For DP-750, know how Delta Lake tables are created and queried, how INSERT, UPDATE, DELETE, and MERGE change rows, and how schema rules, history, OPTIMIZE, and VACUUM affect a table. The practice questions below are original study prompts based on published exam objectives—not actual or protected exam questions.
What table operations should a DP-750 candidate recognize?
Delta Lake supports both ordinary reads and row-level changes. The key distinction is between changing the table’s logical contents, changing its schema, and maintaining the files that store its data. Those operations can have different consequences for later queries and historical snapshots.
As an Amazon Associate I earn from qualifying purchases.
The Microsoft DP-750 study guide topics identified for this article include loading data with merge, insert, and append; schema enforcement and drift; history tables; clustering strategy; and Delta optimization with OPTIMIZE and VACUUM. These are study areas, not a guarantee of how often a topic appears on an exam. Check Microsoft’s live DP-750 study guide for the current skills outline; a listed skills-measured date of October 19, 2026 was still in the future as of October 10, 2026.
How do you create, query, and write a Delta table?
A Delta table can be created through SQL or by writing a DataFrame, then queried through interfaces supported by the engine in use. Azure Databricks documents Delta Lake as the storage layer underlying Databricks tables unless otherwise specified; that platform-specific default should not be generalized to every engine or storage system.
#1 Best Overall
When loading rows, distinguish appending data from applying changes to existing records. Append is suitable when incoming rows should be added as new rows. If records may already exist and need updating, a merge pattern is usually more appropriate. Before choosing, establish whether the incoming data can contain duplicate keys or conflicting records.
How do INSERT, UPDATE, DELETE, and MERGE differ?
| Operation | What it does | Typical decision |
|---|---|---|
| INSERT or append | Adds rows to the table. | Use when the incoming rows are new and should be added, rather than matched against existing records. |
| UPDATE | Changes values in rows that satisfy a match condition. | Use when the target rows already exist and should be modified. |
| DELETE | Removes matching rows from the latest logical table state. | Use when matching rows should no longer appear in the current snapshot. |
| MERGE | Applies actions according to source-to-target matching conditions; it can update matched rows and insert unmatched rows. | Use for upserts or change-data workflows where incoming records may be new or may revise existing records. |
These operations are not interchangeable: append does not, by itself, update an existing record, while MERGE needs a meaningful match condition to determine which source rows correspond to target rows. A merge source should be deduplicated or otherwise constrained so multiple source rows do not attempt conflicting updates to the same target row.
When is MERGE the right choice for an upsert?
Choose MERGE when one batch contains both changes for existing keys and new keys. The match condition defines which records are considered the same; the configured actions then determine what happens to matched and unmatched records. For example, if a customer key matches, the operation can update that target record; if no target key matches, it can insert a new record.
Do not treat the word “upsert” as a substitute for defining key quality and source uniqueness. If the source has multiple competing rows for one target key, the intended update is ambiguous. Resolve that in the source data or with an explicit rule before the merge.
Schema evolution is a separate decision from row matching. A merge’s ability to accommodate new columns depends on the operation, supported syntax, and Delta Lake or runtime version. Use version-appropriate documentation rather than assuming an evolution option works universally.
What do table history and time travel let you do?
A modifying operation creates a new table version. History helps inspect operation metadata; time travel reads an earlier retained snapshot. They answer different questions: history tells you about recorded table operations, while time travel requests the data state associated with a particular earlier version or timestamp.
Historical reads depend on the transaction log and the data files needed for the requested snapshot still being available. Microsoft’s Azure Databricks history documentation says that, in Databricks Runtime 18.0 and later, time-travel requests older than the table’s deletedFileRetentionDuration are blocked. The documented default for that property is seven days. This is a version-scoped Databricks behavior, not a universal Delta Lake retention guarantee.
Free tools Windows power users keep installed
One-click scans. No signup required.
How do schema enforcement and schema evolution work?
Schema enforcement validates incoming writes against the table schema. It helps prevent incompatible data from silently becoming part of the table. When incoming data has a different shape, decide whether the table schema should change or the input should be transformed to fit the existing schema.
- Explicit schema change: use a deliberate schema operation, such as ALTER TABLE, when the intended change should be visible and controlled.
- Automatic schema evolution: use only when the specific write or merge operation supports it and its version-specific behavior matches the intended change.
Do not infer that every missing or additional column is handled identically by every merge. The treatment of columns present on only one side, as well as the syntax for evolution, varies by operation and software version. Confirm the applicable behavior before relying on it.
When should you use OPTIMIZE or clustering?
Many small files can make reads less efficient. In Databricks, OPTIMIZE compacts files to improve read performance. Liquid clustering is another Databricks table-layout option; the choice depends on the workload, table characteristics, and platform support. Databricks also documents OPTIMIZE FULL for applying clustering to existing data.
For eligible Unity Catalog managed tables, Databricks predictive optimization may manage maintenance automatically when enabled. In that case, manual optimization may not be necessary. Treat OPTIMIZE and clustering as physical-layout choices: they are distinct from changing row values or evolving the table schema.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →What does VACUUM remove, and what is the historical trade-off?
DELETE changes which rows belong to the latest logical snapshot; it does not necessarily erase the old data files immediately. VACUUM physically removes eligible obsolete data files according to retention rules. Keeping files available can support historical reads, while cleaning eligible files can reduce stored obsolete data.
Best Value
Before scheduling VACUUM, establish how much historical access the workload requires and confirm the applicable retention settings. Once files needed for an older snapshot have been removed, table history alone cannot restore those file contents. Do not bypass safeguards or change retention casually.
Original DP-750-style practice questions
1. A batch includes updated customer records and customer IDs not yet in the target table. Which operation fits?
Answer: MERGE, with a suitable customer-key match condition. Ensure the source does not contain conflicting duplicate keys.
2. A team needs to inspect what operation ran, then query the table as it existed before that operation. Which capabilities address those tasks?
Answer: Use history to inspect operation metadata and time travel to read an earlier retained snapshot.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →3. A write introduces a column absent from the existing table schema. What should the engineer check before allowing the write?
Answer: Whether an explicit schema change or supported automatic evolution is intended, and whether the syntax and behavior apply to the operation and runtime version in use.
4. Rows no longer appear after DELETE, but storage still contains old files. Which operation is responsible for eligible physical cleanup?
Answer: VACUUM, subject to retention rules and the table’s historical-access requirements.
5. A Databricks table has many small files and slow reads. What maintenance approach might help, and what alternative should be considered?
Answer: OPTIMIZE can compact files; consider whether liquid clustering better fits the layout need, and whether predictive optimization already manages maintenance for an eligible Unity Catalog managed table.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
