Installing Deis Workflow on Amazon Web Services

Check Your Setup

First check that the helm command is available and the version is v2.5.0 or newer.

$ helm version
Client: &version.Version{SemVer:"v2.5.0", GitCommit:"012cb0ac1a1b2f888144ef5a67b8dab6c2d45be6", GitTreeState:"clean"}
Server: &version.Version{SemVer:"v2.5.0", GitCommit:"012cb0ac1a1b2f888144ef5a67b8dab6c2d45be6", GitTreeState:"clean"}

Ensure the kubectl client is installed and can connect to your Kubernetes cluster.

Add the Hephy Chart Repository

Helm changed its UX adding support for OCI, so helm repo add is not a thing anymore.

The Team Hephy Chart Repository contains everything needed to install Hephy Workflow onto a Kubernetes cluster, but Helm's legacy chart repository data structures are chronically unscalable.

As a result, there's no need to ever run this command again:

$ helm repo add hephy https://charts.teamhephy.com/

(Note: at the time of this writing, the chartmuseum host is currently down for the count.)

Install Hephy Workflow

Now that Helm is installed, and the repository has been added install Workflow by running:

$ helm install deis-hephy --namespace deis oci://ghcr.io/kingdonb/hephy-workflow-beta/workflow

Helm will install a variety of Kubernetes resources in the deis namespace. Wait for the pods that Helm launched to be ready. Monitor their status by running:

$ kubectl --namespace=deis get pods

If it's preferred to have kubectl automatically update as the pod states change, run (type Ctrl-C to stop the watch):

$ kubectl --namespace=deis get pods -w

Depending on the order in which the Workflow components initialize, some pods may restart. This is common during the installation: if a component's dependencies are not yet available, that component will exit and Kubernetes will automatically restart it.

Here, it can be seen that the controller, builder and registry all took a few loops before they were able to start:

$ kubectl --namespace=deis get pods
NAME                          READY     STATUS    RESTARTS   AGE
deis-builder-hy3xv            1/1       Running   5          5m
deis-controller-g3cu8         1/1       Running   5          5m
deis-database-rad1o           1/1       Running   0          5m
deis-logger-fluentd-1v8uk     1/1       Running   0          5m
deis-logger-fluentd-esm60     1/1       Running   0          5m
deis-logger-sm8b3             1/1       Running   0          5m
deis-minio-4ww3t              1/1       Running   0          5m
deis-registry-asozo           1/1       Running   1          5m
deis-router-k1ond             1/1       Running   0          5m
deis-workflow-manager-68nu6   1/1       Running   0          5m

Once all of the pods are in the READY state, Deis Workflow is up and running!

Next, configure dns so you can register your first user and deploy an application.

Configure your AWS Load Balancer

After installing Workflow on your cluster, you will need to adjust your load balancer configuration. By default, the connection timeout for Elastic Load Blancers is 60 seconds. Unfortunately, this timeout is too short for long running connections when using git push functionality of Deis Workflow.

Deis Workflow will automatically provision and attach a Elastic Loadbalancer to the router component. This component is responsible for routing HTTP and HTTPS requests from the public internet to applications that are deployed and managed by Deis Workflow.

By describing the deis-router service, you can see what IP hostname has been allocated by AWS for your Deis Workflow cluster:

$ kubectl --namespace=deis describe svc deis-router | egrep LoadBalancer
Type:                   LoadBalancer
LoadBalancer Ingress:   abce0d48217d311e69a470643b4d9062-2074277678.us-west-1.elb.amazonaws.com

The AWS name for the ELB is the first part of hostname, before the -. List all of your ELBs by name to confirm:

$ aws elb describe-load-balancers --query 'LoadBalancerDescriptions[*].LoadBalancerName'
abce0d48217d311e69a470643b4d9062

Set the connection timeout to 1200 seconds, make sure you use your load balancer name:

$ aws elb modify-load-balancer-attributes \
        --load-balancer-name abce0d48217d311e69a470643b4d9062 \
        --load-balancer-attributes "{\"ConnectionSettings\":{\"IdleTimeout\":1200}}"
abce0d48217d311e69a470643b4d9062
CONNECTIONSETTINGS  1200