Metrics at a glance
| Metric | Why it matters | Page |
|---|---|---|
| Traffic served | Bandwidth out — drives bill and capacity planning. | Traffic served |
| Requests served | Volume — sizes infra and tracks adoption. | Requests served |
| Cache hit ratio | Efficiency — high HIT% = low origin cost and latency. | Cache hit ratio |
| Response classes | Health — non-2xx counts catch outages and bad deploys. | Response classes |
Healthy baselines
| Metric | Good | Investigate |
|---|---|---|
| Cache hit ratio (static) | ≥ 95% | < 90% |
| Cache hit ratio (mixed) | ≥ 80% | < 70% |
| 4xx rate | < 1% of total | sustained > 2% |
| 5xx rate | < 0.1% of total | any sustained spike |
| Traffic week-over-week | within ±20% | sudden 2× without launch |
Pull metrics via API
Wire into your stack
| Destination | How |
|---|---|
| Grafana | Use the Tenbyte API as a JSON datasource; chart bandwidth_bytes, requests, cache_hit_ratio. |
| Datadog | Cron a small lambda that pulls the API and pushes custom metrics. |
| Slack alerts | Configure webhook on response-class spike via the distribution settings. |
| Spreadsheet | Pipe daily totals to a CSV for finance / capacity reporting. |
Operational tips
- Compare deploys, not days. A weekday/weekend baseline beats raw deltas.
- Alert on ratios, not absolutes. Traffic doubles, errors stay flat — that’s healthy growth, not an incident.
- Drill in by path. Filter by URL prefix to find the noisy 4xx contributor before it becomes a 5xx.
- Pair with origin metrics. Edge
5xxplus origin5xx= origin issue; edge5xxonly = networking or config.