Kubernetes Core Concepts
Introduction
Kubernetes (K8s) is a container orchestration platform. It takes your containers and
runs them across a cluster of machines, handling scaling, networking, storage, and
failure recovery automatically.
This section focuses on understanding the concepts before touching a real cluster.
Kubernetes has a lot of terminology, and trying to learn by doing without understanding
the model leads to frustration.
Why This Matters
Kubernetes is the dominant platform for running production ML systems:
- Model serving at scale (auto-scaling based on traffic)
- Training job orchestration (Kubeflow, Argo Workflows)
- MLflow, Airflow, and other tools deploy on Kubernetes
- GPU scheduling for training workloads
More importantly, understanding Kubernetes teaches you distributed systems thinking:
- Declarative configuration (desired state vs imperative commands)
- Self-healing systems
- Service discovery and networking
- Resource management and scheduling
This mental model applies even when using managed services that hide Kubernetes.