P95 latency alerts on Railway
Railway shows P95 latency, with P50, P90 and P99, on each service’s Metrics tab, but its monitors only fire on CPU, RAM, disk and network egress. To be alerted when a Railway service gets slow, you need a tool that reads those response times and alerts on P95. For what each percentile means, see P50 vs P95 vs P99 latency explained.
Where Railway shows latency
- Open the project, then the service.
- Go to the Metrics tab.
- The HTTP charts show response time as P50, P90, P95 and P99, the request error rate, and responses by status class (2xx, 4xx, 5xx).
The Railway CLI prints the same numbers, and can narrow them to one method or path:
bash
railway metrics --http
railway metrics --http --method GET --path /api/ordersWhat Railway monitors can and can’t watch
| Monitor on | Railway |
|---|---|
| CPU, RAM, disk, network egress | Yes, above or below a limit, on the Pro plan |
| P95 or P99 response time | No |
| HTTP error rate | No |
Monitors notify by email and in-app, and can call a webhook.
Which percentile to alert on
Alert on P95, held for several minutes, and pair it with the Railway 5xx error rate. P99 on a quiet Railway service is often a single cold or slow request.