Explanation

Correct Answer: B

By default, when data is written into a Delta table, Delta Lake collects file-level statistics—such as minimum, maximum, null counts, and row counts—for the first 32 columns defined in the table schema. During query execution, the Delta engine leverages these statistics to perform data skipping, bypassing Parquet files whose values fall entirely outside the query's filter predicates. This significantly reduces I/O and accelerates query performance.

Why the other options are incorrect:

• A: While Delta Lake supports the declaration of primary and foreign key constraints, it does not currently enforce them. They are primarily used for metadata and documentation purposes; they do not prevent duplicate entries.
• C: Z-ordering is not restricted to numeric columns. It can be applied to any column for which statistics are collected and that can be compared, including strings, dates, and timestamps.
• D: In the Databricks Lakehouse, standard views are logical definitions rather than physical caches. They do not store data themselves. Only Materialized Views or explicitly cached DataFrames hold a snapshot of the data in memory or storage.