1. Periodic Retraining of Models Using Databricks Jobs


    To perform periodic retraining of models using Databricks Jobs, you would typically use a Databricks cluster to execute a notebook or script that re-trains the model. This can be orchestrated using the Databricks Jobs API, which allows you to schedule and run workflows on Databricks.

    A common approach to orchestrating periodic retraining involves creating a job that runs a Databricks notebook or script on a schedule, which could be daily, weekly, monthly, or any custom schedule that suits your needs.

    Below, I'll provide you with a Pulumi program written in Python that sets up a Databricks job that can be used for periodic retraining of models. The job will use a cluster and a notebook/script that you should have already created on Databricks.

    First, I'll define the job specs, including the cluster configuration and the schedule. Then, I'll show you how to set up a Databricks job using these specs with Pulumi and the Databricks provider.

    Here's the Pulumi program that accomplishes this:

    import pulumi import pulumi_databricks as databricks # Create a Databricks job for periodic retraining of models databricks_job = databricks.Job("retraining-job", name="Model Retraining Job", new_cluster=databricks.JobNewClusterArgs( spark_version="7.3.x-scala2.12", node_type_id="Standard_D3_v2", num_workers=2 ), notebook_task=databricks.JobNotebookTaskArgs( notebook_path="/Path/to/your/notebook", ), schedule=databricks.JobScheduleArgs( quartz_cron_expression="0 0 9 ? * MON", # This example sets the job to run every Monday at 9 AM. Adjust it to your needs. timezone_id="America/Los_Angeles" ), max_concurrent_runs=1 # Ensures that only one instance of the job runs at a time ) # Export the job URL so it can be accessed easily pulumi.export('job_url', pulumi.Output.concat("https://", databricks_job.workspace_url, "/?o=", databricks_job.workspace_id, "#job/", databricks_job.id))


    1. We import Pulumi and the Databricks provider.
    2. We create a databricks.Job resource with the necessary parameters:
      • name: A recognizable name for your job.
      • new_cluster: Here, we define a new cluster configuration that will be used when this job runs. Parameters such as spark_version, node_type_id, and num_workers define the cluster's properties.
      • notebook_task: This is where you specify the path to the notebook that contains the code for retraining your model.
      • schedule: Here, we use the Quartz Cron expression to define how often the job should run. In the example given, it's set to run every Monday at 9 AM in the timezone America/Los_Angeles. Modify this to your desired schedule.
      • max_concurrent_runs: This ensures that only one instance of the job runs at a time, which is often the desired behavior for retraining jobs to avoid resource conflicts and data mishandling.
    3. Finally, we export the job URL which incorporates the workspace_url, workspace_id, and id of the job resource, making it easy to navigate directly to the job in the Databricks workspace.

    Make sure to replace the /Path/to/your/notebook with the actual path to the notebook in your Databricks workspace that you intend to use for model retraining. Adjust the quartz_cron_expression to match the retraining schedule you need. Each field in the cron expression represents a different unit of time. The example given 0 0 9 ? * MON translates to 9 AM every Monday.

    For further modifications, you can tune the cluster specifications, add libraries, and control the number of retries upon failure among other settings that the Databricks Job resource allows.

    After running this Pulumi program, you'll have a Databricks job set up that will periodically run based on your specified schedule to retrain a model. If you wish to trigger the job outside of its schedule or need to stop it, you can do so from the Databricks workspace UI using the job URL exported by the program.