Glossary

API server — the cluster's front door. Every read and change, from kubectl or the system itself, goes through it.

Cluster — a set of nodes plus a control plane, managed as one Kubernetes system.

ClusterIP — the default Service type: reachable only from inside the cluster.

ConfigMap — an object holding plain configuration (key/value pairs or files), delivered to pods as environment variables or mounted files.

Container runtime — the software that actually runs containers on a node; usually containerd.

Control plane — the cluster's brain: API server, etcd, scheduler, and controller manager.

Controller — a loop that reconciles one kind of object: compares desired with actual and closes the gap, forever.

CrashLoopBackOff — a container that starts, exits, and restarts repeatedly; the backoff is the delay between attempts.

Deployment — the object that keeps a set of identical pods at a desired count and rolls out template changes gradually.

Desired state — what your manifest says should exist. The cluster's job is making actual state match it.

Endpoints — the live pod addresses currently behind a Service. Empty endpoints, empty answers.

etcd — the control plane's database; the cluster's single source of truth.

Helm — the package manager for bundles of manifests ("charts") with fill-in values.

ImagePullBackOff — the kubelet cannot fetch the container image and is retrying with growing delays.

Ingress — an object describing HTTP routing rules (by host and path) into Services; needs an Ingress controller installed to do anything.

kind — "Kubernetes in Docker": local clusters whose nodes are containers.

kube-proxy — the per-node component that programs network rules so Service addresses reach pods.

kubectl — the command-line client for the API. (kube-control and kube-cuttle are both heard; nobody will correct you either way.)

kubelet — the agent on each node that starts assigned pods and reports their health.

Label — a key/value tag on an object. Selectors match labels to connect objects: Services to pods, Deployments to the pods they count.

LoadBalancer (Service type) — asks the cloud provider for a real external load balancer pointing at the Service.

Manifest — a YAML document describing one object: apiVersion, kind, metadata, spec.

Namespace — a named compartment inside a cluster. Names must be unique only within one; default is the default.

Node — one machine (physical or virtual) in the cluster. Worker nodes run pods; control-plane nodes run the brain.

NodePort — a Service type that also opens a port (30000–32767) on every node, forwarding to the Service.

PersistentVolume (PV) — a chunk of real storage in the cluster. A PersistentVolumeClaim (PVC) is a pod-side request that gets bound to one.

Pod — the smallest schedulable unit: one or more containers sharing a network identity, storage, and lifecycle. Almost always created by a Deployment, not by hand.

ReplicaSet — the counting mechanism under a Deployment: keeps N identical pods in existence.

Rollout — a Deployment's gradual replacement of old pods with new ones; kubectl rollout undo retreats to the previous revision.

Scheduler — the control-plane part that picks a node for each new pod.

Secret — a ConfigMap's sibling for sensitive values. Stored base64-encoded — an encoding, not encryption.

Selector — a label matching rule. The way Services find pods and Deployments count them.

Service — a stable virtual address and DNS name fronting a shifting set of pods, chosen by label selector.

StatefulSet — the controller for stateful workloads: stable pod names, stable storage, ordered starts. Named here so you know it exists.

StorageClass — the recipe a provisioner uses to create PersistentVolumes on demand.

Volume — storage attached to a pod, outliving its containers but not (except via PVs) the pod.