Skip to main content

Deploy to Databricks Private Preview

On Databricks, deploying is part of the build cycle, not the end of it. Chat & build and Simulate both run on a live Genie Space, so you deploy first to create it, then refine and simulate, then certify and redeploy for production.

CES pushes descriptions into Unity Catalog, creates one Metric View per table, and configures a Genie Space backed by those views. Re-deploys replace existing Metric Views in place and patch the Genie Space without recreating it.

Prerequisites​

Before you begin, make sure:

Deploy your context repository​

Deploying creates the Metric Views and Genie Space in your Databricks workspace. You don't need to certify before this first deploy. Once the Genie Space exists, Chat & build and Simulate are available for iteration.

  1. From your context repository, click Deploy.

  2. Confirm the target catalog and schema where Metric Views are created. These default to the values configured on the Configure page.

  3. Enter a name prefix for the Metric Views, or accept the default derived from your repository name. CES creates one Metric View per table, each named <prefix>_<table_alias>. If Metric Views with the same names already exist, CES replaces them.

  4. Select the SQL warehouse that Genie uses to run queries. It must be a pro or serverless warehouse running Databricks Runtime 17.3 or later.

  5. Click Confirm.

After deployment, Metric Views are available at <target_catalog>.<target_schema>.<prefix>_<table_alias> and the Genie Space is linked and ready for use.

Once deployed, Chat & build and Simulate are active. Use them to iterate on the model until it answers your target questions correctly. See Refine with Chat & build and Run simulations.

Certify your context repository​

Certifying marks the model as ready for production. It creates a versioned snapshot at that point in time, so the deployed version is always traceable to a specific certified state. To make changes after certifying, you create a new draft from the certified model, update it, and certify again.

  1. From your context repository, click Certify.
  2. Add a certification note describing what the repository covers and any known limitations.
  3. Click Confirm. The repository moves from Draft to Active.

After certifying, click Deploy and follow the same deploy steps to push the certified model to production.

Grant access and verify​

After the production deploy, grant end users access and confirm the artifacts are visible in Databricks.

  1. Grant your business users access to the Metric Views. Run in a Databricks SQL editor, replacing <consumer_group> with the group assigned to your users.

    GRANT USE CATALOG ON CATALOG <target_catalog> TO `<consumer_group>`;
    GRANT USE SCHEMA ON SCHEMA <target_catalog>.<target_schema> TO `<consumer_group>`;
    GRANT SELECT ON VIEW <target_catalog>.<target_schema>.<metric_view_name> TO `<consumer_group>`;

  2. Grant access to the Genie Space via the Databricks workspace UI.

    • Open the Genie Space in your Databricks workspace.
    • Go to Settings → Permissions.
    • Add end-user groups with CAN RUN (can ask questions) or CAN VIEW (can view conversations only).
    • Confirm end users have consumer access or the Databricks SQL workspace entitlement.

    End users don't need direct warehouse permissions. Queries run under the creator's embedded compute credentials, and Genie respects Unity Catalog row and column security.

  3. Verify the deployment by confirming in the Databricks workspace that:

    • The Metric Views appear in Unity Catalog under <target_catalog>.<target_schema>.
    • The Genie Space is listed under Genie in the workspace sidebar and is accessible.

    You can also navigate to the deployed Metric Views in the Atlan catalog. CES crawls them automatically after deployment.

Update deployed context repositories​

After a repository is live, use this process to make changes without disrupting end users. The existing Metric Views and Genie Space stay active until the new version is certified and deployed.

  1. From your context repository, click Edit. CES creates a new draft from the current certified model.
  2. Make your changes in the YAML editor or via Chat & build.
  3. Re-run your question set through Run simulations to confirm nothing has regressed.
  4. Certify the updated model and click Deploy. CES replaces the existing Metric Views and patches the Genie Space in place. All previous certification versions are retained in the repository history.