Kubeflow Testing

Setup

PS: Short alias for kubectl alias k=kubectl

Install kubeflow python sdk:

pip3 install https://storage.googleapis.com/ml-pipeline/release/0.1.29/kfp.tar.gz --upgrade --user

Install kind

brew install kind

Create cluster

kind create cluster --config=yamls/kind.yaml

Set up registry

docker run -d --restart=always -p "127.0.0.1:5001:5000" --name "kind-registry" registry:2
docker network connect "kind" "kind-registry"

Install kubeflow

export PIPELINE_VERSION=1.8.4
k apply -k "github.com/kubeflow/pipelines/manifests/kustomize/cluster-scoped-resources?ref=$PIPELINE_VERSION"
k wait --for condition=established --timeout=60s crd/applications.app.k8s.io
k apply -k "github.com/kubeflow/pipelines/manifests/kustomize/env/platform-agnostic-pns?ref=$PIPELINE_VERSION"

This will take some time, can check progress with k get pods -n kubeflow

Make log directory

mkdir logs

Set up port forwarding

k port-forward -n kubeflow svc/ml-pipeline-ui 3000:80 >> ./logs/ml-pipeline-ui.logs 2>&1 & \
  k port-forward -n kubeflow svc/minio-service 9000:9000 >> ./logs/minio-service.logs 2>&1 &

Can now go to dashboard

Run Tweet Pipeline

Install components

python setup.py install --user

Run pipeline

python3 code/tweet_pipeline.py

If you make changes to components:

pip3 uninstall components && python setup.py install --user

Build base docker image

docker build . -t 'component_reqs'

Then push to local repository

docker tag <ID> 127.0.0.1:5001/component_reqs:latest
docker push 127.0.0.1:5001/component_reqs:latest

Clean up cluster

kubectl delete pods --field-selector status.phase=Succeeded --all-namespaces
kubectl delete pods --field-selector status.phase=Failed --all-namespaces
kubectl delete pods --field-selector status.phase=Error --all-namespaces

Uninstall kubeflow

Delete everything

export PIPELINE_VERSION=1.8.4
k delete -k "github.com/kubeflow/pipelines/manifests/kustomize/env/platform-agnostic-pns?ref=$PIPELINE_VERSION"
k delete -k "github.com/kubeflow/pipelines/manifests/kustomize/cluster-scoped-resources?ref=$PIPELINE_VERSION"

and

k config delete-cluster kind-ness

Switch back to context

k config use-context docker-desktop

Unistall kind

brew uninstall kind