Running containers by hand on separate machines means configuring and maintaining each host individually. A cluster turns those machines into a shared pool of computing resources. Kubernetes can place containers on that pool, so an application is not tied to one specially configured machine. If a machine fails, another can take over its work. section 1.2
In Kubernetes, the machines are called nodes. They are usually virtual machines, though they can be bare-metal machines. Together, the nodes and the control plane form the cluster. The control plane stores cluster state and coordinates which containers run where; each node runs a kubelet and container runtime to manage its containers. On a managed cloud service, the control plane may be hosted for you, leaving worker nodes as the machines you normally see. section 3.1.1
Compare the hosted control plane with the worker nodes connected to it. This is the useful mental picture when a cloud provider manages the control plane. section 3.1.1
You interact with a cluster through the Kubernetes API, commonly using kubectl. A Deployment manages Pods, which define the containers to run; a Service provides a network endpoint. You declare the result you want in configuration, and Kubernetes works to maintain it. For example, you can request that a service run across multiple availability zones rather than assigning its containers to individual machines yourself. section 3.1 section 1.2 section 3.5
Choose a cluster environment for the job:
- A cloud cluster is suited to publishing a service with a public IP and can provide resources across multiple machines. section 3.2.1 section 3.4
- A local cluster is useful for quick development, learning, and testing deployment configuration, but its fixed resources and lack of a public load balancer change how services are accessed. section 3.2.1 section 3.4
CLUSTER_NAME=my-cluster REGION=us-west1 gcloud container clusters create-auto $CLUSTER_NAME --region $REGION gcloud container clusters get-credentials $CLUSTER_NAME --region $REGION
This GKE Autopilot command sequence creates a cloud cluster and retrieves credentials so kubectl can connect to it. It is one concrete setup path, not a requirement to use GKE. section 3.2.1
