Kubernetes Is Not a Container Platform (And That Changes Everything)

Kubernetes Is Not a Container Platform (And That Changes Everything)

Kubernetes was initially conceived as an extensible, declarative API powered by control loops, not a container orchestrator; this fundamentally alters how developers interact with the system. Understanding this core design impacts everything from application deployment to scalability and observability.

Why This Matters

Many developers approach Kubernetes by focusing on Pods, Deployments, and Helm charts, treating it as a direct container management system. This is a backwards approach that obscures the underlying architecture and limits the ability to leverage Kubernetes’ full potential. Misunderstanding this can lead to fragile deployments, complex YAML configurations, and difficulty scaling beyond basic use cases – costing engineering time and potentially impacting application availability.

Key Insights

• Kubernetes API as Foundation: Kubernetes’ core is its API, allowing for extensibility through Custom Resource Definitions (CRDs).
• CRDs Define New Types: CRDs let users define new resource types understood by Kubernetes, like kind: Backup.
• Operators Provide Control: Operators are controllers that act on CRDs, translating desired state into concrete actions such as creating Jobs or S3 uploads.

Working Example

kind: Backup
apiVersion: example.com/v1alpha1
metadata:
  name: my-backup
spec:
  schedule: "0 0 * * *"
  storageLocation: "s3://my-backup-bucket"

Practical Applications

• Database Operators: Operators like the Percona Operator manage complex database deployments, automating tasks like backups, scaling, and upgrades.
• Pitfall: Treating Helm charts as the core deployment mechanism instead of understanding the underlying controllers can lead to inflexibility and difficulty troubleshooting issues.

References:

• https://dev.to/mxnuchim/kubernetes-is-not-a-container-platform-and-that-changes-everything-4bpa