Kubernetes

Kubernetes Best Practices

Deployment Strategies

• Use Deployment with RollingUpdate as default: set maxSurge and maxUnavailable for controlled rollouts
• Implement readinessProbe and livenessProbe on every container -- HTTP or exec-based
• Use startupProbe for slow-starting applications to prevent premature kill by the liveness probe
• Set resource requests and limits on every pod: CPU/Memory requests for scheduling, limits for QoS
• Apply PodDisruptionBudget to maintain availability during voluntary disruptions (node drains, upgrades)

Service Mesh

• Use a service mesh (Istio, Linkerd) for mTLS, traffic shaping, and observability when microservices exceed ~10
• Define DestinationRule for circuit breaking and connection pooling
• Use VirtualService for canary deployments: route percentages between service versions
• Apply PeerAuthentication for strict mTLS mode across the mesh
• Keep mesh configuration in the same repository as the application manifests

Helm Charts

• Structure charts with templates/, values.yaml, Chart.yaml, and README.md
• Use values.yaml for environment-agnostic defaults; override with -f staging.yaml per environment
• Template with {{ .Values.* }} and {{ .Release.* }}; use {{- to trim whitespace
• Use named templates (_helpers.tpl) for reusable label blocks and common metadata
• Run helm lint and helm template in CI; use helm test for release validation hooks

RBAC and Security

• Apply least-privilege RBAC: create Role/RoleBinding per namespace, avoid ClusterRole unless necessary
• Use NetworkPolicy to restrict pod-to-pod traffic: default deny, explicitly allow required flows
• Run containers as non-root: set runAsNonRoot: true, readOnlyRootFilesystem: true in security contexts
• Use SealedSecrets or external secret managers (Vault, AWS Secrets Manager) -- never plaintext secrets
• Scan images with Trivy or Grype in CI; enforce admission controllers to block vulnerable images

Autoscaling and GitOps

• Configure HorizontalPodAutoscaler with CPU/memory and custom metrics via Prometheus adapter
• Use VerticalPodAutoscaler for right-sizing requests in steady-state workloads
• Implement GitOps with ArgoCD or Flux: declarative repo -> automated cluster reconciliation
• Structure repos as clusters/, apps/, infra/ for GitOps clarity; use app-of-apps pattern in ArgoCD
• Use Kustomize overlays for environment-specific patches without duplicating base manifests

Source: calcosmic/Aether — distributed by TomeVault.