kubeflow/community-distribution

Kubeflow Community Distribution – Quick Overview

What it is: A set of ready‑to‑apply Kubernetes manifests that let you spin up the full Kubeflow machine‑learning platform on a variety of clusters (Kind, Minikube, Rancher, EKS, AKS, GKE, etc.). It bundles all core Kubeflow components—Pipelines, Notebooks, Katib, KServe, Spark Operator, the Central Dashboard, plus shared services like Istio, Knative, Cert‑Manager, Dex, and OAuth2‑Proxy.

Who it’s for:

• Enterprises that need a reproducible, security‑focused ML platform with multi‑tenant support.
• Researchers or hobbyists who want a single‑command way to try the whole Kubeflow stack locally.

How you use it:

1. Install the prerequisites (kind/kubectl/kustomize, Docker/Podman, enough RAM/CPU).
2. Run the provided script to create a Kind cluster (or follow the docs for other clouds).
3. Apply the generated kustomization in one line (the command retries until all CRDs are ready).
4. Log in to the Kubeflow Central Dashboard with the default credentials (user@example.com / 12341234) and change them immediately for production.

You can also install components individually by applying the corresponding kustomize overlays, which is handy for debugging or for building a custom distribution.

Key components & resource hints (from the README):

Component	Approx. CPU	Approx. Memory	PVC storage
Pipelines	970 m	3552 Mi	35 GB
KServe	600 m	1200 Mi	–
Katib	13 m	476 Mi	10 GB
Istio	750 m	2364 Mi	–
…	…	…	…

Maintenance & support

• Two stable releases per year (e.g., 26.03, 26.10) with a 6‑month best‑effort community support window.
• Commercial support options are listed on the Kubeflow website.
• CI runs an end‑to‑end installation test on every PR to keep the manifests reliable.

Why it matters: Kubeflow is the de‑facto open‑source platform for running end‑to‑end ML workflows on Kubernetes. This repository removes the heavy lifting of stitching together dozens of upstream projects, giving you a single source of truth for a production‑grade deployment.

---

TL;DR Installation (single command)

# Create a Kind cluster with the helper script
./tests/install_KinD_create_KinD_cluster_install_kustomize.sh

# Export the kubeconfig
export KUBECONFIG=$(kind get kubeconfig-path --name kubeflow)

# Deploy the whole platform (retries automatically)
while ! kustomize build example | kubectl apply --server-side --force-conflicts -f -; do
  echo "Retrying…"; sleep 20;
 done

After it finishes, open the dashboard at http://<cluster‑ip>/ and log in.

---

Links

• Latest stable release
• Full CI workflow
• Community Slack & meetings

---

All details above are taken directly from the repository’s README; no extra features have been added.