BUILD, SHIP,
OPERATE

Cluster architecture, workload design, scheduling and placement, autoscaling that actually responds, and upgrades that do not become outages. The failures here are rarely mysterious once you know which controller to ask — a Pending pod, a stuck rollout, a removed API that takes manifests with it on the next minor version.

Prometheus, Grafana, log aggregation and distributed tracing, assembled so the signals answer questions rather than merely existing. Alongside that, the practice around them: service levels that mean something, alerts that fire on symptoms rather than causes, and incident handling that produces a blameless postmortem instead of a search for who to blame.

Delivery pipelines and the declarative reconciliation that makes them reviewable — ArgoCD and Flux, progressive delivery, security gates that sit in the pipeline rather than in a document nobody reads. The goal is that what is running can always be traced back to something committed.

The layer underneath all of the above. Performance and capacity work, systemd, kernel and sysctl tuning, namespaces and cgroups, and the kind of troubleshooting that starts with high load or a full disk and ends with a specific cause. Everything in a container is still a Linux process.