Demo Installing Kubeflow

This walkthrough demonstrates installing Kubeflow onto a Kubernetes cluster. It follows a pragmatic, repeatable set of steps so you can reproduce the install locally (using kind) or on a managed cluster (EKS, AKS, GKE). High-level steps

Prepare a Kubernetes cluster (local or cloud).
Install required CLI tools (kind, kustomize / kubectl).
Download the Kubeflow manifests and make any small customizations.
Build and apply the combined manifests to the cluster (kustomize + kubectl).
Verify the installation and wait for key components to be Ready.

Prerequisites

Requirement	Purpose / Example
Kubernetes cluster	Local: `kind` (recommended for this demo). Cloud: EKS/AKS/GKE also supported.
kind CLI	Create local clusters: `go install sigs.k8s.io/kind@v0.31.0` or via package manager.
kustomize or `kubectl kustomize`	Build manifests. `kubectl` includes kustomize support: `kubectl kustomize`.
kubectl	Cluster access and apply resources. Ensure `kubectl` context points to your cluster.
Docker	For kind node images and pulling images (login for private registries).

If you plan to pull images from private registries, ensure Docker is logged in on the machine where you run kubectl.

Install kind and create a cluster (example)

Install kind (example):

go install sigs.k8s.io/kind@v0.31.0

Create a cluster (simple example):

time kind create cluster

Example output (truncated):

Creating cluster "kind" ...
  Ensuring node image (kindest/node:v1.16.3) 🖼️
  Preparing nodes 📦
  Writing configuration 📜
  Starting control-plane 🧭
  Installing CNI ⚙️
  Installing StorageClass 💾
Set kubectl context to "kind-kind"
You can now use your cluster with:

kubectl cluster-info --context kind-kind

Use the Kubeflow-provided kind config

Kubeflow manifests provide an example kind config that sets kubeadm API server args required by some Kubeflow components. Save the snippet below as kind.config.yaml and use it when creating your cluster.

kind: Cluster
apiVersion: kind.x-k8s.io/v1alpha4
nodes:
- role: control-plane
  image: kindest/node:v1.34.0@sha256:7416a61b42b1662ca6ca89f02028ac133a309a2a30ba309614e8ec94d
kubeadmConfigPatches:
- |
  kind: ClusterConfiguration
  apiServer:
    extraArgs:
      "service-account-issuer": "https://kubernetes.default.svc"
      "service-account-signing-key-file": "/etc/kubernetes/pki/sa.key"

For a multi-node cluster, add additional worker nodes:

nodes:
- role: control-plane
  ...
- role: worker
- role: worker

Create the cluster using that config:

kind create cluster --name=kubeflow --config=kind.config.yaml

Verify nodes are Ready:

kubectl get nodes

Kubeflow manifests repository

Choose either a stable release or the master branch. This demo uses the repository master branch to fetch the latest manifests.

Clone the manifests repository and change into its directory:

git clone https://github.com/kubeflow/manifests
cd manifests

The repository README explains the install flow and contains example kind configs and Kustomize overlays.

A dark-themed screenshot of a GitHub README page for the Kubeflow Manifests repository, showing the "Overview of the Kubeflow Platform" section with bullet points and a small directory-purpose table. The browser window and tabs are visible across the top.

Browsing the example kustomization

The example folder is the canonical entry point used by the official install. It pulls in many common components (cert-manager, Istio, oauth2-proxy, Dex, Knative, etc.) and application overlays. Open example/kustomization.yaml to inspect top-level resources and ordering options:

apiVersion: kustomize.config.k8s.io/v1beta1
kind: Kustomization

sortOptions:
  order: legacy
  legacySortOptions:
    orderFirst:
      - Namespace
      - ResourceQuota
      - StorageClass
      - CustomResourceDefinition
      - MutatingWebhookConfiguration
      - ServiceAccount
      - PodSecurityPolicy
      - NetworkPolicy

The example kustomization references a resources: list that includes cert-manager, Istio, OAuth2-proxy overlays, Dex, and many other components. Example snippet:

resources:
# Cert-Manager
- ../common/cert-manager/base
- ../common/cert-manager/kubeflow-issuer/base
# Istio
- ../common/istio/istio-crds/base
- ../common/istio/istio-namespace/base
- ../common/istio/istio-install/overlays/oauth2-proxy
# OAuth2-proxy overlays (choose one depending on your cluster)
- ../common/oauth2-proxy/overlays/m2m-dex-only
# Dex
- ../common/dex/overlays/oauth2-proxy
# ...

You can enable or disable components by editing the example kustomization. For most demos and tests, leaving the defaults is fine.

Create a default user profile (optional)

To create a Profile resource (which creates a user namespace and default resources), edit the profile-instance manifest under common/user-namespace and set a metadata.name and spec.owner.name. Example:

apiVersion: kubeflow.org/v1beta1
kind: Profile
metadata:
  name: kodekloud
spec:
  owner:
    kind: User
    name: sanjeev

Adjust these values to match the username and naming you want. When applied, Kubeflow will create the Profile and corresponding namespace.

Prepare kubeconfig and registry credentials

If you created the cluster with kind and saved kubeconfig to a non-default path, export it:

kind get kubeconfig --name kubeflow > /tmp/kubeflow-config
export KUBECONFIG=/tmp/kubeflow-config

If your cluster needs to pull images from a private registry, create a Docker registry secret (here named regcred) using your ~/.docker/config.json:

docker login    # ensure docker is logged in

kubectl create secret generic regcred \
  --from-file=.dockerconfigjson=$HOME/.docker/config.json \
  --type=kubernetes.io/dockerconfigjson

Install Kubeflow using kustomize + kubectl

The recommended pattern is to build the example kustomization and apply it to the cluster. Because CRDs, webhooks, and controllers can come up in different orders, it’s common to retry the apply until success. Example install loop (standalone kustomize):

while ! kustomize build example | kubectl apply --server-side --force-conflicts -f -; do
  echo "Retrying kustomize build/apply in 10s..."
  sleep 10
done

If you prefer kubectl’s built-in kustomize:

while ! kubectl kustomize example | kubectl apply --server-side --force-conflicts -f -; do
  echo "Retrying kubectl kustomize build/apply in 10s..."
  sleep 10
done

Notes

--server-side --force-conflicts uses server-side apply to reduce client-side merge conflicts and improve behavior when resources already exist.
The loop tolerates transient ordering errors during the install and retries until all resources apply successfully.

Wait for key components

Some components take time to become Ready. Example: wait for cert-manager pods to become Ready:

kubectl wait --for=condition=Ready pods --all -n cert-manager --timeout=180s

Monitor overall progress:

kubectl get pods --all-namespaces
kubectl get deployments -A

You may also watch specific namespaces for pods to transition to Ready.

Example of applied resources

When the build/apply loop finishes successfully you’ll see many server-side applied resources (CRDs, validating webhooks, and application CRs). Example (truncated):

clusterservingruntime.serving.kserve.io/kserve-torchserve serverside-applied
clusterservingruntime.serving.kserve.io/kserve-tritonserver serverside-applied
clusterservingruntime.serving.kserve.io/kserve-xgbserver serverside-applied
clusterstoragecontainer.serving.kserve.io/default serverside-applied
clustertrainingruntime.trainer.kubeflow.org/deepspeed-distributed serverside-applied
validatingwebhookconfiguration.admissionregistration.k8s.io/cert-manager-webhook serverside-applied
validatingwebhookconfiguration.admissionregistration.k8s.io/istio-validator-istio-system serverside-applied
validatingwebhookconfiguration.admissionregistration.k8s.io/katib.kubeflow.org serverside-applied
manifests on ⮂ master [!] on ☁️ (us-east-1)

Post-installation checks and next steps

Verify that key components are Ready (Central Dashboard, Pipelines, Notebook Controller, KServe, Katib).
Confirm the Profile namespace was created if you configured a Profile.
Check logs for any failing pods and iterate on overlays or settings as needed.
Explore the installed Kubeflow applications from the dashboard or CLI.

Customize the example kustomization to enable or disable components (Istio overlays, OAuth2-proxy options, KServe, Katib, etc.) depending on your environment. For cloud-specific clusters (GKE, EKS), pick the recommended overlays in the manifests repository.

This completes the installation walkthrough. Explore the manifests, overlays, and installed resources to understand how the Kubeflow platform composes and deploys on your cluster. For more details and troubleshooting, refer to the Kubeflow Manifests repository and the Kubeflow docs.

Course Introduction

Fundamentals of Kubeflow

Working With Kubeflow

KServe and Katib

Profiles and Multi-Tenancy

Bringing It All Together

Demo Installing Kubeflow

Prerequisites

Install kind and create a cluster (example)

Use the Kubeflow-provided kind config

Kubeflow manifests repository

Browsing the example kustomization

Create a default user profile (optional)

Prepare kubeconfig and registry credentials

Install Kubeflow using kustomize + kubectl

Wait for key components

Example of applied resources

Post-installation checks and next steps

Watch Video

​Prerequisites

​Install kind and create a cluster (example)

​Use the Kubeflow-provided kind config

​Kubeflow manifests repository

​Browsing the example kustomization

​Create a default user profile (optional)

​Prepare kubeconfig and registry credentials

​Install Kubeflow using kustomize + kubectl

​Wait for key components

​Example of applied resources

​Post-installation checks and next steps

Watch Video

Prerequisites

Install kind and create a cluster (example)

Use the Kubeflow-provided kind config

Kubeflow manifests repository

Browsing the example kustomization

Create a default user profile (optional)

Prepare kubeconfig and registry credentials

Install Kubeflow using kustomize + kubectl

Wait for key components

Example of applied resources

Post-installation checks and next steps