Key Takeaways
- Kubernetes orchestrates containers across many machines using a declarative API.
- The core objects are Pods, Deployments, Services and Ingress.
- The control plane runs the desired state and the nodes run the workloads.
- Kubernetes is powerful and expensive to run well — most small teams do not need it.
- Managed offerings hide the pain but not the concepts.
The Core Objects
| Object | What It Represents |
|---|---|
| Pod | One or more containers scheduled together |
| Deployment | A declaration of how many replicas of a Pod should exist |
| Service | A stable network endpoint for a group of Pods |
| Ingress | HTTP routing into the cluster |
| ConfigMap and Secret | Configuration and sensitive data injected into Pods |
| Namespace | Logical grouping and quota boundary |
Every non-trivial workload is expressed as a combination of these objects.
The Control Plane
The control plane is the brain — it accepts declarative manifests and continuously reconciles the actual cluster state to the desired state. It contains the API server, scheduler, controller manager and etcd data store. Nodes run the kubelet, which pulls work from the control plane and runs it on the local container runtime.
Real Workloads On Kubernetes
- Stateless web services behind an Ingress and a horizontal auto-scaler.
- Batch jobs and cron workloads via Job and CronJob.
- Stateful databases via StatefulSets — with caution.
- Data pipelines using Argo or Kubeflow.
When Kubernetes Is Worth It
- More than a few teams sharing infrastructure.
- Workloads with heterogeneous scaling profiles.
- A dedicated platform team that can operate it.
- Multi-region or multi-tenant requirements.
When It Is Not
- A small team running a few services.
- Traffic that a managed serverless platform can handle cheaply.
- A team without operational discipline.
- Any case where boring PaaS options would work.
Common Mistakes
- Adopting Kubernetes before your product needs it.
- Running your own control plane instead of using a managed one.
- Storing state in Pods without persistent volumes.
- Ignoring quotas and limits.
- Treating Helm as an actual configuration language.
Final Summary
Kubernetes is a serious platform. When your workload profile demands it, nothing else quite compares; when it does not, it is often a very expensive way to run a small application.
