Connect bounded RED metrics and SLOs to one question

Observability answers incident questions; it is not a contest to store the most data. Collect Rate, Errors, and Duration by route template, then connect logs and asynchronous work with request context.

Never use the raw 404 path as a metric label: bots can create unbounded time series. Python backend folds unmatched routes into /_not_found. User IDs, filenames, and search terms must not become labels either.

Collection path and cost

Java services expose Actuator Prometheus endpoints and Python services expose standard /metrics. Compose Prometheus scrapes them only in the optional profile. Grafana and a Next.js-specific exporter remain optional until runtimes and deployment platforms share an operations contract; Next.js can use logs, Sentry, and Web Vitals.

When to promise an SLO

Define an SLO from a user path and baseline. If there is no measurement, do not promise an arbitrary 99.9%; agree on error-budget, deployment-pause, and rollback actions after collecting the baseline.

Label budget

Rate ─┐
Error ├─▶ user-path baseline ─▶ SLO candidate ─▶ error budget ─▶ release decision
Duration┘

Time-series count grows as the product of label value counts. Routes 50 × statuses 5 × methods 4 can be bounded; adding request IDs makes storage and query cost grow with request volume.

Related course: RED, correlation, and SLOs