arazzo: 1.0.1
info:
  title: Autodesk Evaluate a Bifrost Graph in the Cloud
  summary: Upload a Bifrost graph, queue an evaluation job with inputs, poll it, and collect the outputs.
  description: >-
    Offloading a Bifrost graph evaluation to the Flow Graph Engine so an effects,
    simulation, or procedural content task runs on cloud hardware instead of an
    artist's workstation. The workflow uploads the .bif graph, reads back the input
    ports it declares, queues a job binding values to those ports, polls until the job
    reaches a terminal state, and collects the outputs. 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: flowGraphEngineApi
  url: ../openapi/autodesk-flow-graph-engine-openapi.yml
  type: openapi
workflows:
- workflowId: evaluate-bifrost-graph
  summary: Upload a graph, run it against inputs, and collect the result.
  description: >-
    Uploads a Bifrost graph file, queues an evaluation job with input parameter values
    matching the graph's declared input ports, waits for the job to succeed, and
    retrieves the outputs and any binary download link.
  inputs:
    type: object
    required:
    - graphFile
    - graphName
    - graphInputs
    properties:
      graphFile:
        type: string
        description: The Bifrost graph file (.bif) to upload.
      graphName:
        type: string
        description: Display name for the uploaded graph.
      graphDescription:
        type: string
        description: Description of what the graph computes.
      graphInputs:
        type: object
        description: >-
          Input parameter values keyed by the graph's input port names, as reported by
          the upload step.
      callbackUrl:
        type: string
        description: Optional URL notified when the job completes.
  steps:
  - stepId: uploadBifrostGraph
    description: >-
      Upload the Bifrost graph. The response declares the graph's input and output
      ports, which is what the job's inputs must be keyed by.
    operationId: uploadGraph
    requestBody:
      contentType: multipart/form-data
      payload:
        file: $inputs.graphFile
        name: $inputs.graphName
        description: $inputs.graphDescription
    successCriteria:
    - condition: $statusCode == 201
    outputs:
      graphId: $response.body#/id
      inputPorts: $response.body#/inputs
      outputPorts: $response.body#/outputs
      fileSize: $response.body#/fileSize
  - stepId: queueEvaluation
    description: >-
      Queue an evaluation job against the uploaded graph, binding values to its input
      ports. A callback URL avoids the polling loop entirely for callers that can
      receive one.
    operationId: createJob
    requestBody:
      contentType: application/json
      payload:
        graphId: $steps.uploadBifrostGraph.outputs.graphId
        inputs: $inputs.graphInputs
        callbackUrl: $inputs.callbackUrl
    successCriteria:
    - condition: $statusCode == 201
    outputs:
      jobId: $response.body#/id
      status: $response.body#/status
      createdAt: $response.body#/createdAt
  - stepId: awaitEvaluation
    description: >-
      Poll the job until it succeeds. The endpoint returns 200 while queued or
      running, so the terminal state is read from the body, and the workflow ends with
      the error message if the evaluation fails or is cancelled.
    operationId: getJob
    parameters:
    - name: jobId
      in: path
      value: $steps.queueEvaluation.outputs.jobId
    successCriteria:
    - condition: $statusCode == 200
    - context: $response.body
      condition: $.status == "succeeded"
      type: jsonpath
    onFailure:
    - name: stillEvaluating
      type: retry
      retryAfter: 15
      retryLimit: 60
      criteria:
      - context: $response.body
        condition: $.status == "queued" || $.status == "running"
        type: jsonpath
    - name: evaluationFailed
      type: end
      criteria:
      - context: $response.body
        condition: $.status == "failed" || $.status == "cancelled"
        type: jsonpath
    outputs:
      status: $response.body#/status
      progress: $response.body#/progress
      completedAt: $response.body#/completedAt
      error: $response.body#/error
  - stepId: collectResult
    description: >-
      Retrieve the evaluation outputs. Returns 404 if called before the job reaches a
      completed state, which the poll above guarantees it has.
    operationId: getJobResult
    parameters:
    - name: jobId
      in: path
      value: $steps.queueEvaluation.outputs.jobId
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      jobId: $response.body#/jobId
      outputs: $response.body#/outputs
      downloadUrl: $response.body#/downloadUrl
  outputs:
    graphId: $steps.uploadBifrostGraph.outputs.graphId
    jobId: $steps.queueEvaluation.outputs.jobId
    status: $steps.awaitEvaluation.outputs.status
    downloadUrl: $steps.collectResult.outputs.downloadUrl