Add docs for KIND deployment

[ckadner]

Add first draft document for deploying MLX on KIND (Kubernetes in Docker).

https://github.com/ckadner/mlx/blob/add_kind_deployment_doc/docs/install-mlx-on-kind.md

---

Deploy MLX on KIND

Kubernetes in Docker (KIND) provides an easy way to deploy MLX locally including
Kubeflow Pipelines which makes it possible to run generated sample pipelines for
any of the registered MLX assets.

Installation

• Docker
• Homebrew (on macOS)
• Kustomize
• kubectl
• KIND

Install Required CLIs (macOS)

After installing Docker and Homebrew (linked above) you can install the kind,
kustomize, and kubectl CLIs with brew install. For Windows and Linux follow
the respective home pages for installation instructions.

brew install kind
kind --version

brew install kubectl
kubectl version --client

brew install kustomize
kustomize version

Note: We successfully tested this KIND deployment with the latest version of kustomize v4.4.0.
However, there have been issues in the past with versions later then v3.2.0. To be on the safe side
you could download the kustomize v3.2.0 binary as described
here

Docker Resources

Increase the default resources for Docker:

• CPUs: 4 Cores
• Memory: 8 GB RAM
• Disk: 32+ GB

Note: We found that on older laptops, like a 2016 15 in MacBook Pro (2.7 GHz i7, 16 GB) the MLX
deployment on KIND may require to give all available resources to the Docker daemon in order to be
able to deploy the manifests and run basic pipelines. Even then, trying to run notebooks or deploying
a model, will cause the laptop to get very slow with fans running full throttle. It may even cause
other application to crash.

Create KIND Cluster

kind create cluster --name mlx
kubectl cluster-info --context kind-mlx
kubectl get pods --all-namespaces

Deploy MLX (Single-User)

git clone https://github.com/IBM/manifests -b v1.4.0-mlx
cd manifests

# run the below command two times if the CRDs take too long to provision.
while ! kustomize build mlx-single-kind | \
  kubectl apply -f -; do echo "Retrying to apply resources"; sleep 10; done

# check pod status
kubectl get pods --all-namespaces

# make the MLX UI available to your local browser on http://localhost:3000/
kubectl port-forward -n istio-system svc/istio-ingressgateway 3000:80

Now paste the URL http://localhost:3000/ into your browser and proceed to
import the MLX catalog.

Delete the mlx cluster when it is no longer needed:

kind delete cluster --name mlx

Install Kubeflow Pipelines (for reference, optional)

kind create cluster --name kfp
kubectl cluster-info --context kind-kfp

# env/platform-agnostic-pns hasn't been publically released, so you will install it from master
export PIPELINE_VERSION=1.7.1
kubectl apply -k "github.com/kubeflow/pipelines/manifests/kustomize/cluster-scoped-resources?ref=$PIPELINE_VERSION"
kubectl wait --for condition=established --timeout=60s crd/applications.app.k8s.io
kubectl apply -k "github.com/kubeflow/pipelines/manifests/kustomize/env/platform-agnostic-pns?ref=$PIPELINE_VERSION"

kubectl get pods --all-namespaces

# make the Kubeflow Pipelines UI available on http://localhost:8080/#/pipelines
kubectl port-forward -n kubeflow svc/ml-pipeline-ui 8080:80

kind delete cluster --name kfp

---

@Tomcli @yhwang

[mlx-bot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: ckadner

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Needs approval from an approver in each of these files:

• OWNERS [ckadner]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[Tomcli]

I have the 2016 15 inch macbook pro. Here are my spec
2.7 GHz Quad-Core Intel Core i7
16 GB 2133 MHz LPDDR3

By giving all my resources to the Docker daemon, I'm able to deploy the manifests and run basic pipelines. However, if I try to run notebooks or after the model is deployed, my laptop will get very slow and loud to a point that start to crash my apps.

[ckadner]

Thanks @Tomcli @yhwang -- I updated the doc with your comments

[Tomcli]

We need kustomize 3.2 for this manifest. Can you point to the kustomize instruction like the one on ibm kubeflow?
https://www.kubeflow.org/docs/distributions/ibm/deploy/deployment-process/#kubeflow-installation

[ckadner]

I added this para:

Note: We successfully tested this KIND deployment with the latest version of kustomize v4.4.0.
However, there have been issues in the past with versions later then v3.2.0. To be on the safe side
you could download the kustomize v3.2.0 binary as described
here

[ckadner]

Thanks @yhwang and @Tomcli for the manifest updates

[ckadner]

@yhwang @Tomcli -- I think the install-mlx-on-kind.md is ready to be merged, unless you have further change requests.

We can improve it in follow up PRs, like coming up with a better alternative to the port-forwarding:

Instead of:

kubectl port-forward -n istio-system svc/istio-ingressgateway 3000:80

Use a kind cluster configuration like this

apiVersion: kind.x-k8s.io/v1alpha4
kind: Cluster
nodes:
- role: control-plane
  extraPortMappings:
  - containerPort: 3000
    hostPort: 80
- role: worker

[ckadner]

@phu since you got the most powerful laptop in the team now, would you mind giving this KIND deployment a spin and let us know your feedback? Deployment takes about 20 minutes on older machines. I am curious what deployment times you'll get :-) Thanks!

[Tomcli]

/lgtm