrics["memory"] = mem.percent
+        metrics["memory_used_gb"] = round(mem.used / (1024**3), 2)
+        metrics["memory_total_gb"] = round(mem.total / (1024**3), 2)
+    except Exception:
+        pass
+
+    # Disk
+    try:
+        disk = psutil.disk_usage("/")
+        metrics["disk"] = disk.percent
+        metrics["disk_used_gb"] = round(disk.used / (1024**3), 2)
+        metrics["disk_total_gb"] = round(disk.total / (1024**3), 2)
+    except Exception:
+        pass
+
+    # Load average (Unix only)
+    try:
+        load1, load5, load15 = psutil.getloadavg()
+        metrics["load_1m"] = round(load1, 2)
+        metrics["load_5m"] = round(load5, 2)
+        metrics["load_15m"] = round(load15, 2)
+    except (AttributeError, OSError):
+        pass
+
+    # Network connections count
+    try:
+        metrics["connections"] = len(psutil.net_connections(kind="inet"))
+    except (psutil.AccessDenied, PermissionError):
+        pass
+
+    # Process count
+    try:
+        metrics["processes"] = len(psutil.pids())
+    except Exception:
+        pass
+
+    return metrics
+
+
+async def run_collector():
+    """Main collector loop with auto-reconnect."""
+    import websockets
+
+    # Build URL with API key if provided
+    url = HUB_URL
+    if API_KEY:
+        separator = "&" if "?" in url else "?"
+        url = f"{url}{separator}key={API_KEY}"
+
+    # Prime CPU percent (first call always returns 0)
+    psutil.cpu_percent(interval=None)
+
+    while True:
+        try:
+            log.info(f"Connecting to {HUB_URL}...")
+            async with websockets.connect(url) as ws:
+                log.info(
+                    f"Connected. Sending metrics every {INTERVAL}s as '{MACHINE_ID}'"
+                )
+
+                while True:
+                    metrics = collect_metrics()
+                    await ws.send(json.dumps(metrics))
+                    log.debug(
+                        f"Sent: cpu={metrics.get('cpu', '?')}% mem={metrics.get('memory', '?')}% disk={metrics.get('disk', '?')}%"
+                    )
+                    await asyncio.sleep(INTERVAL)
+
+        except asyncio.CancelledError:
+            log.info("Collector stopped")
+            break
+        except Exception as e:
+            log.warning(f"Connection error: {e}. Reconnecting in 5s...")
+            await asyncio.sleep(5)
+
+
+def main():
+    log.info("sysmonstm collector starting")
+    log.info(f"  Hub: {HUB_URL}")
+    log.info(f"  Machine: {MACHINE_ID}")
+    log.info(f"  Interval: {INTERVAL}s")
+
+    try:
+        asyncio.run(run_collector())
+    except KeyboardInterrupt:
+        log.info("Stopped")
+
+
+if __name__ == "__main__":
+    main()