Check Your Knowledge

Test what you’ve learned from this section before moving on to physical design for Delta.

1. Why is Oracle’s DATE type a common source of silent migration bugs when mapped to Delta? A. Oracle DATE cannot store negative years B. Oracle DATE always includes a time component, but Delta DATE does not, so mapping it to DATE silently truncates the time C. Oracle DATE is deprecated and should never be migrated D. Delta has no DATE type at all

2. What is the safer default mapping for an unconstrained Oracle NUMBER column, and why? A. DOUBLE, because it requires no precision or scale B. STRING, to avoid any type coercion C. A fixed-precision DECIMAL, because DOUBLE’s binary floating point representation risks rounding error on aggregates D. BOOLEAN, since most NUMBER columns are flags

3. In the Lakebridge workflow, what does the analyzer produce that the converter does not? A. The final converted Databricks SQL DDL B. An object inventory, complexity/effort scoring, and cross-object dependencies – run before conversion, not as a conversion step C. A live connection to the source database D. A reconciliation report comparing row counts

4. Which three transpiler engines does Lakebridge’s converter select between? A. Spark, Hive, and Presto B. BladeBridge, Morpheus, and Switch C. Sqoop, Flume, and NiFi D. Oracle GoldenGate, Qlik Replicate, and Fivetran

5. Why does an undersized target DECIMAL(p,s) cause a worse problem than an oversized one? A. It causes the load job to run out of memory B. It silently rejects or rounds values that exceed its precision, and a rounded financial value doesn’t raise an alarm the way a rejected row does C. It prevents the table from being created at all D. It has no effect on numeric accuracy

6. According to the DDL-auditing lecture, why does Delta Lake not catching dropped foreign keys as an error still matter? A. Foreign keys are automatically reimplemented by the converter B. Delta doesn’t enforce referential integrity at write time, so a dropped FK removes a safety net without any error signaling that loss C. Foreign keys are irrelevant to any analytical workload D. Databricks rejects any schema that lacks foreign keys

7. What commonly happens to source CHECK constraints that reference engine-specific functions during conversion? A. They are automatically rewritten to equivalent Databricks SQL with no review needed B. They are silently dropped or commented out by the converter, and validation logic that lived in the database is now missing unless flagged C. They cause the entire conversion job to fail D. They are converted into Delta NOT NULL constraints

8. For a 200-column legacy table, what does this section recommend over reviewing all 200 columns manually? A. Migrating the table as a single STRING blob column B. Skipping type mapping entirely for wide tables C. Applying the type-mapping matrix programmatically via a metadata join, reserving manual review for ambiguous rows only D. Splitting the table into 200 separate single-column tables

9. What is the standard, engine-version-independent pattern for migrating Oracle SDO_GEOMETRY or Teradata ST_GEOMETRY columns? A. Drop the geometry column entirely since Delta cannot represent spatial data B. Store the geometry as WKT or WKB in a STRING/BINARY column and operate on it with Databricks ST_* functions C. Convert geometry to a DOUBLE representing distance from the equator D. Store the geometry as a Python pickle in a BINARY column

10. Why does the section describe Lakebridge’s transpiler as handling the DDL shape but not the spatial query logic for geospatial columns? A. Because spatial queries never need to change during a migration B. Because functions like Oracle’s SDO_RELATE have no automatic transpilation path and must be manually rewritten against ST_* functions C. Because Lakebridge does not support geospatial source columns at all D. Because Databricks does not support any spatial functions

Answer Key

1. B – Oracle DATE stores a time component that Delta DATE cannot hold, so the mapping silently truncates it; mapping to TIMESTAMP avoids the loss.
2. C – A fixed-precision DECIMAL avoids the binary floating-point rounding error that DOUBLE introduces on financial aggregates.
3. B – The analyzer inventories objects and estimates complexity/effort/dependencies as a pre-conversion assessment step, separate from the converter’s DDL translation.
4. B – BladeBridge, Morpheus, and Switch are Lakebridge’s three transpiler engines, selected via the transpiler configuration.
5. B – Undersized DECIMAL precision causes silent rounding or rejection, and a quietly rounded financial figure is more dangerous than a rejected row because nothing flags it.
6. B – Delta’s lack of write-time FK enforcement means a dropped constraint removes protection against orphaned rows without any error surfacing that fact.
7. B – Engine-specific CHECK constraints are typically commented out or dropped, removing validation logic silently unless the audit process catches it.
8. C – Programmatically joining the source metadata catalog against the type-mapping matrix scales to wide tables; manual review is reserved for genuinely ambiguous columns.
9. B – WKT/WKB stored in STRING/BINARY plus Databricks ST_* functions is the portable pattern that works regardless of native geometry type support.
10. B – Spatial operators like SDO_RELATE have no mechanical equivalent; the underlying query logic must be manually rewritten, not just the column’s DDL shape.

---

| ← Previous: Migrating the Hard Ones: 200-Column Teradata Tables and Geospatial | Next: Physical Design for Delta: Liquid Clustering Over Indexes → | |:—|—:|