This post demonstrates how to use Argo Events and Argo Workflows to achieve event driven workflow execution. Event based architectures are a key part of building solutions where the individual components are decoupled, the line of responsibility for the action which created an event stops at point of generating the event, what happens next is a problem for another system.
This section provides a quick overview of the core components for Argo Events and Argo Workflows. All of these components are Kubernetes resources which are consumed to deploy and configure additional resources on the cluster such as deployments.
eventsources.argoproj.io resource specifies how to consume events from external services such as; GitHub, Slack and Webhooks. The resource also specifies criteria to compare observed events against.
The Event Source process runs within a pod managed by the
eventsource-controller. The process writes to the eventbus when it observes events which match the filtering criteria.
A service account with
[List, Watch] permissions is required for the Event Source to monitor Kubernetes resources.
An Event Bus is a transport service for events to go from an Event Source to a Sensor. The Event Bus process runs in a cluster of three pods managed by the
sensors.argoproj.io resource specifies the events to look for on the Event Bus and the response to trigger when a matching event is observed.
The sensor process runs in a pod managed by the
There are currently 10 predefined triggers for services such as; HTTP requests, Kubernetes Objects and Slack. Additionally, you can create your own triggers as needed.
A service account with sufficient permissions is required if the trigger manipulates Kubernetes Objects.
workflows/argoproj.io/v1alpha1 resource defines a task to be executed by a workflow and the intent to execute. Workflow execution is immediate and a once off event. Use the
cronWorkflows/argoproj.io/v1alpha1 resource to create workflows with delayed execution or execute on a recurring schedule.
The Workflow Controller watches the Kubernetes API server for
cronWorkflows/argoproj.io/v1alpha1 resources. You specify the controllers scope to either it’s own namespace or the entire cluster using the Argo Workflows deployment manifest. A cluster wide scope is required for this integration to work as the Workflow Controller needs to observe the creation of
workflows/argoproj.io/v1alpha1 resources in the argo-events namespace.
A workflows process runs inside a pod called the
executor pod which is created by the Workflow Controller. The executor pod authenticates against with the Kubernetes API server using the service account specified in the pods configuration. The
default service account of the namespace which the executor pod is running in is used if the workflows
spec.serviceAccountName field is empty.
Our use case for this example is to allow uses to deploy a workload based from an external system. The external system provides the users with a request form, once submitted the system sends a HTTP request causing deployment of the workload.
Our task is to deploy a solution which deploys the workflow when the HTTP request is received.
The diagram below is the user level view of our example workflow, they submit a request and a workload is deployed.
This diagram details the data and process flow or our example workflow.
The below diagram is a high level diagram showing where the Argo Event and Argo Workflow service processes run at a namespace level. This diagram can be built on top of to show more detailed and specific information.
The diagram also tells us that the Workflow Controller must be able to read
workflows/argoproj.io/v1alpha1 resources from other namespaces, therefore we must deploy Argo Workflows with a cluster wide scope.
Argo Workflow Deployment
The steps below detail how to deploy Argo Workflows.
After the application is running, if you’re on using a local cluster like Minikube run this command so you can access the web interface.
The web interface address is https://localhost:2746
Argo Events Deployment
The commands below deploy Argo Events controllers and the Event Bus service.
Our workflow needs to be able to receive HTTP Requests from the external system, this is achieved using an WebHook EventSource. Manifest Source
This manifest tells the
- Create a deployment with port 12000 exposed
- Create a service which forwards port 12000 to the Event Source pod on port 12000
Write events matching the follow criteria to the EventBus:
- Received on port 12000
- URI endpoint is /example
- Method is POST
- Any server URL
At this stage our solution is able to receive a HTTP request and write to the EventBus.
Our integration example manipulates Kubernetes resources, therefore we need to create service accounts with the correct access rights to succeed. We need to understand how Argo Events and Argo Workflows behaves in relation to our use case.
Sensor Service Account
The sensor process runs within a pod in the
argo-events namespace and the trigger creates a
workflows/argoproj.io/v1alpha1 resource. The resource spec in the workflow resource does not specify which namespace to create the resource in, therefore it will be created in the
The sensor requires a service account with permission to create
workflows/argoproj.io/v1alpha1 resources in the
argo-events namespace. Manifest Source
Workflow Service Account
The executor pod will be created in the
argo-events namespace because that is where the
workflows/argoproj.io/v1alpha1 resource resides.
The workflow process within the executor pod requires permissions to create a pod (the example workload) in the
Below is the manifest for the service account used by the executor pod and the role and role bindings. Manifest Source
The manifest below specifies the configuration for our Sensor. Manifest source
Apply the Sensor resource.
End to end testing
All the pieces should be in place so our users can execute a workflow from an external system. Each item has been tested individually and we now need to validate the end to end process by sending a HTTP Request to the event source using curl.
The below images are from the Argo Workflows UI.
Technically Argo Events and Argo Workflows do not integrate with each other so it would be more accurate to say that they work well together. Argo Events and Argo Workflows are two independent products, each addressing their specific use cases.
The process of getting Argo Events and Argo Workflows working together wasn’t that difficult, my biggest stumbling points were RBAC and understanding how the components worked together.