NOTES:
- This is an experimental library, do not place a hard dependency on this unless you may be willing to fork it in the future.
- This library is currently only tested with KFP
1.8.18
If you want Graphviz functionality you must install the binaries via instructions here
pip install -e .
These instructions are vaguely based on the ones provided: one, two, three, by Kubeflow
First, perquisites
- Install kubectl: instructions
- Install minikube: instructions
Then make sure you do not have a previous minikube session running, and start a fresh session with
Note: The
--kubernetes-version=v1.20.15
due to the issues with subsequentkubetctl apply
commands related to KFPv1.8.18
see here
minikube stop
minikube delete
minikube start --cpus 4 --memory 8096 --disk-size=40g --kubernetes-version=v1.20.15
Install KFP standalone, here I install 1.8.18
export PIPELINE_VERSION=1.8.18
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"
View the Kubeflow UI by port forwarding and visiting http://localhost:8080/
kubectl port-forward -n kubeflow svc/ml-pipeline-ui 8080:80
And connect with an sdk client via the follow
NOTE: There may be sometime after startup for which this will 504
from kfp import Client
client = Client(host='http://localhost:8080')