Split an Oracle workload into Spark SQL for relational objects and Python/PySpark for procedural code in Fabric Lakehouse.
The job to be done
A migration becomes expensive when every object arrives as a separate mystery. Start with the workload as a whole, then sort the result into three useful categories: what can move directly, what needs adaptation, and what deserves a deliberate engineering decision.
For this path, that means understanding Oracle schemas, built-ins, data types, dependencies, packages, procedures, functions, and triggers. and producing Spark SQL for relational objects and Python or PySpark for procedural logic.. The differences that could affect production behavior are surfaced while the team can still act on them, not after they have become deployment surprises.
Why this path is difficult
Oracle tables and views can become Spark SQL, but packages, procedures, functions, and triggers are executable behavior, not lakehouse tables. Treating every object as SQL either loses the workflow or produces code that does not fit the Spark runtime.
How the work gets done
SQL Tran translates the relational plane into Spark SQL and the procedural plane into PySpark. Oracle expressions, joins, built-ins, and dependencies stay in context, so the split is deliberate rather than a second manual rewrite.
Where it helps
Use this path for Oracle estates moving into Fabric Lakehouse when the target needs both queryable Delta data and PySpark jobs that preserve the source processing logic.
The payoff
The translation runs up front, then becomes an object inventory, coverage view, complexity signal, and findings list. You see the work before you plan it.
Oracle tables and views are emitted as Spark SQL for the Fabric Lakehouse relational plane. Procedures, functions, triggers, packages, and types are emitted as Python/PySpark code for the execution plane.
Generated tests compare source and target behavior for tables, views, functions, and procedures. Performance is recorded alongside correctness, so a passing translation is not mistaken for a fast one.
Where the work happens
These are the source patterns that usually create rework. Each card explains the target-aware treatment, so your team can see what is being adapted and where a decision is still required instead of discovering a mismatch during deployment.
Emit Spark SQL for the relational plane and keep table, view, and query dependencies visible as Oracle objects move toward Fabric Lakehouse.
Translate packages, procedures, functions, and triggers into PySpark workflows instead of forcing executable Oracle behavior into SQL that cannot run in the lakehouse.
Adapt Oracle built-ins, date logic, null behavior, joins, and conversions to Spark functions, with findings for behavior that needs a deliberate implementation.
Keep relational and procedural objects connected through their dependencies, so engineers can see which PySpark jobs read or change which Spark SQL objects.
Generate source-versus-target tests for query results and procedural paths, giving the team evidence that the split preserved the behavior that matters.
The path to confidence
Assessment, translation, and testing stay connected, so your team can see what the software handled automatically, what it changed to fit the target, and which decisions still need a human owner.
Connect or upload Oracle code. The parser reads every object, builds dependencies, scores complexity, and shows where Fabric Lakehouse needs a different construct.
Relational and procedural objects are translated together, with source semantics kept in context. Where Fabric Lakehouse has no direct equivalent, the engine uses a supported rewrite or names the gap clearly.
Deploy the generated target objects to staging, inspect the differences, run generated tests, and use the results to focus engineering time on the exceptions that matter.
The source and target
A script that runs is not proof that a migration worked. The real question is what the target needed in order to preserve the workload’s intent, and whether those changes have been tested. SQL Tran keeps that answer visible at the object level.
Oracle schemas, built-ins, data types, dependencies, packages, procedures, functions, and triggers.
Spark SQL for relational objects and Python or PySpark for procedural logic.
Evidence you can use
This is not a black box. Findings stay attached to the translated workload, so engineers can inspect the change while decision-makers can see what remains before they approve the move.
Your review list
Every migration has a boundary. Surface it early, and reviewers can spend their time on the few decisions that shape the result instead of rereading thousands of generated lines.
Your working set
Take the next step
Run the assessment against your objects, dependencies, and target version. We do that by running full translation and exposing the real result. That tells you exactly what translates cleanly, what changes, and what needs your attention.
Start the assessment