rhoda-notebook-controller

Recipe for RHODA and Kubeflow Notebook-Controller integration

These instructions assume a basic working knowledge of openshift, jupyter notebooks and RHODA.

Step 1. Verify your notebook-controller installation.

oc -n opendatahub get crd/notebooks.kubeflow.org
oc get pods -l app=notebook-controller -n opendatahub
oc get pods -l app=jupyter-web-app -n opendatahub

Find the route to your Jupyter web interface, login, create and connect to a notebook.

oc get route jupyter -n opendatahub

After your notebook has been created, find the notebook name. You will need to know your notebook namespace (i.e. kubeflow-user):

oc -n kubeflow-user get notebooks

Output should be something like:

bluebook-small   47h

Step 2. Add your connected database.

This step assumes the DBaaSInventory has been populated. (Admin or developer created instance. See the RHODA admin and developer workflows for more information. https://access.redhat.com/documentation/en-us/red_hat_openshift_database_access/1/html-single/quick_start_guide/index)

Follow the developer workflow for instructions on adding a connected database. Make sure to add the connected database in your notebook namespace.

Get the name of the connectedb:

oc -n kubeflow-user get dbaasconnection

Example output:

NAME                        AGE
jary-test-0328-ba0d22970f   41h

Step 3. Setup roles for Service Binding.

Add the following rolebinding:

apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
  name: dbaas-nb-edit
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: ClusterRole
  name: notebook-controller-kubeflow-notebooks-edit
subjects:
- kind: ServiceAccount
  name: service-binding-operator
  namespace: redhat-dbaas-operator

This allows the sa service-binding-operator to modify the notebook.

Step. 4 Create the service binding manually.

The service binding will be between the dbaasconnection added in step 3 and the notebook (kind:notebook, not the deployment in opendatahub or statefulset) in the notebook namespace.

Use the dbaasconnection name from above in the service binding:

apiVersion: binding.operators.coreos.com/v1alpha1
kind: ServiceBinding
metadata:
  name: demo-sb
  namespace: kubeflow-user
spec:
  application:
    group: kubeflow.org
    name: bluebook-small # REPLACE with a notebook from “os get notebook -n kubeflow-user”
    resource: notebooks
    version: v1
  bindAsFiles: true # change to false if you require env vars
  detectBindingResources: true
  services:
  - group: dbaas.redhat.com
    kind: DBaaSConnection
    name: jary-test-0328-ba0d22970f  # REPLACE with name of your dbaasconnection: “oc -n kubeflow-user get dbaasconnection”
    version: v1alpha1

Step 5. Verify the notebook and RHODA integration.

To verify your notebook can use/view the connected database binding, enter your notebook and in a cell run:

!env

In the output, search for "binding". You should see the following:

'SERVICE_BINDING_ROOT': '/bindings',

You may consume this binding by using the pyservicebinding library.

Limitations

Each notebook created from the notebook-controller window will require a servicebinding. New notebooks from within a Jupyter window do not require new servicebindings.