Skip to main content

Observability configuration

Verdictan gives a Prometheus-compatible gateway metrics endpoint. Use /metrics to get the metric families from the gateway.

For request policy results and rejection causes, use verdictan events. You can also use GET /v1/events with the necessary since parameter.

Use History for saved session data when you enable capture. Use Trail for the control plane audit record.

Supported surface​

SurfaceUse
GET /metricsThe gateway listener gives metrics in Prometheus format.
GET /verdictan/providers/metricsProvider metrics help you examine routes and upstream operation.
verdictan events tail or GET /v1/events with necessary sinceThese surfaces give runtime request evidence and policy results.
History, when you enable captureThis surface keeps request or session data.
TrailThis surface shows control plane changes and audit evidence.

Basic check​

Start the gateway. Then, get metrics from the same listener:

verdictan gateway run --agent docs-observability --listen 127.0.0.1:41002 --policy-config policy-config.yaml
curl http://localhost:41002/metrics

Get provider metrics from a different endpoint:

curl http://localhost:41002/verdictan/providers/metrics

Dashboard guidance​

At this time, the default /metrics registry includes these metric families:

  • verdictan_gateway_requests_total, labeled by HTTP method, status bucket, and provider
  • verdictan_gateway_policy_eval_duration_seconds, an unlabeled histogram for policy chain latency
  • verdictan_gateway_upstream_errors_total, labeled by provider and error type
  • cache hit and miss counters, the dropped event count, and active connections
  • control plane counters for usage authorization, relay counters, and relay latency

GET /verdictan/providers/metrics gives a JSON snapshot for each provider. The snapshot includes the sample count and the latest health sample.

The snapshot also includes percentiles for time to first token (TTFT) and throughput. Examine Prometheus collection health with the standard up series.

At this time, the gateway does not emit route-labeled request latency, policy verdict counts, or fallback counts. It does not emit dedicated rejection metrics for size limits or rate limits.

Use status-labeled request counts to find total changes. Use verdictan events or the Events API for request results.

Do not make dashboards that use metric names or labels which do not occur in live metric collection.

Configuration boundaries​

Unless a different product page supports a field, do not add callback or health-monitor blocks to policy-config.yaml. Use the listed metrics and evidence surfaces for dashboards and alerts.

Next steps​