Databricks Intelligence Platform
Databricks Intelligence Platform
This domain checks whether you understand how Azure Databricks is assembled: the account and workspace model, control plane versus compute plane, the lakehouse services, and where Unity Catalog fits. Expect scenario questions about choosing compute, locating data, and administering the platform.
Account, workspace, and plane boundaries
- An Azure Databricks account is the billing and identity container. Account admins manage Unity Catalog metastores, users/groups, and workspace assignment. A workspace is the regional UI/API surface where engineers create notebooks, jobs, and clusters.
- Classic compute runs in your Azure subscription and VNet; serverless compute runs in a Databricks-managed plane. Data remains in your ADLS Gen2 storage.
- One Unity Catalog metastore per Azure region can be shared by workspaces in that region. A workspace attached to a metastore gets the three-level namespace
catalog.schema.table.
Compute and workspace objects
- All-purpose clusters are interactive and shared; job clusters are created for a run and terminate after; SQL warehouses serve Databricks SQL and dashboards. Cluster policies constrain size, tags, and runtime.
- Pools pre-provision idle nodes to cut start time; autoscaling changes workers, not driver type.
- Delta Lake is the default table format; managed tables use UC-managed storage, external tables point to an external location, and volumes govern non-tabular files.
- Workspace objects include notebooks, files, Git folders, queries, dashboards, alerts, jobs, and Lakeflow pipelines. Billing is in DBUs by SKU and compute.
Worked example
A team has two workspaces in West Europe and needs one governed table readable from both, plus a nightly job that must not be edited interactively. Reasoning: because both workspaces are in the same region, attach them to the same UC metastore; create a catalog and schema, then a managed or external table. For the nightly run, use a job cluster (or serverless job compute) with a job-owned schedule, not an all-purpose cluster. If the table must live in a specific ADLS container, create a storage credential and external location first.
Traps
- Confusing account admin with workspace admin; UC metastore admin is separate.
- Attaching a workspace to a metastore in another region.
- Using an all-purpose cluster for scheduled production jobs.
- Assuming DBFS root is governed by UC; it is not a UC volume.
- Treating serverless and classic compute as interchangeable for network-bound sources.