This bundle deploys KubeFlow to a Juju k8s model.
KubeFlow consists of:
-
TensorFlow Hub, a JupyterHub for running interactive notebooks with TensorFlow libraries available
-
TensorFlow Dashboard, to manage TensorFlow jobs
-
TensorFlow Training, for training TensorFlow models
-
Ambassador, an API gateway for managing access to the services
Note: This bundle is currently missing Seldon, which will be added soon.
You'll need to snap install both juju and microk8s. Kubeflow
requires at least juju 2.5rc1.
sudo snap install juju --beta --classic
sudo snap install microk8s --edge --classic
Next, either run the commands individually from this script, or copy/paste them into a local script and run it:
#!/usr/bin/env bash
CLOUD=uk8s-kf-cloud
MODEL=uk8s-kf-model
cleanup() {
# Clean up resources
microk8s.kubectl delete ns $MODEL
juju kill-controller localhost-localhost -y -t0
juju remove-cloud $CLOUD
}
trap cleanup EXIT
set -eux
# Set up juju and microk8s to play nicely together
sudo microk8s.enable dns storage
juju bootstrap lxd
microk8s.config | juju add-k8s $CLOUD
juju add-model $MODEL $CLOUD
juju create-storage-pool operator-storage kubernetes storage-class=microk8s-hostpath
# Deploy kubeflow to microk8s
juju deploy cs:~juju/kubeflow
# Exposes the Ambassador reverse proxy at http://localhost:8081/
# The TF Jobs dashboard is available at http://localhost:8081/tfjobs/ui/
# The JupyterHub dashboard is available at http://localhost:8081/hub/
# When you're done, ctrl+c will exit this script and free the created resources
microk8s.kubectl port-forward svc/juju-kubeflow-ambassador -n $MODEL 8081:80
You will first need to create an AWS account for juju to use, and then add the credentials to juju:
$ juju add-credential aws
Enter credential name: kubeflow-test
Using auth-type "access-key".
Enter access-key: <YOUR ACCESS KEY>
Enter secret-key: <YOUR SECRET KEY>
Credential "kubeflow-test" added locally for cloud "aws".
Next, you can run the commands in this script individually, or copy it into a local script and run the entire script.
#!/usr/bin/env bash
# Clean up generated resources on exit
cleanup() {
juju kill-controller aws-us-east-1 -y -t0
}
trap cleanup EXIT
set -eux
CLOUD=aws-kf-cloud
MODEL=aws-kf-model
# Set up Kubernetes cloud on AWS
juju bootstrap aws/us-east-1
juju deploy cs:bundle/canonical-kubernetes
juju deploy cs:~containers/aws-integrator
juju trust aws-integrator
juju add-relation aws-integrator kubernetes-master
juju add-relation aws-integrator kubernetes-worker
# Wait for cloud to finish booting up
juju wait -e aws-us-east-1:default -w
# Copy details of cloud locally, and tell juju about it
juju scp kubernetes-master/0:~/config ~/.kube/config
juju add-k8s $CLOUD
juju add-model $MODEL $CLOUD
# Set up some storage for the new cloud, deploy Kubeflow, and wait for
# Kubeflow to boot up
juju create-storage-pool operator-storage kubernetes storage-class=juju-operator-storage storage-provisioner=kubernetes.io/aws-ebs parameters.type=gp2
juju create-storage-pool k8s-ebs kubernetes storage-class=juju-ebs storage-provisioner=kubernetes.io/aws-ebs parameters.type=gp2
juju deploy cs:~juju/kubeflow
juju wait -e aws-us-east-1:$MODEL -w
# Exposes the Ambassador reverse proxy at http://localhost:8081/
# The TF Jobs dashboard is available at http://localhost:8081/tfjobs/ui/
# The JupyterHub dashboard is available at http://localhost:8081/hub/
# When you're done, ctrl+c will exit this script and free the created resources
kubectl port-forward svc/juju-kubeflow-ambassador -n $MODEL 8081:80
To submit a TensorFlow job to the dashboard, you can run this kubectl command:
kubectl create -n <NAMESPACE> -f path/to/job/definition.yaml
Where <NAMESPACE> matches the name of the Juju model that you're using, and
path/to/job/definition.yaml should point to a TFJob definition similar to the
tf_job_mnist.yaml example found here.
You can submit a model to be served with TensorFlow Serving. See the documentation in the TF Serving Charm for more information.