Databricks Certified Data Engineer Associate Practice Test

Databricks Data Engineer Associate practice test: what's covered

• Data Engineer Associate

Free Databricks Data Engineer Associate sample questions

1. A team currently keeps raw files in a data lake for cheap storage but loads a separate warehouse for ACID transactions and BI. They want to eliminate this duplication. How does the Databricks Lakehouse architecture solve this?

   • A. It adds Delta Lake on top of open-format files in object storage, giving ACID transactions, governance, and BI performance directly on the lake without a separate warehouse copy
   • B. It exports the data lake nightly into a proprietary columnar warehouse format that only Databricks SQL can read
   • C. It mounts the warehouse tables into the lake as read-only views so the warehouse remains the system of record
   • D. It replaces object storage entirely with a managed relational database engine running on the cluster
   Show answer

   Correct: A. The Lakehouse uses Delta Lake (open Parquet plus a transaction log) on cheap object storage to provide ACID, time travel, and warehouse-grade performance, so no separate warehouse copy is needed. The nightly-export option is wrong because it reintroduces duplication and a proprietary format. Mounting the warehouse as the system of record is incorrect because the lake itself becomes authoritative. Replacing storage with a relational engine is not how the Lakehouse works, unlike the open-storage design it actually uses.

2. An organization with several Databricks workspaces needs one governed namespace for tables, volumes, and ML models that is consistent across all of them. Which Unity Catalog characteristic enables this?

   • A. Unity Catalog is an account-level metastore providing a three-level catalog.schema.table namespace shared across workspaces
   • B. Unity Catalog runs one independent Hive metastore per workspace that must be synchronized manually
   • C. Unity Catalog stores objects only inside each cluster and discards them when the cluster terminates
   • D. Unity Catalog can only govern tables in a single default schema named hive_metastore
   Show answer

   Correct: A. Unity Catalog is attached at the account level and gives a three-level namespace (catalog.schema.table) plus centralized lineage and access control across all assigned workspaces. The per-workspace Hive option is incorrect because that describes the legacy metastore Unity Catalog replaces. Cluster-only storage is wrong because objects persist in the metastore, not on ephemeral clusters. The single hive_metastore schema claim is false, unlike the multi-catalog model Unity Catalog provides.

3. A nightly ETL job runs unattended and you want to minimize cost while isolating it from interactive users. Which Databricks compute choice best fits this scheduled workload?

   • A. A job cluster that is created when the scheduled job starts and terminates automatically when it finishes
   • B. A large all-purpose cluster left running continuously so the job never waits for startup
   • C. A shared all-purpose cluster that interactive analysts also attach notebooks to during the day
   • D. A SQL warehouse configured for BI dashboards that the ETL notebook reuses for Spark jobs
   Show answer

   Correct: A. Job clusters spin up for a single scheduled run and tear down afterward, so you pay only for the run and the workload is isolated. A continuously running all-purpose cluster wastes money because it bills while idle. A shared interactive cluster is incorrect because it mixes ETL with interactive users and is costlier. A SQL warehouse cannot run arbitrary Spark ETL notebooks, unlike a job cluster which is built for that.

4. A data engineering team wants notebooks version-controlled in Git so pull requests and CI/CD govern changes, instead of editing notebooks ad hoc in the workspace. Which approach achieves this?

   • A. Use Databricks Repos to clone the Git repository so notebooks are Git-backed with branches and commit history
   • B. Right-click each notebook in the workspace and export it to DBFS as a backup after every edit
   • C. Store the notebooks in the driver node local disk and copy them to S3 nightly with a cron job
   • D. Enable workspace auto-save, which automatically pushes every keystroke to the remote main branch
   Show answer

   Correct: A. Databricks Repos integrates a Git provider so notebooks live on branches with commits, enabling PR review and CI/CD. Exporting to DBFS is incorrect because it is a manual backup with no real version control. Driver-disk storage is wrong because that disk is ephemeral and not source control. Auto-save pushing every keystroke to main does not exist, unlike Repos which uses explicit commits and branches.

5. New JSON files land continuously in a cloud storage folder, and you must ingest only newly arrived files efficiently without rescanning the whole directory each run. Which approach fits?

   • A. Use Auto Loader via spark.readStream.format('cloudFiles') with cloudFiles.format set to json for incremental file ingestion
   • B. Run spark.read.json on the full folder every hour and overwrite the target table each time
   • C. List the directory with dbutils.fs.ls and manually diff filenames against a text log on every run
   • D. Schedule a VACUUM on the source folder to remove already-processed files before each load
   Show answer

   Correct: A. Auto Loader (cloudFiles) incrementally tracks which files have been seen using checkpointed state and scalable file notification, ingesting only new files. Re-reading and overwriting the whole folder is incorrect because it reprocesses everything and is expensive. Manual filename diffing is fragile and not scalable, unlike Auto Loader's managed tracking. VACUUM operates on Delta tables to remove old data files, not to delete source files, so that option is wrong.

Last updated: 2026-10-19

Databricks Data Engineer Associate practice test FAQ