Running Kubernetes in production without proper observability is flying blind. Interviewers use monitoring questions to determine whether candidates can operate and troubleshoot clusters effectively, not just deploy to them.
Foundational questions cover the Metrics Server and its role in supporting kubectl top and the HPA. Candidates should know that the Metrics Server only stores current values (not historical data) and is not a substitute for a full monitoring stack. Follow-up questions typically ask about Prometheus: how it discovers targets using service discovery, the pull-based scraping model, and PromQL basics for writing useful queries.
Logging questions assess whether candidates understand the container logging model in Kubernetes: stdout/stderr streams, log rotation by the container runtime, and patterns for aggregating logs across a cluster. The choice between a DaemonSet-based log collector (Fluentd on every node) versus sidecar containers for application-specific logging is a common discussion point.
Advanced interviews may explore the relationship between monitoring and autoscaling (custom metrics driving HPA), distributed tracing with Jaeger or OpenTelemetry, and how to design SLOs and alerting rules that reduce alert fatigue while catching real incidents. Candidates who can connect observability tools to operational outcomes — faster incident response, data-driven capacity planning — demonstrate the kind of thinking that production teams need.
Prometheus monitors Kubernetes by scraping metrics endpoints from Pods, nodes, and cluster components. It uses Kubernetes service discovery to automatically find targets. The kube-prometheus-stack (Prometheus Operator) is the standard deployment method, providing pre-built dashboards and alerting rules.