Databricks Certified Data Engineer Associate Practice Questions: Databricks Lakehouse Platform Domain

Delta Lake time travel allows querying and restoring a table as of a previous version or timestamp, enabling quick recovery of the correct state without re-running upstream pipelines. OPTIMIZE, VACUUM, or repartitioning change storage layout or clean up files but do not revert data to an earlier logical state.

Databricks secret management is the appropriate place to store sensitive credentials. The notebook can reference the secret at runtime without embedding the username and password directly in code.

Running OPTIMIZE compacts many small files into fewer larger ones, reducing file overhead, and Z-ORDER by `customer_id` clusters data to improve predicate filtering for common queries. This directly addresses the small-file problem and data layout, providing sustained performance gains beyond what more compute or caching alone can offer.

Unity Catalog is the recommended centralized governance layer for Databricks. It provides a three-level namespace (catalog.schema.table), fine-grained permissions, and cross-workspace governance, directly matching the team’s requirements. The legacy Hive metastore, cluster ACLs, and DBFS permissions cannot provide this centralized, object-level control across workspaces.

Delta Lake tables add a transaction log on top of files in cloud object storage, providing ACID transactions, schema enforcement and evolution, and time travel. This allows reliable updates and the ability to query or restore previous table versions for auditing. Plain Parquet, CSV, or JSON files alone do not provide these transactional and versioning capabilities.

Delta Live Tables provides declarative pipeline definitions, built-in expectations for data quality with automatic handling of bad records, automatic dependency management between tables, and integrated monitoring. Orchestrating DLT pipelines via Workflows (Jobs) fits the production scheduling requirement. The other options require manual management of dependencies and data quality or move core pipeline logic outside Databricks.

Running OPTIMIZE compacts many small files into fewer larger files, which improves query performance and can reduce overhead. Using ZORDER on frequently filtered columns further improves data skipping. Changing managed vs external status does not address file layout, simply scaling the cluster increases costs without fixing the root cause, and switching to CSV removes Delta Lake benefits without solving the small-file issue.

Job clusters are created for the duration of a job and then terminated, providing strong isolation between workloads and better cost control for scheduled ETL compared to long-running all-purpose clusters. Sharing a single all-purpose cluster or using personal clusters mixes production with ad-hoc workloads, and serverless SQL alone is not appropriate for all notebook-based ETL patterns.

Unity Catalog provides data-centric permissions at the catalog, schema, table, and view levels. Row- and column-level security is typically implemented using views combined with appropriate grants so that different groups see filtered or masked data while querying a consistent logical object. Cluster-level, notebook-level, or storage-only controls cannot reliably enforce fine-grained, centrally governed access policies.

Unity Catalog is the correct answer because it provides centralized governance for data assets using a hierarchy that includes catalogs and schemas. The requirement is about centrally managing access to tables, which is a governance function, not a workspace organization or development feature. Workspace folders and Repos help organize code and assets, but they do not provide centralized table-level governance.