Spark Pi Example

Spark Pi is a sample project to trigger a short lived Spark CI/CD workflow on Banzai Cloud's Pipeline PaaS platform.

Prerequisites

• Account on one of the following cloud providers: Amazon, Azure, GoogleCloud
• A Banzai Cloud Control Plane needs to be running and accessible
• Oauth2 authentication enabled on the Banzai Cloud Control Plane

Setup

• fork the project into your repository
• make a copy of the flow descriptor template corresponding to your chosen cloud provider:

cp .pipeline.yml.[azure|aws|gke].template .pipeline.yml

• check the flow descriptor file and replace the placeholders with your specific values (cluster name, bucket / folder names etc ...)

In short: the CI/CD flow descriptor file contains the steps needed from provisioning the environment, building the code, running tests to being deployed and executed along with project specific variables (eg.:credentials, program arguments, etc ...).

• navigate to the CI/CD user interface (that usually runs on the Banzai Cloud Control plane instance)
• enable the project build from the list of available repositories
• add the following secrets to the build:

PLUGIN_ENDPOINT = [control-plane]/pipeline/api/v1
PLUGIN_TOKEN = "oauthToken"

The project is configured now for the Banzai Cloud CI/CD flow. On each commit to the repository a new flow will be triggered. You can check the progress on the CI/CD user interface.

Banzai Cloud CI/CD flow descriptor templates

All the templates describe the following CI/CD flow:

• Create a Kubernetes cluster
• Set up monitoring on the cluster (optional, you can skip / remove this part from the template)
• Install spark resources
• Checkout the project
• Build the SparkPi project (the fork of this project)
• Run the SparkPi application (issue a spark-submit command to the above set up environment)

The templates only differ in the steps related to setting up the environment and using the cloud resources (storage, buckets, folders).

Eg.: the create_cluster step:

[Amazon]

create_cluster:
    image: banzaicloud/plugin-pipeline-client:0.3.0
    cluster_name: "[[your-cluster-name]]"
    cluster_provider: "amazon"

    secrets: [plugin_endpoint, plugin_token]

[Azure]

create_cluster:
    image: banzaicloud/plugin-pipeline-client:0.3.0
    cluster_name: "[[your-cluster-name]]"
    cluster_provider: "azure"
    azure_resource_group: "[[az_resource_group]]"

    secrets: [plugin_endpoint, plugin_token]

[Google Cloud]

create_cluster:
    image: banzaicloud/plugin-pipeline-client:0.3.0
    cluster_name: "[[your-cluster-name]]"
    cluster_provider: "google"
    google_project: "[[google-project]]"

    secrets: [plugin_endpoint, plugin_token]

The create_cluster step listed above can be further customised by providing more details like the cluster location, instance type etc... (more details here)

The other major difference is in the configuration related to the persistent storage used by the spark (to write logs to) and the Spark History Server (to read from)

[Amazon]

install_spark_resources:
    image: banzaicloud/plugin-pipeline-client:0.3.0

    deployment_name: "banzaicloud-stable/spark"
    deployment_values:
      historyServer:
        enabled: true
      spark-hs:
        app:
          logDirectory: "s3a://[[your-s3-bucket]]/"

    secrets: [plugin_endpoint, plugin_token]

[Azure]

install_spark_resources:
    image: banzaicloud/plugin-pipeline-client:0.3.0

    deployment_name: "banzaicloud-stable/spark"
    deployment_values:
      historyServer:
        enabled: true
      spark-hs:
        app:
          logDirectory: "wasb://[your-blob-container]]@{{ .PLUGIN_AZURE_STORAGE_ACCOUNT }}.blob.core.windows.net/"
          azureStorageAccountName: "{{ .PLUGIN_AZURE_STORAGE_ACCOUNT }}"
          azureStorageAccountName: "{{ .PLUGIN_AZURE_STORAGE_ACCOUNT_ACCESS_KEY }}"

    secrets: [plugin_endpoint, plugin_token,plugin_azure_storage_account, plugin_azure_storage_account_access_key]

[Google Cloud]

install_spark_resources:
    image: banzaicloud/plugin-pipeline-client:0.3.0

    deployment_name: "banzaicloud-stable/spark"
    deployment_values:
     historyServer:
        enabled: true
     spark-hs:
       app:
         logDirectory: "gs://[[your-gs-bucket]]/"

    secrets: [plugin_endpoint, plugin_token]

Note: currently the storage resources need to be set up manually in case of all cloud providers. In case of azure the storage location is dynamically assembled using the azure secrets.