poc3

app/main.py

@@ -21,6 +21,7 @@ from .flow_tracker import FlowTracker
 from .forensics import ForensicPcapRing
 from .live import EventBus, LiveEventPipeline, TrafficHistory, TrafficNormalizer
 from .maintenance import clear_suricata_logs, storage_info
+from .metrics import PrometheusMetrics
 from .ndr import NDRAnalyzer, ThreatIntelManager
 from .notifier import WebhookNotifier
 from .policy import PolicyEngine
@@ -210,7 +211,7 @@ def main() -> int:
         cfg.redis_snapshot_seconds,
         cfg.redis_aof,
     )
-    if cfg.redis_managed and not redis_supervisor.start(wait_ready_seconds=15):
+    if cfg.redis_managed and not redis_supervisor.start(wait_ready_seconds=120):
         raise RuntimeError(
             f"managed Redis failed to start: {redis_supervisor.status().get('last_error') or 'unknown error'}"
         )
@@ -422,6 +423,45 @@ def main() -> int:
             "runtime": stats.snapshot(),
         }
 
+    version_path = Path(__file__).resolve().parents[1] / "VERSION"
+    try:
+        app_version = version_path.read_text(encoding="ascii").strip() or "unknown"
+    except OSError:
+        app_version = "unknown"
+
+    def metrics_state() -> dict:
+        return {
+            "components": {
+                "web": True,
+                "tzsp": receiver.is_alive() and receiver.sock is not None,
+                "tap": tap.fd is not None,
+                "suricata": suricata.poll() is None,
+                "eve": watcher.is_alive(),
+            },
+            "features": {
+                "auto_block": cfg.auto_block,
+                "ndr": cfg.ndr_enabled,
+                "notifications": notifier.enabled,
+                "routeros": routeros.configured,
+                "redis_managed": cfg.redis_managed,
+                "forensic_pcap": cfg.forensic_pcap_mode != "off",
+            },
+        }
+
+    prometheus_metrics = PrometheusMetrics(
+        stats,
+        version=app_version,
+        mode="full",
+        started_at=started_at,
+        state_provider=metrics_state,
+        flow_tracker=flow_tracker,
+        event_bus=event_bus,
+        live_pipeline=live_pipeline,
+        ndr_analyzer=ndr_analyzer,
+        notifier=notifier,
+        forensic_pcap=forensic_pcap,
+    )
+
     web = WebServer(
         cfg,
         store,
@@ -437,6 +477,7 @@ def main() -> int:
         ndr_analyzer=ndr_analyzer,
         backup_manager=backup_manager,
         forensic_pcap=forensic_pcap,
+        metrics_provider=prometheus_metrics.render,
     )
 
     def housekeeping() -> None: