arazzo: 1.0.1
info:
  title: Azure Databricks Provision a Cluster and Run a Job on It
  summary: Create a cluster, wait until RUNNING, create a job on it, then run it.
  description: >-
    Stands up dedicated compute and immediately runs a notebook job against it.
    The workflow creates a cluster, polls until the cluster state is RUNNING,
    creates a notebook job bound to that cluster, triggers a run, and polls the
    run until its life cycle state is TERMINATED. Every step spells out its
    request inline so the flow can be read and executed without opening the
    underlying OpenAPI description.
  version: 1.0.0
sourceDescriptions:
- name: azureDatabricksApi
  url: ../openapi/azure-databricks-openapi.yml
  type: openapi
workflows:
- workflowId: provision-cluster-and-run-job
  summary: Create a cluster, wait for RUNNING, then create and run a notebook job.
  description: >-
    Chains createCluster, a getCluster poll to RUNNING, createJob bound to the
    new cluster, runJobNow, and a getJobRun poll to TERMINATED.
  inputs:
    type: object
    required:
    - token
    - clusterName
    - sparkVersion
    - nodeTypeId
    - numWorkers
    - jobName
    - taskKey
    - notebookPath
    properties:
      token:
        type: string
        description: Databricks personal access token for the Authorization header.
      clusterName:
        type: string
        description: Name for the new cluster.
      sparkVersion:
        type: string
        description: Databricks Runtime version key for the cluster.
      nodeTypeId:
        type: string
        description: Azure VM node type id for the cluster.
      numWorkers:
        type: integer
        description: Number of worker nodes for the cluster.
      jobName:
        type: string
        description: Name for the job to create.
      taskKey:
        type: string
        description: Unique key identifying the single task within the job.
      notebookPath:
        type: string
        description: Absolute workspace path of the notebook to run.
  steps:
  - stepId: createCluster
    description: >-
      Create the cluster that will host the job. The cluster starts in PENDING
      and the cluster_id is returned.
    operationId: createCluster
    parameters:
    - name: Authorization
      in: header
      value: Bearer $inputs.token
    requestBody:
      contentType: application/json
      payload:
        cluster_name: $inputs.clusterName
        spark_version: $inputs.sparkVersion
        node_type_id: $inputs.nodeTypeId
        num_workers: $inputs.numWorkers
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      clusterId: $response.body#/cluster_id
  - stepId: pollCluster
    description: >-
      Retrieve the cluster state. Repeat until the cluster is RUNNING, then
      create the job; branch to failure if it terminates.
    operationId: getCluster
    parameters:
    - name: Authorization
      in: header
      value: Bearer $inputs.token
    - name: cluster_id
      in: query
      value: $steps.createCluster.outputs.clusterId
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      state: $response.body#/state
    onSuccess:
    - name: running
      type: goto
      stepId: createJob
      criteria:
      - context: $response.body
        condition: $.state == "RUNNING"
        type: jsonpath
    - name: failed
      type: end
      criteria:
      - context: $response.body
        condition: $.state == "TERMINATED" || $.state == "ERROR"
        type: jsonpath
    - name: pending
      type: goto
      stepId: pollCluster
      criteria:
      - context: $response.body
        condition: $.state == "PENDING" || $.state == "RESTARTING"
        type: jsonpath
  - stepId: createJob
    description: >-
      Create a notebook job bound to the newly provisioned cluster and capture
      the job_id.
    operationId: createJob
    parameters:
    - name: Authorization
      in: header
      value: Bearer $inputs.token
    requestBody:
      contentType: application/json
      payload:
        name: $inputs.jobName
        tasks:
        - task_key: $inputs.taskKey
          existing_cluster_id: $steps.createCluster.outputs.clusterId
          notebook_task:
            notebook_path: $inputs.notebookPath
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      jobId: $response.body#/job_id
  - stepId: runJobNow
    description: >-
      Trigger an immediate run of the job and capture the run_id for polling.
    operationId: runJobNow
    parameters:
    - name: Authorization
      in: header
      value: Bearer $inputs.token
    requestBody:
      contentType: application/json
      payload:
        job_id: $steps.createJob.outputs.jobId
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      runId: $response.body#/run_id
  - stepId: pollRun
    description: >-
      Retrieve the run state. Repeat until the run life_cycle_state is
      TERMINATED, then end with the final result_state.
    operationId: getJobRun
    parameters:
    - name: Authorization
      in: header
      value: Bearer $inputs.token
    - name: run_id
      in: query
      value: $steps.runJobNow.outputs.runId
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      lifeCycleState: $response.body#/state/life_cycle_state
      resultState: $response.body#/state/result_state
    onSuccess:
    - name: finished
      type: end
      criteria:
      - context: $response.body
        condition: $.state.life_cycle_state == "TERMINATED" || $.state.life_cycle_state == "INTERNAL_ERROR"
        type: jsonpath
    - name: stillRunning
      type: goto
      stepId: pollRun
      criteria:
      - context: $response.body
        condition: $.state.life_cycle_state == "PENDING" || $.state.life_cycle_state == "RUNNING" || $.state.life_cycle_state == "TERMINATING"
        type: jsonpath
  outputs:
    clusterId: $steps.createCluster.outputs.clusterId
    jobId: $steps.createJob.outputs.jobId
    runId: $steps.runJobNow.outputs.runId
    resultState: $steps.pollRun.outputs.resultState