Integrate Run:ai with Argo Workflows¶

Argo Workflows is an open source container-native workflow engine for orchestrating parallel jobs on Kubernetes.

This document describes the process of using Argo Workflows in conjunction with Run:ai. Argo Workflows submits jobs that are scheduled via Run:ai.

Install Argo Workflows¶

Use the default installation to install Argo Workflows. As described in the documentation, open the Argo Workflows UI by running:

kubectl -n argo port-forward deployment/argo-workflows-server 2746:2746

Then browse to localhost:2746

Create a Run:ai Project¶

Using the Run:ai user interface, create a Run:ai Project. A Project named team-a will create a Kubernetes namespace named runai-team-a.

Run an Argo Workflow with Run:ai¶

Create an Argo Workflows Template¶

Within the Argo Workflows user interface, go to Templates and create a new Template. Add the following metadata:

spec:
  templates:
    - name: <WORKFLOW-NAME>
      metadata:
        labels:
          project: team-a # (1)

Name of Project.

Create and Run the Workflow¶

Create an Argo Workflow from the template and run it. Open the Run:ai user interface, go to Jobs, and verify that you can see the new Job.

Using GPU Fractions with Argo Workflows¶

To run an Argo Workflow using GPU Fractions, you will need to add an annotation:

spec:
  templates:
    - name: <WORKFLOW-NAME>
      metadata:
        annotations:
          gpu-fraction: '0.5' # (1)
        labels:
          project: team-a # (2)

Size of required GPU Fraction.
Name of Project.