Advanced Integrated Monitoring with Prometheus Federation
summary
Concrete setup for Prometheus federation: a VM running a primary Prometheus binary scrapes two Linkerd clusters’ Linkerd Prometheus endpoints over a load balancer at a 15s interval while each Linkerd sidecar retains ~6 hours of metrics. Federation aggregates multiple small Prometheuses into one long‑term store, queried with PromQL; demo shows p95 latency and success‑rate differences between clusters. Storage math uses 1–2 bytes per sample (15 days ≈ 259 MB at 1 byte), one user once allocated ~1 TB RAM, Prometheus 2.13 improves memory, logs reveal some out‑of‑order samples and 404s, Alertmanager and Pushgateway are used, Cortex is suggested as an alternative.
AI-written summary · about