One view across Kubernetes clusters and accounts

01 — The challenge

Where it started.

Clusters sat in separate AWS accounts with no central place to watch them, so problems were found late and availability suffered for it.

Without visibility, the first notice of an incident is a customer.

02 — What we built

The work itself.

Prometheus across the clusters. Metrics collected from every cluster rather than from whichever one someone thought to check.

Alertmanager as one set of rules. Prometheus data centralised and alert rules held in one place, so a rule change applies everywhere.

Grafana on performance and resource. Visualisation built for the DevOps team to read during an incident, not a monthly report.

Sensitive environments fenced off. Access restricted per environment, so central visibility does not mean everyone can reach everything.

03 — Results

What the client has now.

Operational visibility across EKS clusters that were previously watched separately. Problems found and traced before they reached availability. Insight into system performance and user behaviour that was not previously collected.