Skip to content
DP-700 Fabric Engineer Hub

Cheatsheet: DP-700 Exam Traps & Gotchas

Cheatsheet: DP-700 Exam Traps & Gotchas

Source

  • Content type: Quick-Reference Cheatsheet
  • Target Exam: Exam DP-700
  • Date captured: 2026-08-15
  • Last reviewed: 2026-08-15

⚠️ Top 15 DP-700 Exam Traps & Gotchas

  1. SQL Analytics Endpoint is READ-ONLY:
    • Trap: A scenario asks you to run UPDATE, INSERT, DELETE, or CREATE TABLE directly on a Lakehouse SQL analytics endpoint.
    • Reality: The SQL analytics endpoint is strictly read-only. DDL/DML write operations must execute via Spark, Dataflows Gen2, Data Factory Copy pipelines, or within a Fabric Warehouse.
  2. ReadAll Security Bypass:
    • Trap: An analyst is granted ReadAll on a Lakehouse, but has granular Row-Level Security (RLS) configured in SQL.
    • Reality: ReadAll allows users to read raw parquet files directly in OneLake (or via Spark), completely bypassing SQL RLS/CLS. To enforce SQL security, assign Read only (without ReadAll) and manage permissions via SQL GRANT SELECT.
  3. VACUUM Destroys Delta Time Travel:
    • Trap: Running VACUUM with RETAIN 0 HOURS to aggressively reclaim storage before querying historical snapshots.
    • Reality: VACUUM permanently removes unreferenced parquet files older than the retention threshold, causing subsequent TIMESTAMP AS OF time-travel queries to fail with FileNotFoundException.
  4. Nested ForEach Limitation in Data Factory:
    • Trap: Attempting to place a ForEach activity directly inside another ForEach loop in a pipeline.
    • Reality: Data Factory does not support nested loops. You must use an Execute Pipeline activity inside the outer loop to trigger a child pipeline.
  5. Direct Lake Fallback to DirectQuery:
    • Trap: Expecting Direct Lake mode performance when SQL RLS is configured on the Lakehouse SQL endpoint, or when unsupported complex DAX/calculated columns are used.
    • Reality: The semantic model silently falls back from Direct Lake to slower DirectQuery mode.
  6. Shortcuts vs Mirroring vs Ingestion:
    • Shortcut: Zero-copy pointer to external storage (ADLS Gen2, AWS S3, Google Cloud Storage). No compute cost for replication.
    • Mirroring: Continuous Change Data Capture (CDC) replication into Delta Lake for Azure SQL DB, Cosmos DB, and Snowflake.
    • Data Factory Copy / Dataflow Gen2: Point-in-time ETL batch data transfer.
  7. Capacity Smoothing & Throttling Stages:
    • Interactive operations are smoothed over 5 minutes; background jobs over 24 hours.
    • Interactive operations are throttled first: delayed execution when cumulative debt reaches 10 minutes, full rejection when debt reaches 60 minutes.
  8. KQL Update Policy Target Table Attachment:
    • Trap: Attaching the update policy to the raw source table.
    • Reality: The update policy is attached to the target curated table (.alter table CuratedTable policy update ...).
    • Setting "IsTransactional": true ensures that if the transformation query fails, raw ingestion is rolled back.
  9. KQL Query Pipeline Optimization (Early Filtering):
    • Trap: Placing | where filters after | summarize or | join operations.
    • Reality: Always place time and partition filters at the very beginning of the KQL pipeline (| where Timestamp > ago(1h)) to eliminate partitions before expensive operators execute.
  10. Liquid Clustering vs Z-Order & Partitioning:
    • Liquid Clustering (CLUSTER BY (col1, col2)) replaces static directory partitioning and OPTIMIZE ... ZORDER BY.
    • Clustering columns can be redefined dynamically without rewriting table history or repartitioning data.
  11. Native Airflow & Spark Notebook DAG Operators:
    • The native Fabric Airflow operator is FabricRunItemOperator.
    • Native programmatic notebook multi-run execution is invoked via notebookutils.notebook.runMultiple().
  12. The “Recognition vs. Retrieval” Trap in Practice Tests:
    • Official practice questions often test simple keyword association (“low code” -> Dataflows Gen2).
    • The actual DP-700 exam tests multi-constraint enterprise trade-offs (e.g. low code + sub-minute latency + Delta Lake landing -> Eventstream to Lakehouse, NOT Dataflows Gen2).
  13. Case Study Placement Surprise:
    • Case studies can appear at the very start, at the end, or in multiple sections. Never assume question 42/42 is the end of the exam; manage the 100-minute clock to leave at least 15–20 minutes for dense case study tabs.
  14. In-Exam Microsoft Learn Search Limitation:
    • The built-in Pearson VUE Learn search is literal and slow. Do not perform conceptual searches during the exam.
    • Use it solely as an exact-syntax lookup for known functions (e.g. Dynamic Data Masking function syntax, KQL anomaly functions, PySpark notebookutils).
  15. Cross-Database / Cross-Lakehouse Queries:
    • You can query across Lakehouses and Warehouses within the same workspace using 3-part naming: [DatabaseName].[SchemaName].[TableName].
    • Cross-workspace queries require creating OneLake shortcuts.