arazzo: 1.0.1
info:
  title: Apache Airflow Deployment Preflight Check
  summary: Verify scheduler and metadatabase health, capture the version, and surface DAG import errors and warnings before trusting a deployment.
  description: >-
    After a deploy — new DAG files, an upgraded image, a restarted scheduler — the
    question is whether Airflow is actually healthy and whether the DAGs parsed.
    This workflow answers both. It requires the scheduler and metadatabase to
    report healthy, records the version and git build for the change log, and then
    surfaces the two failure modes that a green health check hides: DAG files that
    raised an exception during parsing and therefore silently do not exist, and
    DAG warnings such as non-existent pool references. 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: airflowApi
  url: ../openapi/apache-airflow-openapi.yaml
  type: openapi
workflows:
- workflowId: deployment-preflight
  summary: Gate on Airflow health, then report import errors and DAG warnings.
  description: >-
    Requires healthy scheduler and metadatabase status, captures version
    information, lists import errors and DAG warnings, and counts the active DAGs
    the deployment is serving.
  inputs:
    type: object
    properties:
      errorLimit:
        type: integer
        description: How many import errors to return. Defaults to 100 when omitted.
  steps:
  - stepId: checkHealth
    description: >-
      Require both the metadatabase and the scheduler to report healthy. A
      scheduler whose latest heartbeat has gone stale reports unhealthy here,
      which is the single most useful signal that queued tasks will never start.
    operationId: get_health
    successCriteria:
    - condition: $statusCode == 200
    - context: $response.body
      condition: $.metadatabase.status == "healthy"
      type: jsonpath
    - context: $response.body
      condition: $.scheduler.status == "healthy"
      type: jsonpath
    outputs:
      metadatabaseStatus: $response.body#/metadatabase/status
      schedulerStatus: $response.body#/scheduler/status
      latestSchedulerHeartbeat: $response.body#/scheduler/latest_scheduler_heartbeat
  - stepId: readVersion
    description: >-
      Capture the running Airflow version and git build so the deployment record
      states exactly what is serving traffic.
    operationId: get_version
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      version: $response.body#/version
      gitVersion: $response.body#/git_version
  - stepId: listImportErrors
    description: >-
      List DAG files that failed to parse. This is the check a health endpoint
      cannot make: a DAG with a syntax or import error never registers, so it does
      not appear as broken — it simply is not there.
    operationId: get_import_errors
    parameters:
    - name: limit
      in: query
      value: $inputs.errorLimit
    - name: order_by
      in: query
      value: -timestamp
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      importErrors: $response.body#/import_errors
      importErrorCount: $response.body#/total_entries
  - stepId: listDagWarnings
    description: >-
      List DAG warnings, such as a DAG referencing a pool that does not exist.
      Note that this collection returns its items under an "import_errors"
      property rather than a warnings-specific name — an upstream quirk of the
      Airflow OpenAPI description that the payload genuinely carries.
    operationId: get_dag_warnings
    parameters:
    - name: limit
      in: query
      value: 100
    - name: order_by
      in: query
      value: -timestamp
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      dagWarnings: $response.body#/import_errors
      dagWarningCount: $response.body#/total_entries
  - stepId: countActiveDags
    description: >-
      Count the active DAGs the deployment is serving, giving the preflight a
      concrete number to compare against the expected inventory.
    operationId: get_dags
    parameters:
    - name: only_active
      in: query
      value: true
    - name: limit
      in: query
      value: 100
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      dags: $response.body#/dags
      activeDagCount: $response.body#/total_entries
  outputs:
    schedulerStatus: $steps.checkHealth.outputs.schedulerStatus
    metadatabaseStatus: $steps.checkHealth.outputs.metadatabaseStatus
    version: $steps.readVersion.outputs.version
    importErrorCount: $steps.listImportErrors.outputs.importErrorCount
    importErrors: $steps.listImportErrors.outputs.importErrors
    dagWarningCount: $steps.listDagWarnings.outputs.dagWarningCount
    activeDagCount: $steps.countActiveDags.outputs.activeDagCount