operating-kubernetes_skill

Overview

This skill helps operate production Kubernetes clusters with an operations-first approach to resource management, advanced scheduling, networking, storage, security hardening, and autoscaling. It provides practical patterns, decision matrices, and troubleshooting playbooks to deploy and run reliable workloads. Use it to configure quotas, enforce security policies, tune autoscaling, and diagnose common failures quickly.

How this skill works

The skill inspects workload resource configurations, scheduling constraints, network policies, storage classes, RBAC and pod security settings, and autoscaler configurations to recommend concrete changes. It maps workload types to QoS classes, suggests affinity/taint patterns, evaluates NetworkPolicy defaults, and proposes appropriate HPA/VPA/KEDA strategies. It also provides step-by-step troubleshooting actions for Pending, CrashLoopBackOff, ImagePullBackOff, and service connectivity issues.

When to use it

• Deploying applications to Kubernetes and defining requests/limits
• Setting up autoscaling for web apps, queue processors, or databases
• Implementing NetworkPolicies and Gateway API for zero-trust networking
• Configuring StorageClasses, PVCs, and CSI drivers for stateful apps
• Hardening clusters with RBAC, Pod Security Standards, and policy engines
• Troubleshooting pods stuck in Pending, CrashLoopBackOff, or networking failures

Best practices

• Always set CPU and memory requests and limits; use VPA for rightsizing
• Enforce namespace ResourceQuotas and LimitRanges for multi-tenancy
• Apply topologySpreadConstraints and taints/tolerations for availability
• Use default-deny NetworkPolicies and prefer Gateway API for new apps
• Define StorageClasses per performance tier and enable snapshots
• Enforce Pod Security Standards (restricted) and RBAC least privilege

Example use cases

• Right-size a high-traffic web service with HPA and VPA to control cost
• Lock down a production namespace with default-deny NetworkPolicies and RBAC
• Configure GPU nodes with taints and pod tolerations for ML workloads
• Set up a fast SSD StorageClass for a database and migrate PVCs
• Create a KEDA ScaledObject to autoscale workers from a message queue