Skip to main content

Install Lakemeter from Databricks Marketplace

Lakemeter can be installed directly from Databricks Marketplace as a managed Databricks App. The Marketplace package includes the application and its pricing reference data, so no local CLI, installer notebook, personal access token, or separate data upload is required.

Preferred installation method

Installing from Databricks Marketplace is the preferred way to install Lakemeter. Use the source installer only when Marketplace is unavailable in your workspace or you need to operate a customized source deployment.

Prerequisites​

Before starting:

  • A workspace administrator must perform the installation. Workspace administrators can install apps from Marketplace.
  • Create a new, empty Lakebase database in the same Databricks workspace where Lakemeter will be installed. The default Lakebase configuration and default databricks_postgres database are supported.
  • Have a running Claude Model Serving endpoint available in the same workspace.

The administrator does not need to grant the app permissions before starting. During installation, Databricks grants the new app identity access to the selected Lakebase database and Model Serving endpoint.

Lakebase database​

Only an empty Lakebase database in the same workspace is required. Do not create tables or load pricing data manually. Lakemeter creates its tables, calculation functions, and reference data automatically when the app first starts.

Use a database without an existing Lakemeter schema

Each installation receives its own Databricks App service principal, which creates and owns the lakemeter schema. Do not select a database containing a lakemeter schema created by another app identity. Reusing that schema can cause a permission denied for schema lakemeter startup error.

Claude Model Serving endpoint​

Any running Claude endpoint available for the intended users' use can be selected, provided it supports the Databricks chat-completions request format.

During installation, Lakemeter binds the selected endpoint as claude-endpoint with CAN_QUERY. The app service principal uses this permission when a user invokes the AI assistant. Users do not provide endpoint credentials to Lakemeter.

Check the endpoint's availability and rate limits before installation. A throttled endpoint affects the optional AI assistant but not manual estimates, calculations, or Excel export.

Installation flow​

1. Find Lakemeter in Marketplace​

In the Databricks workspace, open Marketplace, search for Lakemeter, and select Lakemeter - Databricks Cost Estimation App.

Find Lakemeter in Databricks Marketplace

Select Install on the listing page.

2. Configure the app resources​

Under Database, select the new Lakebase project, writable branch, and empty database prepared for this installation.

Under Serving endpoint, select the prepared Claude endpoint. Keep the default app compute configuration unless your organization requires a different size or instance count.

Configure the Lakebase database and Claude endpoint

During installation, Databricks grants the app:

  • CAN_CONNECT_AND_CREATE on the selected Lakebase database.
  • CAN_QUERY on the selected Model Serving endpoint.

Select Next.

3. Review authorizations​

Review the permissions shown by Databricks and confirm that the selected Lakebase database and Model Serving endpoint are the intended resources. Continue only after the authorization review is complete.

4. Add metadata and install​

Enter an app name and description. The app name cannot be changed after the app is created. Select the serverless usage policy required by your organization, then select Install.

Add the app metadata and install Lakemeter

5. Wait for the app to start​

Databricks creates the app identity, grants the two bound resources, deploys the Marketplace package, and starts the app. On first start, Lakemeter automatically:

  1. Connects to Lakebase with a short-lived OAuth database credential.
  2. Creates the lakemeter schema and application tables.
  3. Installs the calculation functions and reference records.
  4. Loads the packaged pricing data, including cloud VM pricing.
  5. Records the installed application version and pricing checksum.

Wait until the app status is Running and the active deployment is Succeeded, then select Open app. No additional bootstrap or pricing data step is required.

6. Grant users access to Lakemeter​

After installation, open the Lakemeter app's Permissions page:

  1. Select Add user, group, or service principal.
  2. Select the users or workspace groups that should use Lakemeter.
  3. Grant CAN USE.

Only users with permission to use the app can open its URL. Prefer granting access to a workspace group when Lakemeter will be used by a team.

Troubleshooting​

App does not start​

Open the app's Deployments or Logs page and review the latest startup error. Confirm that the selected Lakebase database and Model Serving endpoint still exist in the same workspace and that the installation completed its resource grants.

Permission denied for schema lakemeter​

The selected database likely contains a schema owned by another Lakemeter app identity. Bind a new empty database to the app, or migrate the schema ownership and grants deliberately if existing data must be preserved.

Lakebase connection timeout​

A scale-to-zero Lakebase endpoint can take time to wake. Lakemeter retries transient connection timeouts during startup. If the retries continue to fail, confirm that the Lakebase endpoint is available and review the app deployment logs.

AI assistant is unavailable​

Confirm that the selected Claude endpoint is running and that the app retains CAN_QUERY. Model Serving rate limits can affect the AI assistant without affecting the core estimate workflow.

Another user cannot open the app​

On the app's Permissions page, grant the user or one of their workspace groups CAN USE. Installing Lakemeter does not automatically grant every workspace user access to it.

Upgrade a Marketplace installation​

When the provider publishes a new Lakemeter version, Databricks displays an update on the installed app's page. A workspace administrator reviews and applies the update from Databricks Marketplace.

Marketplace upgrades preserve the existing app identity, resource bindings, permissions, and Lakebase data. Lakemeter runs its idempotent bootstrap after deployment to apply compatible packaged updates. Users do not need to reinstall the app or recreate their estimates.

If a release changes its required resources or permissions, Databricks shows those changes for administrator review before the update is applied.