Metrics and monitoring
Airgate exposes its health and activity in four ways: a health endpoint for load balancers and uptime checks, Prometheus metrics for capacity and error rates, alerts in the console, and export of events and logs to your SIEM. This page is for operators who run Airgate on their own infrastructure and connect it to their monitoring.
Health endpoint
Section titled “Health endpoint”GET /healthz on the public HTTPS listener (for example https://nac.contoso.com/healthz) needs no authentication.
| Status | Body | Meaning |
|---|---|---|
200 |
{"status": "ok", "events_dropped": 0, "event_relay": "connected"} |
Airgate serves and its database answered within 2 seconds |
503 |
{"error": "database unavailable", "code": "db_unavailable"} |
Airgate runs but cannot reach its database |
event_relay is connected, disconnected or off (single replica). events_dropped counts internal events dropped because a subscriber’s queue was full; it should stay at 0.
Use it for uptime monitoring and as a load balancer health check.
Prometheus metrics
Section titled “Prometheus metrics”nacd serves metrics in the Prometheus text format at GET /metrics on its internal listener (port 8091), next to the internal API the RADIUS service uses. The internal listener is bound to the deployment’s private network and is never published on the appliance’s network interface or to the internet.
- Without a token the endpoint is open on that private network, like
/healthz. - With
METRICS_TOKENset in the deployment’s settings, every scrape must sendAuthorization: Bearer <token>; other requests get401.
Attach your Prometheus server to the deployment’s private network and scrape each nacd replica:
scrape_configs: - job_name: airgate scheme: http authorization: type: Bearer credentials_file: /etc/prometheus/airgate-metrics-token static_configs: - targets: ['nacd:8091']Counters are per nacd replica and start at zero when the process starts; use rate() and increase(). Each metric family holds at most 4,096 label combinations; further combinations are counted under the label value other.
Sign-ins and accounting
Section titled “Sign-ins and accounting”| Metric | Type | Labels | Meaning |
|---|---|---|---|
nac_authentications_total |
counter | tenant, protocol, kind, result, method |
Access decisions written to the Sign-in log: RADIUS Access-Request, TACACS+ authentication and authorization, web and application logins. tenant is the tenant ID; result is for example ACCEPT or REJECT |
nac_accounting_records_total |
counter | tenant, protocol |
RADIUS and TACACS+ accounting records |
RADIUS decisions
Section titled “RADIUS decisions”| Metric | Type | Labels | Meaning |
|---|---|---|---|
nac_radius_requests_total |
counter | route, status |
Calls from the RADIUS service answered by nacd, by call site (route) and status class: 2xx accept or continue, 4xx reject, 5xx no answer. rate() is the decision rate |
nac_radius_request_duration_seconds |
histogram | route |
Decision latency (_bucket, _sum, _count) |
nac_radius_handler_panics_total |
counter | route |
Internal errors while deciding; each was answered with a deny |
nac_radius_conversations |
gauge | Multi-round EAP conversations held in memory | |
nac_radius_effects_queued |
gauge | Background work waiting for a worker (Sign-in log records, sessions, CoA) | |
nac_radius_effects_overflow_total |
counter | Background work that found the queue full |
Site connectors
Section titled “Site connectors”| Metric | Type | Labels | Meaning |
|---|---|---|---|
nac_edge_connectors_online |
gauge | tenant |
Enabled, enrolled connectors seen within the offline threshold (cluster-wide) |
nac_edge_connectors_enrolled |
gauge | tenant |
Enabled, enrolled connectors (cluster-wide) |
nac_edge_jobs_total |
counter | kind, result |
Jobs relayed to connectors. kind: coa (CoA or Disconnect to a device behind a connector), ntlm_auth (Active Directory password check). result: ok, timeout, nas_timeout, unavailable, canceled, error |
nac_edge_tunnel_radius_total |
counter | result |
RADIUS packets connectors sent through the HTTPS tunnel |
nac_edge_tacacs_tunnels_open |
gauge | TACACS+ sessions tunnelled through this replica now |
Process and database
Section titled “Process and database”| Metric | Type | Meaning |
|---|---|---|
nac_up |
gauge | 1 while nacd serves |
go_goroutines |
gauge | Goroutines in the process |
nac_events_dropped_total |
counter | Internal events dropped because a subscriber queue was full |
nac_db_pool_max_conns |
gauge | Maximum size of the database pool (DB_MAX_CONNS) |
nac_db_pool_total_conns |
gauge | Open database connections |
nac_db_pool_acquired_conns |
gauge | Database connections in use |
nac_db_pool_idle_conns |
gauge | Idle database connections |
nac_db_pool_empty_acquire_total |
counter | Times a request had to wait for a free database connection |
Example alert rules
Section titled “Example alert rules”groups: - name: airgate rules: - alert: AirgateDown expr: up{job="airgate"} == 0 or nac_up == 0 for: 2m - alert: AirgateRejectRateHigh expr: | sum(rate(nac_authentications_total{result="REJECT"}[5m])) / sum(rate(nac_authentications_total[5m])) > 0.3 for: 10m - alert: AirgateRadiusErrors expr: increase(nac_radius_requests_total{status="5xx"}[5m]) > 0 - alert: AirgateDecisionSlow expr: | histogram_quantile(0.95, sum by (le) (rate(nac_radius_request_duration_seconds_bucket[5m]))) > 1 for: 10m - alert: AirgateSiteConnectorOffline expr: sum(nac_edge_connectors_online) < sum(nac_edge_connectors_enrolled) for: 5m - alert: AirgateDatabasePoolExhausted expr: increase(nac_db_pool_empty_acquire_total[5m]) > 0 for: 10m - alert: AirgateEventsDropped expr: increase(nac_events_dropped_total[10m]) > 0Adjust thresholds to your normal reject rate and traffic.
Alerts in the console
Section titled “Alerts in the console”Monitor › Alerts and events shows alerts Airgate raises itself, among them:
- Misconfigured Network Devices: a device sends with a wrong shared secret or from an unknown address; see Logs and debugging.
- A site connector that stops sending heartbeats (offline after 90 seconds by default), and again when it is back online.
- Failed device-source synchronisations (Intune, Jamf, CrowdStrike, SentinelOne), failing webhook and syslog targets, and HTTP enforcement actions that could not run.
Monitor › Overview shows sign-in volume, results and connected sessions; Monitor › Reports gives trends over time.
Send events and logs to your SIEM
Section titled “Send events and logs to your SIEM”- Settings › Log export sends events (sign-ins, sessions, alerts, audit records and more, chosen by topic) to syslog or SIEM targets in RFC 5424, CEF, LEEF or JSON format.
- Settings › Webhooks posts events to an HTTPS endpoint, signed so the receiver can check they come from Airgate.
See Webhooks and log export API for the formats.
Monitoring the appliance itself
Section titled “Monitoring the appliance itself”On the appliance:
| Check | Command |
|---|---|
| Service status and health | sudo airgate-compose ps |
| Service logs | sudo airgate-compose logs -f nacd |
| Disk space | df -h / |
| Last backup | Settings › Backups |
| Installed version | Settings › Software update, or sudo airgate-update status |
The VMware Tools (open-vm-tools) in the appliance report its state and IP address to vCenter, so you can also monitor it with vCenter alarms.