Monitoring & Observability

• Console pages: server CPU/memory/disk, PostgreSQL connections & slow queries, Redis memory & hit rate, cron job registry — plus a service-health overview card (concurrent /ready probes across all services, unhealthy first, 10 s auto-refresh) and a distributed-job heartbeats card.
• Health: every service exposes /api/v1/health/live and /ready (DB ping), wired into compose health checks; monitor's /monitor/services aggregates the probes for the overview card.
• Job heartbeats: in-process crons, standalone workers and host shell scripts all report liveness via shared/pkg/jobbeat (one call per run into ops_job_heartbeats); anything silent for over 2× its interval is flagged stale in the console. Shell scripts join with a single curl.
• Metrics: Prometheus scraping (zero-dependency shared/pkg/metrics: HTTP counters/error rates/latency histograms + Go runtime + DB pool) + node_exporter host metrics + Grafana dashboards under platform/deploy/ (optional profile).
• Alerting loop (optional): Prometheus alert rules (service down / low disk / high memory / 5xx surge, with for windows absorbing rolling-update noise) → Alertmanager (grouping, dedup, resolved notifications, inhibition) → notify webhook → in-console notifications. If notify isn't running, failed delivery is just logged.
• Tracing (optional): OpenTelemetry SDK pre-wired; set OTEL_EXPORTER_OTLP_ENDPOINT to enable, with request_id propagated through logs and responses.
• Logs: login logs (durable NATS consumption), operation logs (middleware-level with latency and request_id, CSV export), audit logs — all queryable via the audit service.