Intro

Welcome! By following the instructions below, you should be able to deploy a kubernetes cluster running on Google Cloud Platform GKE, a VPC, Kubeflow, and Istio.

Kubeflow is a portable and scalable machine learning toolkit, that eases the task of deploying machine learning workflows on Kubernetes.
Kubeflow provides out-of-the-box:

Jupyter notebooks
TensorFlow model training job operator
TensorFlow Serving container to export trained models to Kubernetes
Kubeflow Pipelines for deploying and managing end-to-end ML workflows
Istio + Ambassador for ingress
And more

Project structure

README.md
You are here
Makefile
High level cli interface
gcp/
infrastructure-as-code for GKE cluster and Google VPC
gcp.tf
Initialization of terraform module declared in .gcp
k8s/kubeflow
Kubeflow k8s files

Pre-requisites

A GCP account
A GCP Project associated with a billing account
GCloud SDK
kubectl

Step by step deployment of a GKE cluster with Kubeflow and Istio

Double check each item at the Pre-requisites section
Install kfctl locally with make deps-macos or make deps-linux
Run make auth-gcloud and grant gcloud authorization to interact with GCP's k8s API
Run make gcp-build to deploy a VPC and a GKE cluster
You might need to wait a few minutes for the VPC and GKE cluster to be ready
Run make kubeflow-refresh to refresh your kubeflow infrastructure's code
Run make kubeflow-build to install kubeflow

Or cross your fingers and run make all ;-)

Shutdown and remove all the infrastructure

make clean

Backlog