Containers, Kubernetes & Platform Engineering

A pending pod has not been scheduled, often because of insufficient resources, affinity rules, taints, storage claims, or quota. Checks to make Inspect pod events and scheduler messages first. Compare requested CPU and memory with node capacity, check node selectors and tolerations, inspect PVC binding, and review namespace quotas. Change the specific...

A readiness probe can fail because path, port, timing, or dependency assumptions do not match application behavior. Checks to make Inspect pod events, test endpoint inside the container, compare port and path, and allow startup time. Keep readiness distinct from liveness. When to get help Container operations can tune health checks and rollout...

A Kubernetes rolling update can interrupt service if new pods take too long to become ready, old pods stop too early, or the cluster lacks room for temporary surge capacity. A deployment may be technically successful while users experience errors during the transition. Review rollout conditions Check replica count, readiness probes, max unavailable,...

A readiness probe that is too strict can mark a usable pod unready and reduce service capacity. If it checks a slow dependency or runs before application warm-up completes, healthy instances may be removed from traffic during routine load or deployment. Compare probe results with application health Review probe path, timeout, period, failure threshold,...

A Kubernetes pod that remains Pending has not yet been scheduled to a node. The application image may be valid; the scheduler may instead be unable to satisfy resource requests, node selectors, taints, affinity rules, or persistent storage requirements. Read the scheduling events Use kubectl describe pod and inspect the latest events. Check whether...

A Kubernetes container may be terminated with an out-of-memory event when its memory limit is below its real working set. Setting every limit very high is not a safe fix because a busy node can run out of memory and affect multiple workloads. Requests and limits should be based on measured behavior. Measure under representative load Review peak and...

A Kubernetes pod in a restart loop may be failing during startup, exceeding a memory limit, or being restarted by an overly aggressive health probe. Increasing resources without evidence can mask the real issue or move the problem to another node. Start with the pod's previous container state and recent events. Inspect the failure Use kubectl describe...

A service with no endpoints commonly has selector mismatch, unready pods, or a namespace and label assumption that differs from deployment. Checks to make Inspect service selectors and pod labels, check readiness and endpoint objects, verify namespace and port names, then compare the intended release labels. Fix the mismatch through the deployment...