Free DP-750 practice test: 30 questions with answers

Auto Loader incrementally discovers new files and can evolve the schema as new fields appear, which suits continuous ingestion. A weekly COPY INTO misses timely loads, manual full reads reprocess everything, and Delta Sharing is for sharing data with other consumers.

A checkpoint records progress so the query resumes from the last committed batch, and Delta transactions make writes idempotent. Without a checkpoint the job reprocesses data, disabling transactions removes atomicity, and widgets do not persist offsets reliably.

Declarative pipelines let engineers define datasets as queries and manage dependencies, retries and expectations automatically. Manual streaming code and cron-driven notebooks require the team to manage that orchestration, and Power BI dataflows are a different product.

OPTIMIZE rewrites small files into larger ones, improving read performance. VACUUM removes files no longer referenced, DESCRIBE HISTORY shows table versions, and SHOW TABLES lists tables without changing file layout.

MERGE INTO updates matched rows and inserts unmatched ones atomically based on the key condition. Overwriting everything each day is costly, CTAS does not upsert, and truncate-then-insert in separate jobs risks partial results.

Delta time travel queries earlier versions of a table by version number or timestamp. Restoring from another region is slow and not needed for this, and reading raw Parquet files does not respect the transaction log.

Broadcasting the small table sends it to every executor and avoids shuffling the large table. A single CSV file and a single worker remove parallelism, and huge shuffle partition counts add overhead.

The header option tells Spark the first row contains column names, and sep sets the delimiter to a semicolon. The other options set the wrong delimiter or treat the header incorrectly.

Delta uses optimistic concurrency with conflict detection on the transaction log, so conflicting commits fail and can be retried. Disabling transactions removes safety, separate copies break the single source of truth, and text-file locks are unreliable.

Processing only new or changed rows using a watermark or incremental ingestion avoids reprocessing the whole dataset. Full overwrites, full history reads and disabled pruning all increase cost and time.

Databricks Jobs runs tasks on a schedule and can send notifications on failure, which automates operations. Calendar reminders, sleeping cells and manual runs rely on people and are not reliable.

Task dependencies in a multi-task job define the order and ensure each task waits for its upstream tasks. Random schedules, merged cells and manual starts do not enforce the required order.

Auto-termination stops idle clusters after a set time, which removes unnecessary compute cost. More workers increase cost, disabling logs saves little and hides problems, and running every minute multiplies cost.

Adaptive query execution can detect and split skewed partitions at runtime, which rebalances the work. A single core removes parallelism, removing partitions does not fix skew, and driver memory does not affect executor-side skew.

Defining jobs as code in a bundle and deploying through CI/CD keeps environments consistent and reviewable. Manual recreation, emailed JSON and direct production edits create drift and untracked change.

Scaling compute and tuning trigger and parallelism lets the stream process data faster than it arrives. Stopping the stream lets lag grow, disabling checkpoints risks data loss or duplicates, and one partition limits parallelism.

Expectations evaluate a rule on each row and can warn or drop violations while recording metrics, which fits this requirement. Manual checks after the fact, table comments and deleting every row with any null value do not meet the requirement.

The Spark UI shows stage-level skew and spills, and query profiles or plans show which operations are expensive. Restarting the workspace, deleting logs and changing fonts do not diagnose performance.

Task retries with a maximum count and interval handle transient failures without endless loops. Continuous loops waste resources, silent failures hide problems, and a zero timeout fails every run.

A file arrival trigger starts the job when new files arrive, which matches the event-driven requirement. Frequent cron runs waste compute, and manual or cluster-start triggers do not respond to the arriving data.

Row filters in Unity Catalog are enforced by the governance layer for each query, so users only see permitted rows. Copies multiply data, comments are not enforcement, and editable views allow users to remove the restriction.

Column masks return masked values to most users while allowing the authorized group to read the raw data. Deleting the column removes data needed by compliance, workspace-wide encryption does not control column visibility, and public links expose the data.

USE SCHEMA with SELECT allows reading the tables and no modification. ALL PRIVILEGES and MANAGE grant administrative rights, and CREATE TABLE allows creating new tables rather than reading existing ones.

Audit logs record access events for Unity Catalog objects with the principal and time, and they can be delivered for retention. Cell output, the Spark UI and table comments do not provide a reliable access record.

A storage credential backed by a managed identity or access connector avoids keys in code and enables governed access through Unity Catalog. Keys in notebooks, shared SAS tokens and public containers all expose access.

Granting access through groups and limiting privileges to what each group needs applies least privilege, and scoping privileges to specific catalogs and schemas keeps access narrow. Broad readability across the tenant, admin tokens and secrets in notebooks all expand exposure.

Unity Catalog captures lineage and manages permissions centrally across workspaces attached to a metastore. Governance covers SQL and Python alike, copies are unnecessary, and audit data is not limited to driver logs.

Granting the usage and create privileges on the schema allows creating tables without broad ownership of others' objects. Metastore-wide privileges, admin groups and catalog ownership all exceed the requirement.

Delta Sharing gives recipients governed read access to shared data without copying it, across accounts and platforms. Exports, email extracts and shared admin passwords create copies or unacceptable access.

Governed tags on columns and tables let teams find sensitive data and apply policies by tag consistently. Name prefixes are fragile, notebook comments are not searchable governance metadata, and cluster names describe compute, not data.