Orchestrator on OpenShift

  8 minute read  

Installing the Orchestrator is facilitated through an operator available in the Red Hat Catalog in the OLM package. This operator is responsible for installing all of the Orchestrator components. The Orchestrator is based on the SonataFlow and the Serverless Workflow technologies to design and manage the workflows. The Orchestrator plugins are deployed on a Red Hat Developer Hub instance, which serves as the frontend.

When installing a Red Hat Developer Hub (RHDH) instance using the Orchestrator operator, the RHDH configuration is managed through the Orchestrator resource.

To utilize Backstage capabilities, the Orchestrator imports software templates designed to ease the development of new workflows and offers an opinionated method for managing their lifecycle by including CI/CD resources as part of the template.

Orchestrator Documentation

For comprehensive documentation on the Orchestrator, please visit https://www.parodos.dev.

Installing the Orchestrator Helm Operator

Deploy the Orchestrator solution suite in an OCP cluster using the Orchestrator operator.
The operator installs the following components onto the target OpenShift cluster:

  • RHDH (Red Hat Developer Hub) Backstage
  • OpenShift Serverless Logic Operator (with Data-Index and Job Service)
  • OpenShift Serverless Operator
    • Knative Eventing
    • Knative Serving
  • (Optional) An ArgoCD project named orchestrator. Requires an pre-installed ArgoCD/OpenShift GitOps instance in the cluster. Disabled by default
  • (Optional) Tekton tasks and build pipeline. Requires an pre-installed Tekton/OpenShift Pipelines instance in the cluster. Disabled by default

Important Note for ARM64 Architecture Users

Note that as of November 6, 2023, OpenShift Serverless Operator is based on RHEL 8 images which are not supported on the ARM64 architecture. Consequently, deployment of this operator on an OpenShift Local cluster on MacBook laptops with M1/M2 chips is not supported.

Prerequisites

  • Logged in to a Red Hat OpenShift Container Platform (version 4.13+) cluster as a cluster administrator.
  • OpenShift CLI (oc) is installed.
  • Operator Lifecycle Manager (OLM) has been installed in your cluster.
  • Your cluster has a default storage class provisioned.
  • A GitHub API Token - to import items into the catalog, ensure you have a GITHUB_TOKEN with the necessary permissions as detailed here.
    • For classic token, include the following permissions:
      • repo (all)
      • admin:org (read:org)
      • user (read:user, user:email)
      • workflow (all) - required for using the software templates for creating workflows in GitHub
    • For Fine grained token:
      • Repository permissions: Read access to metadata, Read and Write access to actions, actions variables, administration, code, codespaces, commit statuses, environments, issues, pull requests, repository hooks, secrets, security events, and workflows.
      • Organization permissions: Read access to members, Read and Write access to organization administration, organization hooks, organization projects, and organization secrets.

Deployment with GitOps

If you plan to deploy in a GitOps environment, make sure you have installed the ArgoCD/Red Hat OpenShift GitOps and the Tekton/Red Hat Openshift Pipelines Install operators following these instructions. The Orchestrator installs RHDH and imports software templates designed for bootstrapping workflow development. These templates are crafted to ease the development lifecycle, including a Tekton pipeline to build workflow images and generate workflow K8s custom resources. Furthermore, ArgoCD is utilized to monitor any changes made to the workflow repository and to automatically trigger the Tekton pipelines as needed.

  • ArgoCD/OpenShift GitOps operator

    • Ensure at least one instance of ArgoCD exists in the designated namespace (referenced by ARGOCD_NAMESPACE environment variable). Example here
    • Validated API is argoproj.io/v1alpha1/AppProject

  • Tekton/OpenShift Pipelines operator

    • Validated APIs are tekton.dev/v1beta1/Task and tekton.dev/v1/Pipeline
    • Requires ArgoCD installed since the manifests are deployed in the same namespace as the ArgoCD instance.

    Remember to enable argocd and tekton in your CR instance.

Installation

  1. Deploy the PostgreSQL reference implementation for persistence support in SonataFlow following these instructions

  2. Create a namespace for the Orchestrator solution:

    oc new-project orchestrator

  3. Create a namespace for the Red Hat Developer Hub Operator (RHDH Operator):

    oc new-project rhdh-operator

  4. Download the setup script from the github repository and run it to create the RHDH secret and label the GitOps namespaces:

    wget https://raw.githubusercontent.com/parodos-dev/orchestrator-helm-operator/main/hack/setup.sh -O /tmp/setup.sh && chmod u+x /tmp/setup.sh

    Run the script:

    /tmp/setup.sh --use-default

    NOTE: If you don’t want to use the default values, omit the --use-default and the script will prompt you for input.

    The contents will vary depending on the configuration in the cluster. The following list details all the keys that can appear in the secret:

    • BACKEND_SECRET: Value is randomly generated at script execution. This is the only mandatory key required to be in the secret for the RHDH Operator to start.
    • K8S_CLUSTER_URL: The URL of the Kubernetes cluster is obtained dynamically using oc whoami --show-server.
    • K8S_CLUSTER_TOKEN: The value is obtained dynamically based on the provided namespace and service account.
    • GITHUB_TOKEN: This value is prompted from the user during script execution and is not predefined.
    • GITHUB_CLIENT_ID and GITHUB_CLIENT_SECRET: The value for both these fields are used to authenticate against GitHub. For more information open this link.
    • ARGOCD_URL: This value is dynamically obtained based on the first ArgoCD instance available.
    • ARGOCD_USERNAME: Default value is set to admin.
    • ARGOCD_PASSWORD: This value is dynamically obtained based on the first ArgoCD instance available.

    Keys will not be added to the secret if they have no values associated. So for instance, when deploying in a cluster without the GitOps operators, the ARGOCD_URL, ARGOCD_USERNAME and ARGOCD_PASSWORD keys will be omited in the secret.

    Sample of a secret created in a GitOps environment:

    $> oc get secret -n rhdh-operator -o yaml backstage-backend-auth-secret
apiVersion: v1
data:
  ARGOCD_PASSWORD: ...
  ARGOCD_URL: ...
  ARGOCD_USERNAME: ...
  BACKEND_SECRET: ...
  GITHUB_TOKEN: ...
  K8S_CLUSTER_TOKEN: ...
  K8S_CLUSTER_URL: ...
kind: Secret
metadata:
  creationTimestamp: "2024-05-07T22:22:59Z"
  name: backstage-backend-auth-secret
  namespace: rhdh-operator
  resourceVersion: "4402773"
  uid: 2042e741-346e-4f0e-9d15-1b5492bb9916
type: Opaque

  5. Use the following manifest to install the operator in an OCP cluster:

    apiVersion: operators.coreos.com/v1alpha1
kind: Subscription
metadata:
  name: orchestrator-operator
  namespace: openshift-operators
spec:
  channel: alpha
  installPlanApproval: Automatic
  name: orchestrator-operator
  source: redhat-operators
  sourceNamespace: openshift-marketplace

  6. Run the following commands to determine when the installation is completed:

    wget https://raw.githubusercontent.com/parodos-dev/orchestrator-helm-operator/main/hack/wait_for_operator_installed.sh -O /tmp/wait_for_operator_installed.sh && chmod u+x /tmp/wait_for_operator_installed.sh && /tmp/wait_for_operator_installed.sh

    During the installation process, Kubernetes cronjobs are created by the operator to monitor the lifecycle of the CRs managed by the operator: RHDH operator, OpenShift Serverless operator and OpenShift Serverless Logic operator. When deleting one of the previously mentioned CRs, a job is triggered that ensures the CR is removed before the operator is. In case of any failure at this stage, these jobs remain active, facilitating administrators in retrieving detailed diagnostic information to identify and address the cause of the failure.

    Note: that every minute on the clock a job is triggered to reconcile the CRs with the orchestrator resource values. These cronjobs are deleted when their respective features (e.g. rhdhOperator.enabled=false) are removed or when the orchestrator resource is removed. This is required because the CRs are not managed by helm due to the CRD dependency pre availability to the deployment of the CR.

  7. Apply the Orchestrator custom resource (CR) on the cluster to create an instance of RHDH and resources of OpenShift Serverless Operator and OpenShift Serverless Operator Logic. Make any changes to the CR before applying it, or test the default Orchestrator CR:

    oc apply -n orchestrator -f https://raw.githubusercontent.com/parodos-dev/orchestrator-helm-operator/refs/heads/main/config/samples/_v1alpha1_orchestrator.yaml

Additional information

Additional Workflow Namespaces

When deploying a workflow in a namespace different from where Sonataflow services are running (e.g., sonataflow-infra), several essential steps must be followed:

  1. Label the Workflow Namespace: To allow Sonataflow services to accept traffic from workflows, apply the following label to the desired workflow namespace:

       oc label ns $ADDITIONAL_NAMESPACE rhdh.redhat.com/workflow-namespace=""

  2. Identify the RHDH Namespace: Retrieve the namespace where RHDH is running by executing:

       oc get backstage -A

    Store the namespace value in RHDH_NAMESPACE in the Network Policy manifest below.

  3. Identify the Sonataflow Services Namespace: Check the namespace where Sonataflow services are deployed:

       oc get sonataflowclusterplatform -A

    If there is no cluster platform, check for a namespace-specific platform:

       oc get sonataflowplatform -A

    Store the namespace value in SONATAFLOW_PLATFORM_NAMESPACE.

  4. Set Up a Network Policy: Configure a network policy to allow traffic only between RHDH, Sonataflow services, and the workflows. The policy can be derived from here

    oc create -f <<EOF
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
  name: allow-rhdh-to-sonataflow-and-workflows
  # Sonataflow and Workflows are using the same namespace.
  namespace: $ADDITIONAL_NAMESPACE
spec:
  podSelector: {}
  ingress:
    - from:
      - namespaceSelector:
          matchLabels:
            # Allow RHDH namespace to communicate with workflows.
            kubernetes.io/metadata.name: $RHDH_NAMESPACE
      - namespaceSelector:
          matchLabels:
            # Allow Sonataflow services to communicate with workflows.
            kubernetes.io/metadata.name: $SONATAFLOW_PLATFORM_NAMESPACE
EOF

  5. Ensure Persistence for the Workflow: If persistence is required, follow these steps:

  • Create a PostgreSQL Secret: The workflow needs its own schema in PostgreSQL. Create a secret containing the PostgreSQL credentials in the workflow’s namespace:
    oc get secret sonataflow-psql-postgresql -n sonataflow-infra -o yaml > secret.yaml
sed -i '/namespace: sonataflow-infra/d' secret.yaml
oc apply -f secret.yaml -n $ADDITIONAL_NAMESPACE
  • Configure the Namespace Attribute: Add the namespace attribute under the serviceRef property where the PostgreSQL server is deployed.
    apiVersion: sonataflow.org/v1alpha08
kind: SonataFlow
  ...
spec:
  ...
  persistence:
    postgresql:
      secretRef:
        name: sonataflow-psql-postgresql
        passwordKey: postgres-password
        userKey: postgres-username
      serviceRef:
        databaseName: sonataflow
        databaseSchema: greeting
        name: sonataflow-psql-postgresql
        namespace: $POSTGRESQL_NAMESPACE
        port: 5432
    Replace POSTGRESQL_NAMESPACE with the namespace where the PostgreSQL server is deployed.

By following these steps, the workflow will have the necessary credentials to access PostgreSQL and will correctly reference the service in a different namespace.

GitOps environment

See the dedicated document

Deploying PostgreSQL reference implementation

See here

ArgoCD and workflow namespace

If you manually created the workflow namespaces (e.g., $WORKFLOW_NAMESPACE), run this command to add the required label that allows ArgoCD to deploy instances there:

oc label ns $WORKFLOW_NAMESPACE argocd.argoproj.io/managed-by=$ARGOCD_NAMESPACE

Workflow installation

Follow Workflows Installation

Cleanup

/!\ Before removing the orchestrator, make sure you have first removed any installed workflows. Otherwise the deletion may become hung in a terminating state.

To remove the operator from the cluster, delete the subscription:

oc delete subscriptions.operators.coreos.com orchestrator-operator -n openshift-operators

Note that the CRDs created during the installation process will remain in the cluster.

To clean the rest of the resources, run:

oc get crd -o name | grep -e sonataflow -e rhdh | xargs oc delete
oc delete namespace orchestrator sonataflow-infra rhdh-operator

If you want to remove knative related resources, you may also run:

oc get crd -o name | grep -e knative | xargs oc delete

Last modified September 25, 2024: Add main directory for main branch content (e35c6e8)