Check Your Knowledge

Test your understanding of this section’s pattern translations before moving on.

1. Why is a cursor loop translated as for row in df.collect(): ... considered an anti-pattern?
   • A. collect() is deprecated in current Spark versions
   • B. It pulls the entire DataFrame to the driver, discarding distributed execution and risking out-of-memory failures
   • C. Python for-loops cannot iterate over Spark DataFrames at all
   • D. It requires a paid Databricks license feature
2. Which cursor pattern maps directly to a window function like SUM(...) OVER (ORDER BY ...)?
   • A. A per-row external API call
   • B. A running accumulation or running total maintained across rows in order
   • C. A cursor that never reads any rows
   • D. A cursor used only for logging
3. What replaces an Oracle row-level trigger on Databricks, since Delta tables don’t execute code on write?
   • A. A second trigger defined on the same table
   • B. Delta Change Data Feed, consumed explicitly via table_changes() or readChangeFeed
   • C. A stored procedure called manually after every write
   • D. There is no replacement; triggers cannot be migrated
4. What is the main advantage of Lakeflow Jobs over DBMS_SCHEDULER highlighted in this section?
   • A. Lakeflow Jobs run faster in all cases
   • B. Scheduling, retries, and dependencies are declared visibly and every run produces a queryable history
   • C. Lakeflow Jobs don’t require any configuration
   • D. DBMS_SCHEDULER cannot run PL/SQL code
5. When should a session temp table be translated into a CTE rather than a persisted Declarative Pipeline table?
   • A. Always – CTEs should replace every temp table
   • B. When the intermediate result is used once, immediately, and never inspected independently
   • C. When the intermediate result is shared across many downstream steps
   • D. Never – CTEs are not supported on Databricks
6. What does Delta’s MERGE INTO do differently from Oracle’s MERGE when the join condition matches more than one source row to a target row?
   • A. It silently picks the first matching row
   • B. It throws an error rather than applying a nondeterministic update
   • C. It ignores the match entirely
   • D. It automatically deduplicates the source table
7. Which clause does Delta’s MERGE INTO support that Oracle’s MERGE does not have a direct equivalent for?
   • A. WHEN MATCHED
   • B. WHEN NOT MATCHED
   • C. WHEN NOT MATCHED BY SOURCE
   • D. USING
8. What parameter does create_auto_cdc_flow use to correctly handle late-arriving or out-of-order change records in an SCD Type 2 flow?
   • A. keys
   • B. target
   • C. sequence_by
   • D. stored_as_scd_type
9. Why is unconditionally materializing every temp table as its own persisted Declarative Pipeline table listed as an anti-pattern?
   • A. Declarative Pipelines don’t support persisted tables
   • B. It inflates storage and pipeline DAG complexity with nodes that add no value for single-use staging
   • C. It’s illegal under Unity Catalog governance rules
   • D. It causes MERGE statements to fail
10. What do all five anti-patterns in the gallery have in common?
    • A. They all involve syntax errors that fail immediately
    • B. They preserve the legacy code’s literal shape instead of its intent, and fail only at production scale or under real-world conditions
    • C. They only affect tables smaller than one gigabyte
    • D. They are all related to Unity Catalog permissions

Answer Key

1. B – collect() brings all rows to the driver, defeating distributed execution and risking out-of-memory errors at real data volume.
2. B – A running total or accumulation walked in row order is exactly what a window function computes without any loop.
3. B – Delta tables don’t execute code on write, so Change Data Feed makes row-level changes queryable explicitly instead.
4. B – Lakeflow Jobs make scheduling, retries, and dependencies visible in the job definition and produce a queryable run history.
5. B – A single-use, immediately-consumed intermediate result doesn’t need persisted storage; a CTE computes it inline.
6. B – Delta’s MERGE INTO rejects ambiguous matches with an error instead of silently applying a nondeterministic update.
7. C – WHEN NOT MATCHED BY SOURCE handles target rows with no matching source row, a clause Oracle’s MERGE lacks directly.
8. C – sequence_by tells the AUTO CDC flow which column determines correct ordering, handling out-of-order arrivals correctly.
9. B – Materializing single-use staging as a persisted table adds storage and DAG nodes that provide no benefit over a CTE.
10. B – Each anti-pattern compiles and runs but preserves the legacy shape rather than the intent, failing only at production scale or under real arrival conditions.

---

| ← Previous: Anti-Pattern Gallery and Pattern Cheat Sheet | Next: AI-Assisted Migration: Lakebridge plus LLMs With a Human Gate → | |:—|—:|