Most Recent Microsoft DP-750 Exam Dumps

CORRECT ANSWE R: C - A Lakeflow Spark Declarative Pipelines (SDP) pipeline.

According to Microsoft Learn, Lakeflow Spark Declarative Pipelines (formerly Delta Live Tables) is purpose-built for exactly this scenario: ingesting files from cloud storage, applying cleansing and enrichment transformations, and writing Delta tables for analytics. SDP provides built-in monitoring (pipeline UI with expectation metrics and event logs), automatic retries on failure, data quality enforcement via expectations, and schema evolution --- all with minimal development effort using declarative Python or SQL. Option A (Spark Structured Streaming) requires manually implementing monitoring, retry logic, and alerting. Option B (scheduled notebook job) lacks built-in monitoring and retry capabilities. Option D (Azure Data Factory with data flows) requires additional Azure service setup, licensing, and does not integrate natively with Unity Catalog.

CORRECT ANSWE R: D - Create separate tasks for ingestion, cleansing, and curation.

According to Microsoft Learn, Lakeflow Jobs (formerly Databricks Workflows) supports multi-task pipelines where each task encapsulates a single, well-defined step. The official documentation states that best practice is to decompose complex pipelines into discrete tasks --- ingestion, cleansing, and curation --- so that each task contains only the logic appropriate for that stage. This approach aligns with the Contoso planned change to 'implement scalable data pipeline orchestration' and the requirement to 'define a clear execution order and dependencies.' Option A is incorrect because combining all logic in one notebook violates the single-responsibility principle and makes retry/recovery difficult. Option B is incorrect because duplicating identical logic across tasks wastes resources and defeats the purpose of a modular pipeline. Option C is incorrect because a single SQL MERGE task cannot cleanly separate the ingestion, cleansing, and curation concerns required.

CORRECT ANSWE R: D - Add a NOT NULL constraint to transaction_id and a CHECK constraint to amount.

According to Microsoft Learn on Delta Lake table constraints, Delta tables support two types of data quality constraints: NOT NULL constraints and CHECK constraints. A NOT NULL constraint ensures that the specified column cannot contain null values --- applying this to transaction_id satisfies 'transaction_id must never be null.' A CHECK constraint enforces a boolean expression on a column --- applying CHECK (amount > 0) ensures that only positive amounts are written. When a row violates either constraint, Delta Lake rejects the write operation, satisfying 'invalid records must be rejected when data is written to the Sales table.' These are table-level constraints enforced by the Delta engine on every write. Option A (SELECT with WHERE) is a read-time filter, not a write-time constraint. Options B and C (RLS and views) are read-time access controls, not write constraints.

CORRECT ANSWE R: C - The Databricks CLI.

According to Microsoft Learn on Databricks Asset Bundles deployment, the Databricks CLI (version 0.205+) is the official tool for deploying DABs to any environment. The deployment commands 'databricks bundle deploy' and 'databricks bundle run' are part of the CLI and support automated, repeatable, and environment-aware deployments. The CLI reads the databricks.yml configuration and deploys the bundle resources to the specified target environment. Option A (Azure Developer CLI / azd) is for deploying Azure infrastructure and does not natively support Databricks Asset Bundles. Option B (Git folders) is a workspace feature for syncing notebook code from Git but does not handle full DAB deployment. Option D (Azure CLI) manages Azure infrastructure and resources but does not have native support for deploying Databricks Asset Bundles.

CORRECT ANSWE R: B - Implement checkpointing.

According to Microsoft Learn on Apache Spark Structured Streaming fault tolerance, checkpointing stores the streaming query's progress in durable storage so that when a failure occurs and the job restarts, it resumes processing from exactly the point of failure without reprocessing previous data. This satisfies all three requirements: 'Resumes processing from the point that Job1 failed' (checkpoint stores last committed offset), 'Minimizes how long it takes to restart' (no data replay needed), and 'Minimizes costs' (no reprocessing of already-processed data). Option A (decrease retry interval) reduces the wait before retry but does not ensure resuming from the failure point. Option C (alert and manual restart) adds human latency. Option D (increase minimum nodes) increases cluster cost without addressing the root cause of transient errors.