Databricks Certified Data Engineer Professional Practice Exam

A nightly batch job merges a new batch of records into an events Delta table, where event_id is a unique key. New records that share an event_id with an existing row should update it, and brand-new event_ids should be inserted. Which operation expresses this directly?

1. INSERT INTO events SELECT * FROM new_events
2. MERGE INTO events USING new_events ON events.event_id = new_events.event_id WHEN MATCHED THEN UPDATE … WHEN NOT MATCHED THEN INSERT …
3. CREATE OR REPLACE TABLE events AS SELECT * FROM new_events
4. DELETE FROM events; INSERT INTO events SELECT * FROM new_events

A long-running nightly batch ETL processes very large JSON volumes into Delta. Performance and reliable completion are the top priorities, with cost a secondary concern. Which compute choice fits best?

1. A job cluster that autoscales across multiple workers for the duration of the run
2. An all-purpose cluster kept always-on for low-latency startup
3. A single-node cluster with minimal workers to cut cost
4. A high-concurrency cluster designed for interactive SQL

Predictive Optimization is enabled by default for Unity Catalog managed tables to keep Delta tables performant. Which two maintenance operations does it run on your behalf? (Choose two.)

1. OPTIMIZE (compaction)
2. PARTITION BY
3. BUCKETING
4. VACUUM

A streaming pipeline reads a source that occasionally redelivers the same record. The engineer must drop duplicates within the stream based on a unique key while keeping state bounded over time. Which approach fits best?

1. Collect the full stream to the driver and call distinct() in memory
2. dropDuplicates on the key combined with a watermark to bound state
3. Disable checkpointing so old keys are forgotten automatically
4. Write all records and deduplicate manually once a year

PII (email, phone, IP) arrives in both daily batch files and a real-time stream. It must be masked before storage, handled consistently across both pipelines, and remain auditable and reproducible. Which design meets all requirements?

1. Store PII unmasked in Bronze for lineage, then mask only in Gold reporting tables
2. Apply a shared, version-controlled masking/anonymization transformation in both batch and streaming paths before writing, governed and audited via Unity Catalog
3. Mask only the batch path, since streaming data is transient
4. Email a masked extract to analysts and keep raw PII everywhere else

A team needs a modular, version-controlled way to package and promote Lakeflow Jobs and pipelines across dev, test, and prod through automated CI/CD. Which Databricks tooling supports this directly?

1. Manually exporting and importing notebooks per environment
2. Databricks Asset Bundles defined in code, versioned in Git, validated and deployed via the Databricks CLI in CI/CD
3. Relying on notebook revision history as version control
4. Copying jobs through the workspace UI by hand

An engineer must track historical query and job behavior across the workspace — who ran what, how long it took, and how costs trend — to build alerting on regressions. Which Databricks feature is the governed, queryable source for this?

1. Screenshots of the Spark UI saved manually
2. System tables (the system catalog) in Unity Catalog
3. Printing logs to notebook output
4. Asking each engineer to self-report

For auditing, a team must retain the full history of how each customer’s address changed over time. An engineer argues a Type 1 dimension plus Delta time travel is enough; another prefers Type 2. Which fact is most critical to the decision?

1. Type 1 with time travel is the simplest and always sufficient for long-term history
2. Delta time travel is not designed as a long-term versioning/audit solution — retention and cost/latency make it unsuitable, so a Type 2 dimension is the durable way to keep full history
3. Type 2 dimensions cannot represent historical changes
4. Time travel retains all versions forever at no cost

An organization must share live Delta data with an external partner who uses a different platform, without copying the data or building a custom API. Which Databricks capability is designed for this?

1. Emailing nightly CSV exports
2. Delta Sharing
3. Granting the partner full workspace admin access
4. Mounting the partner’s laptop to the cluster

An engineer must restrict an analytics group so it can read a customers table but cannot see the rows for EU customers, enforced centrally and reusably across multiple tables. Which Unity Catalog capability fits best?

1. Make a manual filtered copy of the table for the group
2. A row filter / ABAC policy in Unity Catalog applied to the relevant tables
3. Tell the group not to query EU rows
4. Revoke all access and email results on request