Kubernetes Best Practices for Startups

Start Simple

Startups don’t need a multi-cluster service mesh on day one. Start with a single managed Kubernetes cluster (EKS, GKE, AKS) and grow incrementally.

Essential Components

Resource Limits

Always set CPU and memory limits. Without them, one bad deployment can take down your entire cluster:

resources:
  requests:
    memory: "256Mi"
    cpu: "250m"
  limits:
    memory: "512Mi"
    cpu: "500m"

Health Checks

Configure liveness and readiness probes for every pod:

livenessProbe:
  httpGet:
    path: /healthz
    port: 8080
  initialDelaySeconds: 5
  periodSeconds: 10
readinessProbe:
  httpGet:
    path: /ready
    port: 8080
  initialDelaySeconds: 3
  periodSeconds: 5

Namespace Isolation

Use namespaces to separate environments and teams.

Cost Optimization

• Use cluster autoscaler with spot instances for worker nodes
• Right-size your requests based on actual usage
• Use Karpenter for dynamic node provisioning
• Enable HPA (Horizontal Pod Autoscaler) for all stateless services

Monitoring Stack

A minimal but effective stack:

• Metrics: Prometheus + Grafana
• Logging: Loki + Promtail
• Alerting: Alertmanager
• Tracing: OpenTelemetry (start with a sampling rate of 10%)

Conclusion

Kubernetes can be simple. Start with the essentials, add complexity only when you need it.