Skip to content

Latest commit

 

History

History
63 lines (45 loc) · 2.43 KB

KUBEFLOW.md

File metadata and controls

63 lines (45 loc) · 2.43 KB
Raw

Deploying Kubeflow

back to the argo 3d

Guide for those looking to run Kubeflow Version 0.2.2 on their Kuberenetes Cluster

Kubeflows guide in its current state needs lots of work. However, their guide is progressively being updated at Kubeflow.org

Install ksonnet command line utility

brew install ksonnet/tap/ks

RBAC

Before deploying kubeflow make sure you have cluster admin privileges on your kubernetes cluster

kubectl create clusterrolebinding YOURNAME-cluster-admin-binding --clusterrole=cluster-admin [email protected]

Install Kubeflow

Kubeflow is required in order to deploy the custom resources that use Tensorflow: TFJobs.

To install Kubeflow run the following command:

export KUBEFLOW_VERSION=0.2.2
export KUBEFLOW_DEPLOY=false
curl https://raw.githubusercontent.com/kubeflow/kubeflow/v$KUBEFLOW_VERSION/scripts/deploy.sh | bash

This command will automatically deploy a bash script which will set up a ksonnet directory called ksonnet_ks_app with all the necessary built-in packages and deploy them. The export KUBEFLOW_DEPLOY=false command disables ks from deploying to your default namespace.

Now all thats left to deploy kubeflow is a couple ksonnet commands which can only be run inside the kubeflow_ks_app directory

#The first two commands are for running deploying kubeflow in a new namespace.
#Without these commands kubeflow will deploy to the default namespace
cd kubeflow_ks_app
NAMESPACE=mykubeflow
kubectl create ns $NAMESPACE
ks env set default --namespace=$NAMESPACE
ks apply default

If you would like to delete an old instance of kubeflow, or delete it all together use the following commands inside the kubeflow_ks_app directory:

#Undeploy
ks delete $NAMESPACE

#delete ks application (Only necessary if you really f*k'd up)
cd ..
rm -rf kubeflow_ks_app kubeflow_repo

Creating your own TFJob/TFService

This part of the tutorial is simply documentation for creating your own tensorflow application since websites like kubeflow.org are often hard to follow or outdated as of June, 31, 2018.

Kubeflows guide in its current state needs lots of work. Their guide is progressively being updated at Kubeflow.org