In this lab we will install Concourse to new worker nodes within the shared cluster via a Helm chart. The following modifications to the default chart values need to be made:
- Use Contour Ingress
- Generate certificate via Let's Encrypt
- Updated URLs
Concourse will also be managed via Tanzu Mission Control in a dedicated workspace.
The following section should be added to or exist in your local params.yaml file:
concourse:
namespace: concourse
fqdn: concourse.tkg-shared.<your-domain>
tmc-workspace: concourse-workspace
Once these are in place and correct, run the following to export the following into your shell:
export TMC_CLUSTER_GROUP=$(yq r params.yaml tmc.cluster-group)
export CONCOURSE_NAMESPACE=$(yq r params.yaml concourse.namespace)
export CONCOURSE_TMC_WORKSPACE=$TMC_CLUSTER_GROUP-$(yq r params.yaml concourse.tmc-workspace)
export CLUSTER_NAME=$(yq r params.yaml shared-services-cluster.name)
export IAAS=$(yq r params.yaml iaas)
export VMWARE_ID=$(yq r params.yaml vmware-id)
Rather than creating a new cluster, which we normally would recommend for Concourse, to save on resources we can create a couple of worker nodes in our shared cluster and taint them. What this does is enable us to use these new nodes exclusively for Concourse. The Concourse web and worker pods will be given a toleration for the taint, and use node selectors to ensure that they are "pinned" to the new cluster cworker nodes. To do this, we will:
- Scale the Shared Services cluster
- Label the new nodes
- Taint the labelled nodes to prevent any other workload from running there
- Apply the toleration and node selector to the Concourse Helm chart
First, create a pair of new worker nodes in the tkg cluster:
NEW_NODE_COUNT=$(($(yq r params.yaml shared-services-cluster.worker-replicas) + 2))
tkg scale cluster $CLUSTER_NAME -w $NEW_NODE_COUNT
After a few minutes, check to see which new nodes were created. These will be the nodes that are the most recent in the AGE output of "kubectl get nodes". Label these nodes uisng a space to separate the list of nodes in the command:
kubectl get nodes
kubectl label nodes ip-10-0-0-57.us-east-2.compute.internal ip-10-0-0-105.us-east-2.compute.internal type=concourse
Now add the taint to the same nodes. This will prevent other workloads from being scheduled on the new nodes. Note that any DaemonSet may still have a pod running on these nodes because the taint was not added until after node creation.
kubectl taint nodes -l type=concourse type=concourse:PreferNoSchedule --overwrite
In order to deploy the Helm chart for Concourse to a dedicated namespace, we need to create it first. To do this, we can use Tanzu Mission Control, as it is already running on our shared services cluster. This will create a "managed namespace", where we can assert additional control over what is deployed.
tmc workspace create -n $CONCOURSE_TMC_WORKSPACE -d "Workspace for Concourse"
tmc cluster namespace create -c $VMWARE_ID-$CLUSTER_NAME-$IAAS -n $CONCOURSE_NAMESPACE -d "Concourse product installation" -k $CONCOURSE_TMC_WORKSPACE
Create the Concourse namespace and generate the deployment file. This file (generated/tkg-shared/concourse/concourse-values-contour.yaml) will contain oeverrides to the default chart values.
./scripts/generate-concourse.sh
Add the repository to helm and use the generated deployment file to deploy the chart.
helm repo add concourse https://concourse-charts.storage.googleapis.com/
helm repo update
helm upgrade --install concourse concourse/concourse --values generated/tkg-shared/concourse/concourse-values-contour.yaml -n $CONCOURSE_NAMESPACE
Lastly, add the newly created namespace for the Concourse main team to TMC's managed workspace we created earlier.
tmc cluster namespace attach -c $VMWARE_ID-$CLUSTER_NAME-$IAAS -n concourse-main -k $CONCOURSE_TMC_WORKSPACE
- All Concourse pods are in a running state, on the tainted nodes:
kubectl get po -n $CONCOURSE_NAMESPACE -o wide
- Certificate is True and Ingress created:
kubectl get cert,ing -n $CONCOURSE_NAMESPACE
- Open a browser and navigate to the FQDN you defined for Concourse above. The default user is test/test. This can be controlled by editing the deployment values.