Step 11
Step 11 — Follow one request with RED, correlation, and SLOs
0 views
Step 11 — Follow one request with RED, correlation, and SLOs
More logs do not automatically make incidents easier. Start from a user question and connect request rate, latency, errors, and the same request's context.
- Rate: requests per route template
- Errors: status classes and business failures
- Duration: route-template latency distribution
- Context:
X-Request-Idthrough HTTP response, logs, and Kafka headers
Never use raw bot URLs or user IDs as metric labels. Python backend folds 404s into a fixed /_not_found route, while Java and Python expose RED endpoints for Prometheus. Logs, Sentry, and Web Vitals remain distinct signals when Prometheus is not running.
Authentication failures on the internal revalidation webhook record only a bounded reason, never the secret, payload, or IP. Observability should help find an incident without becoming a new store for private data or credentials.
An SLO such as “99% of public reads succeed within one second” should drive error-budget and rollback decisions. If there is no baseline, measure first instead of presenting an arbitrary number as a product promise.