Skip to content

GRAFANA.md

Latest commit

 

History

History
23 lines (14 loc) · 1.14 KB

GRAFANA.md

File metadata and controls

23 lines (14 loc) · 1.14 KB
Raw

Grafana dashboard

Metrics

quorabust-serve exposes Prometheus text at /metrics. Relevant series:

  • quorabust_predictions_total{variant="a"|"b"} — Counter incremented per successful POST /predict.
  • quorabust_predict_latency_seconds_*{variant=...} — Histogram of handler latency (seconds).

Scrape the service with Prometheus (job pointing at http://<host>:8000/metrics), then attach Grafana to that Prometheus data source.

Import

  1. Grafana → Dashboards → New → Import.
  2. Upload grafana/dashboards/quorabust-serving.json (or paste its contents).
  3. When prompted, pick your Prometheus data source for DS_PROMETHEUS.

Panels cover request rate by variant, p50/p95 latency, total predictions, and a single-stat p95 for SLO spot-checks. Adjust time range and thresholds to match your environment.

Notes

  • Histogram buckets are the Prometheus client defaults; for stricter SLOs, configure custom buckets in code and re-export the dashboard queries if names change.
  • If only variant a is deployed, graphs still work; variant b series appear when QUORABUST_MODEL_B is configured.