Skip to content

Argo Workflows

Argo Workflows is a workflow engine for orchestrating parallel jobs on Kubernetes which we use to define workflows. Each step in the workflow is a container. Argo workflows handles the (de)provisioning of pods and can be used to pass artifacts between them.

We use workflow templates and strict workflows to ensure that workflows cannot be directly sent to Argo Workflows, but instead that only workflow templates can be referenced to start a workflow. This means that no workflows can be started on the Workflows tab of the UI, and instead the Workflow Templates tab must be used to first create a workflow template, which must then be submitted.

The UI of Argo Workflows can be accessed with a local Tilt setup at localhost:2746 or by creating a port forward from the remote cluster. If your browser complains that the server did not send any data, make sure that you have entered the URL with https. Logging in works with token authentication, and the broker's service account may be used to generate a token. After running the local setup, you may call:

kubectl --context <CONTEXT> -n control-plane create token controller-service-account

It is important to note that the value you paste should be Bearer $TOKEN. If you are not automatically redirected to the correct namespace when you open the workflow or workflow templates windows, enter your namespace after a slash in the URL bar of your browser, e.g. localhost:2746/workflows/data-plane.

Since token auto-mounting of Kubernetes pods are disabled, Argo Workflows requires a service account name for the executor to be set. Hence, if you generate a new template using the Create New Workflow Template button, you may encounter a serviceAccountName error. To handle this error, add the following block under the top level spec field of the auto-generated template:

  serviceAccountName: "data-plane-service-account"

Currently, all workflows orchestrated by the Argo Workflow controller are archived for 30 days in a PostgresSQL database.