DatabricksCreateJobsOperator¶

Use the DatabricksCreateJobsOperator to create (or reset) a Databricks job. This operator relies on past XComs to remember the job_id that was created so that repeated calls with this operator will update the existing job rather than creating new ones. When paired with the DatabricksRunNowOperator all runs will fall under the same job within the Databricks UI.

Using the Operator¶

There are three ways to instantiate this operator. In the first way, you can take the JSON payload that you typically use to call the api/2.2/jobs/create endpoint and pass it directly to our DatabricksCreateJobsOperator through the json parameter. With this approach you get full control over the underlying payload to Jobs REST API, including execution of Databricks jobs with multiple tasks, but it’s harder to detect errors because of the lack of the type checking.

The second way to accomplish the same thing is to use the named parameters of the DatabricksCreateJobsOperator directly. Note that there is exactly one named parameter for each top level parameter in the api/2.2/jobs/create endpoint.

The third way is to use both the json parameter AND the named parameters. They will be merged together. If there are conflicts during the merge, the named parameters will take precedence and override the top level json keys.

Currently the named parameters that DatabricksCreateJobsOperator supports are:

name
description
tags
tasks
job_clusters
email_notifications
webhook_notifications
notification_settings
timeout_seconds
schedule
max_concurrent_runs
git_source
access_control_list

Forwarding Airflow Dag params as Databricks job parameters¶

The Databricks api/2.2/jobs/create endpoint accepts a top-level parameters field that defines job-level parameters — a list of objects with a name (the parameter name) and a default (its default value), for example [{"name": "env", "default": "prod"}].

If parameters is not set in json and the operator’s params dict is non-empty, each key/value pair in params is converted into one such {"name": key, "default": value} entry, so that Airflow Dag params can be forwarded as Databricks job parameters without hardcoding the API shape in json. If json already contains parameters, it is left untouched.

# Airflow Dag params (key/value pairs)
params = {"env": "prod", "batch_size": "100"}

create_job = DatabricksCreateJobsOperator(
    task_id="create_job",
    json={"name": "my-job", "tasks": [...]},
    params=params,
)

# The Databricks job created/reset by the operator will have:
#   parameters=[
#       {"name": "env",        "default": "prod"},
#       {"name": "batch_size", "default": "100"},
#   ]
# i.e. each "<key>: <value>" in params becomes one
# {"name": "<key>", "default": "<value>"} entry in the job definition.

Examples¶

Specifying parameters as JSON¶

An example usage of the DatabricksCreateJobsOperator is as follows:

tests/system/databricks/example_databricks.py[source]

    # Example of using the JSON parameter to initialize the operator.
    job = {
        "tasks": [
            {
                "task_key": "test",
                "job_cluster_key": "job_cluster",
                "notebook_task": {
                    "notebook_path": "/Shared/test",
                },
            },
        ],
        "job_clusters": [
            {
                "job_cluster_key": "job_cluster",
                "new_cluster": {
                    "spark_version": "7.3.x-scala2.12",
                    "node_type_id": "i3.xlarge",
                    "num_workers": 2,
                },
            },
        ],
    }

    jobs_create_json = DatabricksCreateJobsOperator(task_id="jobs_create_json", json=job)

Using named parameters¶

You can also use named parameters to initialize the operator and run the job.

tests/system/databricks/example_databricks.py[source]

    # Example of using the named parameters to initialize the operator.
    tasks = [
        {
            "task_key": "test",
            "job_cluster_key": "job_cluster",
            "notebook_task": {
                "notebook_path": "/Shared/test",
            },
        },
    ]
    job_clusters = [
        {
            "job_cluster_key": "job_cluster",
            "new_cluster": {
                "spark_version": "7.3.x-scala2.12",
                "node_type_id": "i3.xlarge",
                "num_workers": 2,
            },
        },
    ]

    jobs_create_named = DatabricksCreateJobsOperator(
        task_id="jobs_create_named", tasks=tasks, job_clusters=job_clusters
    )

Pairing with DatabricksRunNowOperator¶

You can use the job_id that is returned by the DatabricksCreateJobsOperator in the return_value XCom as an argument to the DatabricksRunNowOperator to run the job.

tests/system/databricks/example_databricks.py[source]

    # Example of using the DatabricksRunNowOperator after creating a job with DatabricksCreateJobsOperator.
    run_now = DatabricksRunNowOperator(task_id="run_now", job_id=jobs_create_named.output)

    jobs_create_named >> run_now