Running the platform

This page describes how to deploy the Renku platform on kind. Running GitLab (which is a part of Renku) is quite resource intensive. For this reason we describe a setup which excludes GitLab from the Renku deployment and instead uses a GitLab instance running elsewhere, for example, gitlab.com as the GitLab backend. For the Renku graph features to function properly in this setup, the GitLab instance needs to be able to contact the locally running platform through a webhook. This can be achieved through many services like ngrok or the like. We are going to use localhost.run, a minimalistic method based on ssh tunneling which allows for a somewhat static hostname.

The renku source code is hosted on github: https://github.com/SwissDataScienceCenter/renku.

Prerequisites

You will need the following tools installed on your machine:

Clone the repo

Get the code by cloning the Renku repository:

$ git clone https://github.com/SwissDataScienceCenter/renku.git
$ cd renku/helm-chart

Install python dependencies

We recommend using pipenv for managing your virtual python environment. Install all the necessary python dependencies:

$ pipenv install --dev

Start kind

Next we start kind. We make it so that ports 8080 and 8443 on the host machine are forward respectively to port 80 and 443 to the cluster. This will allow you to access renku.

$ cat <<EOF | kind create cluster --name kind --config=-
kind: Cluster
apiVersion: kind.x-k8s.io/v1alpha4
nodes:
- role: control-plane
  kubeadmConfigPatches:
  - |
    kind: InitConfiguration
    nodeRegistration:
      kubeletExtraArgs:
        node-labels: "ingress-ready=true"
  extraPortMappings:
  - containerPort: 80
    hostPort: 8080
    protocol: TCP
EOF

Once kind has started, make sure you can access it by running:

$ kubectl get node
NAME                 STATUS   ROLES                  AGE   VERSION
kind-control-plane   Ready    control-plane,master   43s   v1.21.1

Deploy the NGINX ingress controller

We need an ingress controller to expose HTTP and HTTPS routes from outside the cluster to services within the cluster. We use nginx ingress. Deploy the ingress controller and then run the wait command below to make sure it becomes fully available before proceeding further.

$ kubectl apply -f https://raw.githubusercontent.com/kubernetes/ingress-nginx/main/deploy/static/provider/kind/deploy.yaml
$ kubectl wait --namespace ingress-nginx \
    --for=condition=ready pod \
    --selector=app.kubernetes.io/component=controller \
    --timeout=300s

Build and pull all necessary charts

$ helm repo add renku https://swissdatasciencecenter.github.io/helm-charts
$ pipenv run chartpress
$ helm dep update renku

Set up localhost.run

We use a service called localhost.run to expose the platform by establishing an ssh tunnel.

$ ssh -R 80:localhost:8080 ssh.localhost.run

This will start the tunnel and display your hostname of the style http://<some-name>.lhr.domains. In subsequent steps this domain will be added to the configuration for Renku.

Note: When you stop and restart the tunnel without waiting for too long, you should receive the same subdomain that you have previously had. If this is not the case you will have to reconfigure the gitlab.com client application and recreate your Renku deployment (see next steps).

Note: The second port number in the ssh command above (i.e. 8080) should be the same as the port number you connected on your host machine to the kind cluster. If you use a different port number in the command when you create your kind cluster then make sure you use that port number here as well. Using a common port on the host side like 80 is not recommended because it can cause issues with the routing between Renku, the host and the kind cluster.

Set up the gitlab client application

Browse to GitLab, then click on the user icon and select settings > applications. Register a new client application with the following settings:

  1. Name: renku-local

  2. Scopes: all except sudo

  3. Redirect URIs (for <renku-domain> use the value provided by the localhost.run service, from earlier - this domain usually has a format like <some-name>.lhr.domains):

https://<renku-domain>/login/redirect/gitlab
https://<renku-domain>/api/auth/gitlab/token

Copy the Application Id and the Secret that appear at the end of this procedure.

Configure the values.yaml file

The values.yaml file will depend on the URL that GitLab is running under. We have two templates available in the cloned repository in the folder helm-chart/example-configurations/.

The flagship gitlab.com instance is available at the root of the domain. If you are using gitlab.com or any other instance where GitLab is accessible at the root of the domain, use the file helm-chart/example-configurations/gitlab_dot_com-gitlab-values.yaml as a template and copy this file to helm-chart/example-configurations/external-gitlab-values.yaml.

If you are using renkulab.io/gitlab or another instance of GitLab that is accessible under the path /gitlab, then use helm-chart/example-configurations/renkulab-gitlab-values.yaml as a template and copy this file to helm-chart/example-configurations/external-gitlab-values.yaml.

(If the GitLab instance is available under a different path than either the root or /gitlab, use the gitlab_dot_com-gitlab-values.yaml file as a template and adapt the path.)

To complete the template, you should open helm-chart/example-configurations/external-gitlab-values.yaml and enter the required variables that are indicated by TODO:. You need to provide the Application Id and the Secret from the GitLab client application created in the previous step.

Deploy Renku to kind

Start the renku platform using helm:

$ kubectl create namespace renku
$ helm upgrade renku --install --namespace renku ./renku \
  -f minikube-values.yaml -f example-configurations/external-gitlab-values.yaml \
  --timeout 3600s

Executing this command for the first time can easily take a long time depending on your internet connection speed (even 30+ minutes) as all the necessary docker images need to be pulled. Do not interrupt the upgrade command.

If you want to verify that things are moving on, open a new terminal any type kubectl -n renku get pod. The renku-keycloak-* pod is the slowest. If the Ready column reports 0/1 and Restarts is not higher than 1, all is going as expected. This would be a good moment to grab yourself a coffee…

Once the above command has returned you should be able to access the platform using your browser under http://<some-name>.lhr.domains.

Manage Keycloak

Keycloak is an identity management solution which is deployed as a part of Renku. You can access the admin interface at http://<some-name>.lhr.domains/auth. The admin username is “admin” and the admin password can be looked up in the corresponding kubernetes secret:

$ kubectl get secrets -n renku keycloak-password-secret -o json | \
    jq -r '.data["keycloak-password"]' | base64 --decode

Configure gitlab.com as the identity provider (optional)

It is possible to configure gitlab.com or another external GitLab deployment as an identity provider for your local Renku deployment. This has the benefit of having only one source for user information and making the login process for the platform more convenient. While logged into the Keycloak admin interface, go to Identity Providers and choose “GitLab” from the “Add Provider…” drop-down menu. Add the Application Id and Secret from the Renku-local application created earlier. Save the identity provider and logout from the Keycloak admin panel.

Configure GitLab CI

For each created project you will have to modify the .gitlab-ci.yaml in the repository for the image build to work. In the image_build job

  • remove the image-build tag

  • add the following entry:

services:
  - docker:dind