Ship small, ship often: Practical Kubernetes CI/CD on a budget (GitHub Actions + Helm)

Audience: freelancers and small teams who need reliable, inexpensive delivery to Kubernetes. This is a long, hands-on guide with lots of copy-pasteable code and extra explanations.

---

TL;DR

• Build Docker images in GitHub Actions, tag each image with the commit SHA, push to GitHub Container Registry (GHCR) using the built-in GITHUB_TOKEN, and deploy to Kubernetes with Helm. This keeps infra simple and costs low because your code, CI, registry, and permissions all live inside GitHub. (GitHub Docs)

• Rely on Kubernetes Deployments for rolling updates, and wire readiness + liveness probes so pods don’t get traffic until they’re ready (and get restarted if they hang). Make the pipeline wait with helm --wait --atomic and kubectl rollout status so a “green” job actually means the app is healthy. (Kubernetes, Helm)

• Three recurring mistakes I see: (1) shipping :latest, (2) not waiting for rollouts, (3) giving CI cluster-admin. Fixes: tag or digest-pin images, wait on the rollout, and scope deploy permissions with RBAC in the target namespace. (Kubernetes)

---

Table of Contents

1. Why this stack fits freelancers

2. The architecture in one picture

3. One-time setup & context you’ll need

4. A tiny sample app (Node.js) you can actually run

5. Dockerfile: multi-stage build, explained

6. Helm chart structure & values: what’s going on and why

7. CI: GitHub Actions workflow to build/tag/push (with caching)

8. CD: Helm upgrade, safe timeouts, and rollout checks

9. Health probes that prevent surprise downtime

10. Pitfalls I see weekly (and how to avoid them)

11. Minimal RBAC for CI (no more cluster-admin)

12. Private registries & imagePullSecrets (GHCR note)

13. Rollbacks, who does what, and a quick “am I healthy?”

14. A reusable repo layout & a secrets checklist

15. Optional: the same loop with Kustomize

16. Appendix: extra CI steps (tests, concurrency, smoke checks)

17. References

---

1) Why this stack fits freelancers

When you’re solo or working with a tiny team, the time you don’t spend gluing tools together is time you can bill. The combination below keeps blast radius small and the moving parts understandable:

• GitHub Actions + GHCR: you already store your code in GitHub; Actions runs your builds and GHCR keeps your images in the same trust boundary. With proper workflow permissions, the built-in GITHUB_TOKEN is enough to push images—no extra tokens to rotate, no additional accounts to manage. (GitHub Docs)

• Helm: clients expect Helm charts. Charts are just templates with a clear place for defaults (values.yaml), and a simple way to override them per environment. (Helm)

• Kubernetes Deployments: rolling updates are the default—Kubernetes gradually replaces old pods with new ones and only declares success after the new ReplicaSet becomes ready. That’s near-zero downtime when you wire probes and wait for the rollout. (Kubernetes)

---

2) The architecture in one picture

The last step matters: your pipeline blocks until the Deployment is actually healthy (or times out), which means green pipelines line up with a healthy app, not just “manifests applied”. (Kubernetes)

---

3) One-time setup & context you’ll need

• kubectl on the runner and on your machine. Use the official install so versions match your cluster’s supported skew (typically ±1 minor). (Kubernetes)

• Helm v3 on the runner. The official install page documents the script many people use in CI. (Helm)

• GHCR enabled in your org/repo. Add the OCI label org.opencontainers.image.source in your Dockerfile so GHCR auto-links the package to your repo; use GITHUB_TOKEN with packages: write scoped by workflow permissions. (GitHub Docs)

Why versions/skew matters: if your kubectl is too far ahead/behind, you’ll chase weird errors. The official docs call out the “within one minor” rule. (Kubernetes)

---

4) A tiny sample app (Node.js) you can actually run

Two endpoints that map directly to Kubernetes probes:

// app/server.js
import http from 'node:http';
const port = process.env.PORT || 3000;
let ready = false;

// Simulate warm-up work: DB connection, JIT, caches...
setTimeout(() => (ready = true), 3000);

const server = http.createServer((req, res) => {
  if (req.url === '/livez') return res.end('ok');                  // liveness probe
  if (req.url === '/healthz') return res.end(ready ? 'ok' : 'starting'); // readiness probe
  res.writeHead(200, { 'content-type': 'text/plain' });
  res.end('hello');
});

server.listen(port, () => console.log(`listening on ${port}`));

A minimal package.json so CI can run tests quickly:

{
  "name": "demo-app",
  "type": "module",
  "scripts": {
    "start": "node app/server.js",
    "test": "node -e \"const http=require('http');http.get('http://127.0.0.1:3000',r=>process.exit(r.statusCode===200?0:1))\""
  },
  "dependencies": {}
}

The point of /healthz and /livez: Kubernetes treats them differently—readiness gates traffic; liveness restarts stuck containers. We’ll wire both into the Deployment in a moment. (Kubernetes)

---

5) Dockerfile: multi-stage build, explained

Why multi-stage? Build tools (compilers, bundlers) aren’t needed at runtime. Splitting build and runtime stages makes smaller, faster, and safer images. The official Docker docs explicitly recommend multi-stage builds for production. (Docker Documentation)

# Dockerfile

# -------- build stage --------
FROM node:20 AS build
WORKDIR /app
COPY package*.json ./
RUN npm ci --omit=dev
COPY . .
# leave only prod deps
RUN npm prune --omit=dev

# -------- runtime stage --------
FROM node:20-slim
WORKDIR /app
COPY --from=build /app ./
EXPOSE 3000
CMD ["node","app/server.js"]

# helps GHCR link this image to your repo automatically
LABEL org.opencontainers.image.source="https://github.com/your-org/your-repo"

If you ever need multi-arch images (arm64 laptop, amd64 server), Buildx and QEMU make that possible with one workflow. We’ll stick to a single arch for speed. (Docker Documentation)

---

6) Helm chart structure & values: what’s going on and why

Helm charts are folders with a specific layout; defaults live in values.yaml, and templates live under templates/. You can override defaults per environment with --values path.yaml or inline --set key=value. (Helm)

chart/Chart.yaml

apiVersion: v2
name: app
description: Minimal demo app
type: application
version: 0.1.0         # chart version
appVersion: "0.1.0"    # your app version (for humans; not enforced)

chart/values.yaml (defaults you’ll override in CI)

image:
  repository: ghcr.io/your-org/your-repo/app
  tag: "CHANGE_ME"         # CI will set this to the commit SHA
  pullPolicy: IfNotPresent

service:
  type: ClusterIP
  port: 80
  targetPort: 3000

replicaCount: 2

resources:
  requests: { cpu: 100m, memory: 128Mi }
  limits:   { cpu: 500m, memory: 256Mi }

# optional if using private images
# imagePullSecrets:
#   - name: ghcr-pull

chart/templates/deployment.yaml (probes + rolling update)

apiVersion: apps/v1
kind: Deployment
metadata:
  name: {{ include "app.fullname" . }}
  labels:
    app.kubernetes.io/name: {{ include "app.name" . }}
spec:
  replicas: {{ .Values.replicaCount }}
  strategy: { type: RollingUpdate }
  selector:
    matchLabels:
      app.kubernetes.io/name: {{ include "app.name" . }}
  template:
    metadata:
      labels:
        app.kubernetes.io/name: {{ include "app.name" . }}
    spec:
      {{- if .Values.imagePullSecrets }}
      imagePullSecrets:
      {{- range .Values.imagePullSecrets }}
        - name: {{ .name }}
      {{- end }}
      {{- end }}
      containers:
        - name: app
          image: "{{ .Values.image.repository }}:{{ .Values.image.tag }}"
          imagePullPolicy: {{ .Values.image.pullPolicy }}
          ports:
            - containerPort: {{ .Values.service.targetPort }}
          readinessProbe:
            httpGet: { path: /healthz, port: {{ .Values.service.targetPort }} }
            initialDelaySeconds: 5
            periodSeconds: 5
          livenessProbe:
            httpGet: { path: /livez,  port: {{ .Values.service.targetPort }} }
            initialDelaySeconds: 15
            periodSeconds: 10

chart/templates/service.yaml

apiVersion: v1
kind: Service
metadata:
  name: {{ include "app.fullname" . }}
spec:
  type: {{ .Values.service.type }}
  selector:
    app.kubernetes.io/name: {{ include "app.name" . }}
  ports:
    - port: {{ .Values.service.port }}
      targetPort: {{ .Values.service.targetPort }}

If you prefer a more DRY approach, add a _helpers.tpl to standardize names/labels. Helm’s best-practices guide covers how to structure values and helpers cleanly. (Helm)

---

7) CI: GitHub Actions to build, tag, and push (with caching)

We’ll use Docker’s official Actions: setup-buildx, login, and build-push. We tag the image with the commit SHA so every deploy is traceable and immutable. Enable packages: write for GITHUB_TOKEN. (GitHub, GitHub Docs)

.github/workflows/ci.yml

name: ci
on:
  push:
    branches: [ "main" ]

permissions:
  contents: read
  packages: write   # allow pushing to GHCR (least privilege)

jobs:
  build-and-push:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4

      # optional: speed up local tests
      - uses: actions/setup-node@v4
        with:
          node-version: '20'
          cache: 'npm'
      - run: npm ci
      - run: npm test

      - name: Set up Docker Buildx
        uses: docker/setup-buildx-action@v3

      - name: Log in to GHCR
        uses: docker/login-action@v3
        with:
          registry: ghcr.io
          username: ${{ github.actor }}
          password: ${{ secrets.GITHUB_TOKEN }}

      - name: Build and push image
        uses: docker/build-push-action@v6
        with:
          context: .
          push: true
          tags: ghcr.io/${{ github.repository }}/app:${{ github.sha }}
          cache-from: type=gha
          cache-to: type=gha,mode=max

A few gotchas here:

• Workflow permissions must allow writing packages, or GHCR will reject pushes. You can set defaults in repo settings and still restrict per workflow/job. (GitHub Docs)

• Labeling images with org.opencontainers.image.source helps GHCR connect the package back to the repo UI. (GitHub Docs)

• Buildx is the recommended way to build images in Actions, with caching and multi-arch support documented in Docker’s CI guide. (Docker Documentation)

Why SHA tags instead of :latest? Kubernetes treats image references literally. Mutable :latest tags make it harder to know what’s running and can trip pull-policy defaults. Use an immutable tag (SHA) or even a digest (image@sha256:...) for perfect reproducibility—Kubernetes supports digests natively. (Kubernetes)

---

8) CD: Helm upgrade, safe timeouts, and rollout checks

Deploy right after a successful push. helm upgrade --install handles both first-time and repeat deploys. Add --wait (block until ready), --timeout (don’t wait forever), and --atomic (automatic rollback on failure). Then, as an extra safety net, run kubectl rollout status to stream progress into the job log. (Helm, Kubernetes)

  deploy:
    needs: build-and-push
    runs-on: ubuntu-latest
    environment: production
    steps:
      - uses: actions/checkout@v4

      - name: Install kubectl (official)
        run: |
          curl -LO "https://dl.k8s.io/release/$(curl -Ls https://dl.k8s.io/release/stable.txt)/bin/linux/amd64/kubectl"
          chmod +x kubectl && sudo mv kubectl /usr/local/bin/

      - name: Install Helm
        run: curl -fsSL https://raw.githubusercontent.com/helm/helm/main/scripts/get-helm-3 | bash

      - name: Configure kubeconfig
        env: { KUBECONFIG_DATA: ${{ secrets.KUBECONFIG_DATA }} }
        run: |
          mkdir -p $HOME/.kube
          echo "$KUBECONFIG_DATA" | base64 -d > $HOME/.kube/config

      - name: Helm upgrade (create or update)
        run: |
          helm upgrade --install app ./chart \
            --namespace prod --create-namespace \
            --set image.tag=${{ github.sha }} \
            --wait --timeout=10m --atomic

      - name: Rollout status (extra verification)
        run: kubectl rollout status deploy/app -n prod --timeout=180s

Why both --wait and kubectl rollout status? Helm’s --wait bubbles up a release-level success/fail; kubectl rollout status prints per-step progress and gives fast, readable feedback when something’s off. Helm’s --atomic turns failures into an automatic rollback—useful when you’d rather be safe than “half updated”. (Helm, Kubernetes)

---

9) Health probes that prevent surprise downtime

You can have a perfect rolling update and still serve errors if the app accepts traffic before it’s warmed up. That’s what readiness probes prevent: they gate Service endpoints until your pod says “ready.” Liveness probes are your safety rope if the app wedges itself; the kubelet restarts the container after repeated failures. The official docs cover these semantics and when to use startup probes for slower apps. (Kubernetes)

Practical tuning guidance:

• If your app needs 20s to warm up, set initialDelaySeconds on readiness accordingly.

• If your app occasionally blocks (e.g., deadlock), prefer an HTTP or TCP liveness probe that hits a trivial endpoint or port.

• If your app is slow to start, add a startup probe to delay liveness until boot completes. (Kubernetes)

---

10) Pitfalls I see weekly (and how to avoid them)

10.1 Shipping :latest

• What goes wrong: The same tag points to different bits over time; pull cache and policy quirks make it unclear whether a node actually pulled the new image.

• The fix: Tag images by commit SHA or use digests (@sha256:...). Kubernetes supports digest references; it will always fetch the exact image. The official image docs explain how Kubernetes resolves image references, tags, and digests. (Kubernetes)

10.2 Not waiting for rollouts

• What goes wrong: Your pipeline finishes “green,” but users see errors for 30–60 seconds because the Deployment hasn’t converged yet.

• The fix: Add --wait --timeout (and optionally --atomic) to Helm, and run kubectl rollout status so failures show up clearly in logs. (Helm, Kubernetes)

10.3 Over-privileged CI (cluster-admin)

• What goes wrong: A leaked token equals full cluster compromise.

• The fix: Create a ServiceAccount + Role + RoleBinding scoped to the target namespace, with only the verbs and resources your chart needs. The RBAC docs call this model out explicitly. (Kubernetes)

---

11) Minimal RBAC for CI (copy/paste)

# rbac/deployer.yaml
apiVersion: v1
kind: ServiceAccount
metadata:
  name: helm-deployer
  namespace: prod
---
apiVersion: rbac.authorization.k8s.io/v1
kind: Role
metadata:
  name: helm-deployer
  namespace: prod
rules:
  - apiGroups: ["apps"]
    resources: ["deployments","replicasets"]
    verbs: ["get","list","watch","create","update","patch"]
  - apiGroups: [""]
    resources: ["services","configmaps","secrets"]
    verbs: ["get","list","watch","create","update","patch"]
  - apiGroups: ["networking.k8s.io"]
    resources: ["ingresses"]
    verbs: ["get","list","watch","create","update","patch"]
---
apiVersion: rbac.authorization.k8s.io/v1
kind: RoleBinding
metadata:
  name: helm-deployer
  namespace: prod
subjects:
  - kind: ServiceAccount
    name: helm-deployer
    namespace: prod
roleRef:
  kind: Role
  name: helm-deployer
  apiGroup: rbac.authorization.k8s.io

This follows the least privilege principle: the SA can only touch the namespace and resource types your chart manages. If a token leaks, the damage is limited to those verbs/resources in prod. RBAC is the supported authorization model in Kubernetes. (Kubernetes)

---

12) Private registries & imagePullSecrets (GHCR)

• Public GHCR packages: anyone can pull anonymously; no secret required.

• Private GHCR packages: create a Docker registry secret and reference it in your pod spec (or attach it to your ServiceAccount). Kubernetes documents both the Secret type and the imagePullSecrets mechanism. (Kubernetes)

Create a pull secret and use it in prod:

kubectl create secret docker-registry ghcr-pull \
  --docker-server=ghcr.io \
  --docker-username=YOUR_GITHUB_USERNAME \
  --docker-password=YOUR_GHCR_TOKEN_OR_PAT \
  --docker-email=you@example.com \
  -n prod

Then in values.yaml for prod:

imagePullSecrets:
  - name: ghcr-pull

Official docs show the same flow for private registries; GHCR uses standard Docker auth. (Kubernetes)

---

13) Rollbacks, who does what, and a quick “am I healthy?”

Two layers can help you:

• Helm release history → helm history/rollback.

• Deployment history → kubectl rollout undo/status.

Typical commands:

# See release history managed by Helm
helm history app -n prod

# Roll back to the previous release
helm rollback app 1 -n prod

# Verify the Deployment converges
kubectl rollout status deploy/app -n prod --timeout=180s

Helm’s --atomic already auto-rolls back on failures during the upgrade. kubectl rollout status is still useful for transparent logs during manual ops or in CI. (Helm, Kubernetes)

---

14) A reusable repo layout & a secrets checklist

repo-root/
  app/                         # your code
  Dockerfile
  chart/
    Chart.yaml
    values.yaml
    values-staging.yaml
    values-prod.yaml
    templates/
      deployment.yaml
      service.yaml
      _helpers.tpl
  rbac/
    deployer.yaml
  .github/workflows/ci.yml

Secrets (GitHub → Settings → Secrets and variables → Actions):

• KUBECONFIG_DATA – base64 kubeconfig for the helm-deployer ServiceAccount.

• GITHUB_TOKEN – auto-injected; set workflow permissions: packages: write to push to GHCR (use repo/org defaults or YAML). (GitHub Docs)

---

15) Optional: the same loop with Kustomize

Prefer overlays without templating? Kustomize is built into kubectl. Your pipeline becomes: build → push → kubectl apply -k overlays/prod → kubectl rollout status. It’s great for single-service repos and small customizations; Helm tends to be handier when you need packagability or your client already uses charts. (Kubernetes)

Quick sketch:

kustomize/
  base/
    deployment.yaml
    service.yaml
    kustomization.yaml
  overlays/
    prod/
      kustomization.yaml   # patches image tag to $GIT_SHA

---

16) Appendix: extra CI steps (tests, concurrency, smoke checks)

A tiny smoke check after deploy (port-forward for 1 request):

      - name: Port-forward and smoke test
        run: |
          kubectl -n prod port-forward svc/app 8080:80 &
          PF_PID=$!
          sleep 3
          curl -fsS http://127.0.0.1:8080/healthz | grep -q "ok"
          kill $PF_PID

Digest pinning (even stricter than tags):

# Check the pushed image's digest
docker buildx imagetools inspect ghcr.io/${OWNER}/${REPO}/app:${GITHUB_SHA}

# Then set in values-prod.yaml:
# image:
#   repository: ghcr.io/${OWNER}/${REPO}/app
#   tag: ""   # leave empty when using a digest
#   digest: "sha256:..."

If you want to template digest usage into Helm, adjust your image stanza and template:

# values.yaml (allow either tag OR digest)
image:
  repository: ghcr.io/your-org/your-repo/app
  tag: "CHANGE_ME"
  digest: ""

# templates/deployment.yaml (image reference)
image: "{{ .Values.image.repository }}{{- if .Values.image.digest -}}@{{ .Values.image.digest }}{{- else -}}:{{ .Values.image.tag }}{{- end }}"

Kubernetes understands @sha256: digests directly. It’s the most deterministic way to roll out exactly what you built. (Kubernetes)

Multi-arch builds (if you need arm64): add platforms: linux/amd64,linux/arm64 to build-push-action and include QEMU if needed. Docker’s docs show both. (Docker Documentation)

---

17) References

• Kubernetes Deployments (rolling update behavior & concepts). (Kubernetes) Rolling updates overview (tutorial). (Kubernetes) kubectl rollout status (command reference). (Kubernetes) Probes (liveness, readiness, startup). (Kubernetes) Images: tags, digests, and imagePullSecrets. (Kubernetes) RBAC authorization (official). (Kubernetes) Install kubectl and version skew note. (Kubernetes)

• Helm helm upgrade and flags (--install, --wait, --atomic). (Helm) Installing Helm (CI script). (Helm) Chart structure & values reference and best practices. (Helm)

• GitHub Actions & GHCR Publishing Docker images with Actions. (GitHub Docs) Using GITHUB_TOKEN with least-privilege permissions. (GitHub Docs) Working with the Container registry (GHCR) and OCI labels. (GitHub Docs) Docker official GitHub Actions: setup-buildx, login, build-push. (GitHub) Docker’s CI guide for GitHub Actions. (Docker Documentation)

• Docker Multi-stage builds (why/how), best practices, multi-platform. (Docker Documentation)

---

Closing thoughts

This is the path I reach for when a client says: “we want reliable deploys this week, not a platform rebuild.” It scales from a single service to many, and every part of it is explainable to non-platform folks:

• One chart, one workflow, one registry.

• Immutable images (SHA tags or digests), probes that reflect real app health, and rollouts that wait.

• Scoped deploy permissions, not cluster-admin.

Stay Connected

If you enjoyed this article and want to explore more about web development, feel free to connect with me on various platforms:

dev.to

hackernoon.com

hashnode.com

twitter.com

instagram.com

personal portfolio v1

Your feedback and questions are always welcome. Keep learning, coding, and creating amazing web applications.