Argo Workflows

Argo Workflows  enables multi-step agent operations with parallel execution, conditional logic, and workflow visualization.

Charts that are preconfigured to enable Argo workflows on Ark are available so that you can rapidly build out use cases.

Installing

This installs Argo in single-namespace mode within the Ark tenant namespace (typically default in development mode). Production installations will typically use a cross-tenant Argo operator with workflow execution distributed across Ark tenant namespaces.

Install via Helm, or for local development mode use devspace:

helm upgrade --install argo-workflows \
  oci://ghcr.io/mckinsey/agents-at-scale-ark/charts/argo-workflows
 
# Or for local development, use devspace:
cd services/argo-workflows
devspace dev
 
# Check status
kubectl get pods -l app.kubernetes.io/part-of=argo-workflows
 
# Port-forward to dashboard (this is done automatically when using devspace).
kubectl port-forward svc/argo-workflows-server 2746:2746
# Dashboard: http://localhost:2746

Minio Artifact Storage

Argo Workflows can use Minio for storing workflow artifacts like intermediate outputs, logs, and data files. This is useful for passing data between workflow steps or preserving workflow outputs.

To enable Minio artifact storage, first install the Minio Operator:

helm upgrade minio-operator operator \
  --install \
  --repo https://operator.min.io \
  --namespace minio-operator \
  --create-namespace \
  --version 7.1.1
 
# Check status
kubectl get pods -n minio-operator

Then enable Minio in the Argo Workflows chart:

helm upgrade --install argo-workflows \
  oci://ghcr.io/mckinsey/agents-at-scale-ark/charts/argo-workflows \
  --set minio.enabled=true
 
# Or for local development:
cd services/argo-workflows
devspace dev  # Select 'true' when prompted to enable Minio
 
# Check Minio tenant status
kubectl get tenant

When using devspace dev, you’ll be prompted whether to enable Minio artifact storage. If enabled, DevSpace automatically installs the Minio Operator and tenant.

Uninstalling

Uninstall Argo workflows:

# Uninstall a Helm installation
helm uninstall argo-workflows
 
# Or purge if using devspace
devspace purge

Note: by default, the Argo chart does NOT remove installed CRDs to prevent accidental data deletion. This means re-installing to a different namespace will cause a clash.

To remove Argo CRDs (including Workflow, WorkflowTemplate, etc.):

kubectl get crd -o name | grep argoproj.io | xargs kubectl delete

This will delete all workflows, templates, and related data from the cluster. Use with caution.

Uninstalling Minio

If you enabled Minio artifact storage, you can optionally uninstall it:

# Uninstall Minio Operator
helm uninstall minio-operator --namespace minio-operator

Note: Similar to Argo, removing Minio CRDs will delete stored artifacts and data:

kubectl get crd -o name | grep min.io | xargs kubectl delete

Opening the Argo and Minio Dashboards

Wait for services to be ready before accessing dashboards.

You can now check the Argo Workflows dashboard as well as the Minio dashboard:

# Show the Argo Workflows dashboard:
kubectl port-forward svc/argo-workflows-server 2746:2746
# http://localhost:2746
 
# Show the minio dashboard:
kubectl port-forward svc/myminio-console 9443:9443
# https://localhost:9443
# Username: minio
# Password: minio123

Running Workflows

Start by creating a workflow or workflow template.

A sample template is provided that shows how to run Ark queries:

kubectl apply -f services/argo-workflows/samples/query-fanout-template.yaml

Open the Argo dashboard with at http://localhost:2746 and choose “Workflows > New Workflow > From Template > Query Fanout”. The workflow will look something like this when running:

Workflows can also be run using the Argo CLI:

# List workflow templates if needed.
argo template list
 
# output:
# NAME
# query-fanout-template
 
# Run a workflow.
argo submit --from workflowtemplate/query-fanout-template \
  -p question="Describe how to monitor argo workflows from the command-line" \
  -p evaluator-model="default"
 
# Watch workflow progress
argo watch @latest
 
# View results, or status of all workflows.
argo logs @latest
argo list workflows

This workflow demonstrates how to run Ark and Argo operations:

1. Take input paramters
2. Lists Ark models
3. Fan-out a workflow into a step per model
4. Run Ark queries
5. Fan-in and evalate results (quality, token count, etc)

Samples

Ark Tools

The ark-tools-template.yaml provides a workflow for running Ark CLI commands in containers. The ark-tools image includes the Ark CLI pre-installed, enabling workflows to query agents and capture results as workflow outputs.

kubectl apply -f services/argo-workflows/samples/ark-tools-template.yaml

Use this template to integrate Ark agent queries into your workflows without building custom containers.

Minio Artifacts

The minio-artifact-template.yaml demonstrates storing and retrieving artifacts using Minio. This workflow creates an artifact in one step and retrieves it in another, showing how to pass data between workflow steps.

kubectl apply -f services/argo-workflows/samples/minio-artifact-template.yaml

Screenshot:

A2A Arithmetic Workflow

The a2a-arithmetic-workflow.yaml demonstrates combining A2A agents, Ark agents, and Python scripts. The workflow validates prerequisites, performs arithmetic operations, and queries agents for results.

Install mock-llm  with A2A support - this will create a simple countdown-agent that returns an A2A task that counts down from a given input number of seconds:

helm upgrade --install mock-llm oci://ghcr.io/dwmkerr/charts/mock-llm \
  --set ark.a2a.enabled=true
 
# Check the A2A server is ready
kubectl get a2aserver mock-llm-countdown
# NAME                   ADDRESS                                                 READY
# mock-llm-countdown     http://mock-llm:6556/a2a/agents/countdown-agent         true

Install and run the workflow:

kubectl apply -f services/argo-workflows/samples/a2a-arithmetic-workflow.yaml
argo submit --from workflowtemplate/a2a-arithmetic-workflow -p a="2" -p b="3"
 
# After the workflow runs, check the agent it created
kubectl get agent number-curiosa
# NAME              MODEL     READY
# number-curiosa    default   true

Screenshot:

Troubleshooting

When you get errors or unexpected results:

# List workflows
kubectl get workflows
 
# View logs
argo logs @latest