Helm

Every lab so far has had you hand-write and kubectl apply plain YAML. That's fine for one Deployment and a Service — it gets unwieldy fast once a real app is a dozen objects that all need to change together for a release, or vary slightly between dev/staging/prod. Helm is Kubernetes's most widely used answer to that problem.

Charts: templates plus values

A chart is a directory with a Chart.yaml (name/version metadata), a values.yaml (the default settings), and a templates/ folder of YAML files with Go-template placeholders like {{ .Values.image.tag }}. Helm's whole job is substitution: take the templates, plug in values, produce plain Kubernetes YAML.

From chart to cluster

Chart.yaml + templates/

Go-template YAML, with {{ .Values.x }} placeholders

values.yaml

the substitutions — a missing key just renders empty

helm template

renders plain YAML — nothing touches the cluster yet

helm install / upgrade

applies the rendered YAML, tracked as a release

Nothing here is magic — helm template <chart> does exactly that substitution and prints the result, with zero cluster contact. That command is your best debugging tool: if something's wrong after a helm install, render it first and read the YAML Kubernetes actually received, before you even look at kubectl describe.

Releases: install, upgrade, rollback

helm install <release-name> <chart> renders and applies a chart for the first time, giving it a release name Helm tracks. helm upgrade re-renders against new values and applies the diff — this is the one you'll use far more often, the same way kubectl apply is your everyday verb and create is a one-time thing.

Helm keeps a revision history per release, and helm rollback <release> <revision> reverts to an older one — directly analogous to kubectl rollout undo from Module 3, just operating on a whole bundle of objects instead of one Deployment's pod template.

What Helm doesn't save you from

Templates are just text substitution — Helm doesn't know or care whether .Values.image.tag actually exists in your values.yaml. Reference a key that isn't there and you don't get an error; you get an empty string silently spliced into the output. That's a real, common failure mode, and it's exactly what you'll hit in this lesson's lab.

The infrastructure analogy

A chart is closest to a parameterized CloudFormation/Terraform module — same idea of "a template plus a set of inputs, versioned and releasable together" — except Helm's unit of "release" is a live, continuously reconciled set of Kubernetes objects rather than a one-shot provisioning run.