Skip to main content

Grafana Monitoring

Grafana can visualize KoreShield metrics by connecting to Prometheus and building dashboards around request volume, block rate, and provider latency.

Use Cases​

  • Operator dashboards for LLM proxy health
  • Executive reporting on safety outcomes
  • Latency analysis per provider and model

Prerequisites​

  • Prometheus scraping KoreShield metrics at /metrics
  • Grafana connected to your Prometheus data source

Add Prometheus Data Source​

In Grafana:

  1. Configuration -> Data sources
  2. Add data source -> Prometheus
  3. Set the URL to your Prometheus server
  4. Save and test

Suggested Panels​

Create panels using the following PromQL queries:

  • Requests per minute:
rate(koreshield_requests_total[1m])
  • Blocked requests per minute:
rate(koreshield_requests_blocked_total[1m])
  • Attack detections per minute:
rate(koreshield_attacks_detected_total[1m])
  • P95 request latency:
histogram_quantile(0.95, sum(rate(koreshield_requests_duration_seconds_bucket[5m])) by (le))
  • Provider latency by provider:
histogram_quantile(0.95, sum(rate(koreshield_provider_latency_seconds_bucket[5m])) by (le, provider))

Alert Ideas​

  • High block rate
  • Sudden spike in attacks detected
  • Provider latency above threshold

Troubleshooting​

  • No data: confirm Prometheus scrape targets are healthy
  • Wrong labels: review scrape config and relabeling rules
  • Slow panels: increase Prometheus resources or narrow query ranges

Next Steps​