project initialization

backend/monitoring/management_commands/start_metrics_collection.py

@@ -0,0 +1,132 @@
+"""
+Django management command to start metrics collection.
+"""
+
+import time
+import signal
+import sys
+from django.core.management.base import BaseCommand
+from django.conf import settings
+
+from ..exporters import metrics_collector
+from ..alerts import alert_manager
+
+
+class Command(BaseCommand):
+    help = 'Start metrics collection and alert monitoring'
+
+    def add_arguments(self, parser):
+        parser.add_argument(
+            '--interval',
+            type=int,
+            default=30,
+            help='Metrics collection interval in seconds (default: 30)'
+        )
+        parser.add_argument(
+            '--alert-interval',
+            type=int,
+            default=60,
+            help='Alert checking interval in seconds (default: 60)'
+        )
+        parser.add_argument(
+            '--port',
+            type=int,
+            default=8001,
+            help='Metrics server port (default: 8001)'
+        )
+        parser.add_argument(
+            '--metrics-only',
+            action='store_true',
+            help='Only collect metrics, no alerts'
+        )
+        parser.add_argument(
+            '--alerts-only',
+            action='store_true',
+            help='Only check alerts, no metrics collection'
+        )
+        parser.add_argument(
+            '--quiet',
+            action='store_true',
+            help='Run quietly'
+        )
+
+    def handle(self, *args, **options):
+        self.interval = options['interval']
+        self.alert_interval = options['alert_interval']
+        self.port = options['port']
+        self.metrics_only = options['metrics_only']
+        self.alerts_only = options['alerts_only']
+        self.quiet = options['quiet']
+
+        # Set up signal handlers for graceful shutdown
+        signal.signal(signal.SIGINT, self.signal_handler)
+        signal.signal(signal.SIGTERM, self.signal_handler)
+
+        self.running = True
+
+        if not self.quiet:
+            self.stdout.write(
+                self.style.SUCCESS('Starting metrics collection and alert monitoring...')
+            )
+            self.stdout.write(f'Metrics interval: {self.interval} seconds')
+            self.stdout.write(f'Alert interval: {self.alert_interval} seconds')
+
+        try:
+            # Start metrics collection
+            if not self.alerts_only:
+                if not self.quiet:
+                    self.stdout.write('Starting metrics collection...')
+                metrics_collector.start_collection(self.interval)
+
+            # Start alert monitoring
+            if not self.metrics_only:
+                if not self.quiet:
+                    self.stdout.write('Starting alert monitoring...')
+                self.start_alert_monitoring()
+
+            # Keep the command running
+            if not self.quiet:
+                self.stdout.write('Monitoring started. Press Ctrl+C to stop.')
+
+            while self.running:
+                time.sleep(1)
+
+        except KeyboardInterrupt:
+            if not self.quiet:
+                self.stdout.write('\nReceived interrupt signal, stopping...')
+        finally:
+            self.shutdown()
+
+    def start_alert_monitoring(self):
+        """Start alert monitoring in a separate thread."""
+        import threading
+
+        def alert_monitor():
+            while self.running:
+                try:
+                    alert_manager.check_rules()
+                    time.sleep(self.alert_interval)
+                except Exception as e:
+                    if not self.quiet:
+                        self.stdout.write(
+                            self.style.ERROR(f'Error in alert monitoring: {e}')
+                        )
+                    time.sleep(self.alert_interval)
+
+        alert_thread = threading.Thread(target=alert_monitor, daemon=True)
+        alert_thread.start()
+
+    def signal_handler(self, signum, frame):
+        """Handle shutdown signals."""
+        if not self.quiet:
+            self.stdout.write(f'\nReceived signal {signum}, shutting down...')
+        self.running = False
+
+    def shutdown(self):
+        """Shutdown the monitoring system."""
+        if not self.quiet:
+            self.stdout.write('Shutting down metrics collection...')
+        metrics_collector.stop_collection()
+
+        if not self.quiet:
+            self.stdout.write(self.style.SUCCESS('Monitoring stopped.'))