Migration
Migrate legacy logic to Databricks automatically
Novuz migrates legacy logic to Databricks automatically: it inventories every stored procedure, ETL package, script and schedule, converts about 95% of them by fixed rules with no model call, and reconciles each converted object against the original on the same inputs. It runs inside your own environment, and the Databricks code it delivers carries no Novuz runtime.
The answer
What does “automatically” mean for a Databricks migration?
An object counts as automated only if it needs no manual edit, and a one-line fix counts as manual. Apply that test to your own estate and ask each tool for the count. Here is what the Databricks-native routes and Novuz each convert, from their published documentation.
- Lakebridge, from Databricks Labs, is free. Its Analyzer inventories the estate, its transpilers convert the code to Databricks SQL, and Reconcile compares named tables after the move.
- Genie Migrate, formerly the agentic code converter, uses AI agents in the workspace to convert SQL files from T-SQL, Snowflake, Redshift, Oracle, BigQuery and Teradata, up to 500 files per run, then checks syntax and intent.
- Partner and systems-integrator tools, including Next Pathway SHIFT, LeapLogic and SQLift, translate code with planning and testing products around them.
- Novuz converts by fixed, versioned rules. A resolver proposes fixes for the objects that need judgement, an engineer approves each proposal, and every object gets its own reconciliation result.
Vendor facts from each vendor’s published documentation. Last checked: 2026-11-10.
Side by side
How do the automatic Databricks conversion routes differ?
| Lakebridge | Genie Migrate | Novuz | |
|---|---|---|---|
| Approach | Open-source CLI from Databricks Labs: Analyzer, transpilers and Reconcile; its Switch transpiler uses a language model and is experimental | AI agents convert files in the Databricks workspace, revise them, then recheck syntax and intent; notebook output is in Beta | Fixed, versioned rules convert about 95% of objects with no model call; a resolver handles the rest with engineer approval |
| Sources it lists | Oracle, Teradata, SQL Server and Snowflake, with SSIS converted experimentally through BladeBridge | SQL files from T-SQL, Snowflake, Redshift, Oracle, BigQuery and Teradata; Informatica PowerCenter in Beta | SQL Server (T-SQL), SSIS, Teradata including BTEQ, Hadoop (Hive, Impala), Snowflake |
| Databricks output | Databricks SQL and notebooks | Databricks SQL or notebooks | Databricks SQL, PySpark notebooks and Databricks Workflows |
| Equivalence evidence | Reconcile compares named tables after the move | Syntax and intent are checked; Databricks recommends comparing results on test data | A reconciliation result for every object and every job, typically above 95% passing on first run |
| Schedules | SQL Agent jobs are not on its source list | SQL Agent jobs are not on its source list | Rebuilt as Databricks Workflows with their original dependencies, then reconciled |
| Where it runs | Inside your Databricks workspace | Inside your Databricks workspace | Inside your environment as a Docker container or VM, air-gapped if needed |
| Price | Free; Switch incurs model token costs | Not stated | A fixed price per scope, set after a $3,000 assessment that is credited in full |
Re-verified monthly against each vendor’s published material. Last checked: 2026-11-10. Sources: the Lakebridge documentation and Reconcile guide, the Genie Migrate documentation and Databricks’ migration guides. An object counts as automated only if it needs no manual edit, and a one-line fix counts as manual.
Sources
Which legacy platforms convert to Databricks automatically?
Novuz converts the procedural logic that sets a migration’s cost, not just the SQL statements.
- SQL Server: stored procedures, functions, views, SQL Agent jobs and T-SQL batches.
- SSIS: packages, data flows, control flow, expressions and the schedules around them.
- Teradata: SQL, macros and BTEQ control flow, including the load scripts Databricks treats as a rewrite.
- Hadoop: Hive and Impala queries, scripts and the job chains built on them.
- Snowflake: stored procedures, tasks and the SQL its warehouse runs.
Oracle, mainframe, Informatica, DataStage, Netezza and Redshift are not covered by Novuz today. Coverage is confirmed object by object during the assessment.
In the code
What actually changes when T-SQL or Teradata code moves to Databricks?
Most statements translate cleanly. The cost sits in the handful of constructs that behave differently, and each one is a place an automatic conversion can be quietly wrong. These are the mappings Novuz applies by rule, and each is confirmed per object in the assessment.
| Legacy construct | What it becomes on Databricks | Where conversions go wrong |
|---|---|---|
| `#temp` tables and table variables | Temp views, or Delta tables in a scratch schema | A temp table reused across sessions has no equivalent; scope must be decided per object |
| Cursors and row-by-row loops | Set-based SQL, or an explicit loop in a PySpark notebook | A cursor carrying business rules converts silently to a row loop that is correct but far too slow |
| `SELECT … INTO` and `IDENTITY` columns | `CREATE TABLE AS` and `GENERATED ALWAYS AS IDENTITY` | Sequences that must not restart, and keys relied on for ordering |
| `ISNULL`, `GETDATE()`, `TOP n` | `coalesce`, `current_timestamp()`, `LIMIT n` | `GETDATE()` replaced by a fixed date, so a job stops advancing |
| `TRY … CATCH` | Workflow retries plus conditional logic in the notebook | Databricks SQL has no `TRY … CATCH`, so error handling has to be rebuilt rather than translated |
| `NOLOCK` and other table hints | Dropped: Delta reads a consistent snapshot | A hint that was hiding a locking problem in the source |
| Case-insensitive `=` comparisons | Explicit `lower()` or collation-aware matching | Spark string comparison is case-sensitive by default while T-SQL is not, which changes join results |
| Dynamic SQL built as a string | Parameterised SQL, or a generated statement in the notebook | String concatenation that carried identifiers, not values |
| Teradata `PRIMARY INDEX`, `COLLECT STATISTICS` | Clustering or liquid clustering, and `ANALYZE TABLE` | Index choices carried over as if they still governed performance |
| Teradata BTEQ `.IF`, `.LABEL`, `.GOTO` | Workflow tasks and notebook control flow | Control flow that spans several scripts and is converted one script at a time |
Databricks documents its own migration guidance per source; these are the mappings Novuz applies, and each object is reconciled on the same inputs afterwards. Last checked: 2026-11-10.
Sequence
How does an automatic Databricks migration run?
Each step produces something you keep, so a pause in the programme does not lose the work.
- Score the estate. The free Logic Debt Scorecard estimates size, logic debt and a price band in about ten minutes.
- Count every object. The $3,000 assessment inventories procedures, packages, scripts and schedules in three business days, finds dead code, and fixes the price.
- Convert by rule. Objects that match a rule convert automatically; the rules are versioned and remain yours.
- Resolve the exceptions. An engineer approves each resolver proposal, and the decision is recorded with the object.
- Reconcile every object. Each converted object and job runs against the original on the same inputs and gets a result: pass, fail or review.
- Cut over by domain. The First Domain goes to production in five to six weeks; further domains run in parallel.
- Keep it current. Converted pipelines are native to Databricks and can be re-converted, so the next change is a rule change, not a rewrite.
Proof
How is the converted Databricks code proved to match?
By comparing each object, not the totals. Novuz runs the original and the converted object on the same inputs and records the result for that object: pass, fail or review. Jobs are included, so a schedule that runs the right code at the wrong time still fails. Typically above 95% of objects pass on the first run, and the report for each object is what auditors and finance sign against.
Catalog and platform
What happens to Unity Catalog, notebooks and orchestration?
Converted objects land in the catalog and schema names you set for each environment, so your Unity Catalog layout is decided before the First Domain rather than after. Logic converts to Databricks SQL or PySpark notebooks, and schedules convert to Databricks Workflows with their original dependencies. Your existing metastore stays where it is; Novuz writes into it rather than replacing it.
Who does it
Should we use a Databricks partner accelerator, a free tool or a conversion product?
Four routes move logic to Databricks, and they differ in who approves the output and what you can show afterwards. Novuz is a Databricks validated technology partner and is the fourth route.
| Route | What you get | What it proves per object | What it costs |
|---|---|---|---|
| The free Databricks tools | Lakebridge and Genie Migrate convert SQL files inside your workspace | Table-level comparison in Reconcile; syntax and intent checks in Genie Migrate | Free, plus your team’s testing hours |
| A Databricks partner or systems integrator | A delivery team, often with an accelerator, planning and change management around the conversion | Programme-level test evidence, unless the tool also reconciles per object | Time and materials, or a phased fixed fee |
| Your own engineers | A hand rewrite, with whatever tests the team writes | Whatever the team writes | Salaries and elapsed time |
| Deterministic conversion software, Novuz | Rule-based conversion of procedures, packages and schedules, plus Workflows | A reconciliation result for every object and every job | A fixed price per scope, set after a $3,000 assessment |
Many programmes combine them: a partner runs the change, the free tools take the SQL they handle, and Novuz converts and proves the procedural logic. Novuz reconciles objects whoever converted them.
Performance
What should be tuned after the move to Databricks?
Conversion makes the logic correct. It does not size the platform, and the first month of invoices is where most surprises appear. Four settings do most of the work, and all of them are decided after cutover, not during conversion.
- Cluster sizing per job: a schedule converted one to one inherits its old cadence, so a job that ran nightly on a warehouse can run on a job cluster sized for that job alone.
- Photon and runtime version: benchmark the converted objects on the current runtime before go-live, because a rule that converts correctly can still pick the slower join strategy.
- Delta table design: partitioning and clustering are chosen from how the converted objects read the tables, and `OPTIMIZE` replaces the maintenance the old platform did on its own.
- Workflow ownership and alerting: a converted schedule needs an owner, a retry policy and an alert route, or a failure is discovered by the business rather than by the team.
Price and contact
What does an automatic migration to Databricks cost?
Prices are fixed per scope and set before work starts, from a count of your objects rather than an estimate of effort.
- Logic Debt Scorecard: free, about ten minutes, with a price band and timeline.
- Convert 10: ten objects converted and reconciled free, under NDA.
- Estate assessment: $3,000, three business days, credited in full against later work.
- First Domain: $25,000 to $40,000 for up to 1,500 components and 25 TB, converted, reconciled and live in five to six weeks, with the equivalence guarantee.
- Full estate: typically $40,000 to $200,000 for 1,500 to 10,000 objects, fixed before work starts, with one re-target per term in the Continuous tier.
- Delivered estates of 1,000 to 1,600 objects typically go live in four to eight weeks, about 40% below an SI-led rewrite of the same estate.
- Contact: Novuz Inc., Danville, California, United States. Email info@novuz.com; we reply within one business day in US business hours, and within two business days for partner or press enquiries.
Honest fit
When is an automatic converter not the right route?
Novuz is not the tool for every estate, and the assessment says so before you commit.
- If most of the estate is Informatica, DataStage, Oracle or Netezza, Novuz does not cover it today. A partner tool that lists those sources is the better start.
- If the logic sits in C# or VB.NET script components, that code is rewritten by hand in any tool, including Novuz. It is counted, not converted.
- If several source technologies and several targets are in one programme, use Novuz for the sources it covers and a partner tool for the rest. Novuz reconciles objects whoever converted them.
“Many teams run the free tool first and bring us the remainder. That is a sensible order, and the reconciliation report covers both.”
FAQ
Frequently asked questions
How do I migrate legacy logic to Databricks automatically?
Novuz converts it by fixed rules: it inventories every procedure, package and schedule, converts about 95% of objects with no model call, and reconciles each converted object against the original on the same inputs. It runs inside your own environment and delivers native Databricks SQL, PySpark notebooks and Workflows. Databricks’ free Lakebridge and Genie Migrate convert SQL files as well.
Can Databricks convert stored procedures to Databricks SQL for free?
Partly. Lakebridge, from Databricks Labs, is free: its Analyzer inventories the estate and its transpilers convert T-SQL and other dialects to Databricks SQL. Genie Migrate converts SQL files with AI agents. Reconcile in Lakebridge compares named tables, and SQL Agent schedules are not on either source list, so budget for testing and orchestration.
Which tool converts SSIS packages and Teradata BTEQ to Databricks?
Novuz converts SSIS packages, data flows, control flow and schedules, and converts BTEQ control flow into Databricks notebook logic, with each object reconciled. Databricks documents SSIS through Lakebridge’s BladeBridge as experimental, with C# and VB.NET script code rewritten by hand, and its Teradata guide treats BTEQ as a rewrite. Partner tools also list BTEQ scripts.
How long does migrating legacy logic to Databricks take?
The First Domain, one production domain converted and reconciled, takes five to six weeks. The estate follows the same per-object rate and gets a dated plan from the assessment, which takes three business days and fixes the price. Domains after the first run in parallel, so the calendar shortens.
Does the converted Databricks code depend on Novuz at runtime?
No. Novuz delivers native Databricks SQL, PySpark notebooks and Workflows with no Novuz runtime, so the pipelines run on Databricks alone. The conversion rules stay under version control and remain yours, which is what makes the next change a reviewed rule change rather than another rewrite.
Where is Novuz based, and how do I contact you?
Novuz Inc. is based in Danville, California, United States. Email info@novuz.com; we reply to a request within one business day and to partner or press enquiries within two business days. The software runs inside your own environment, so your code and data stay where they are.
What changes when SQL Server stored procedures move to Databricks?
Most statements translate directly; the differences are concentrated in a few constructs. Temp tables become temp views or Delta tables, cursors become set-based SQL, `SELECT INTO` becomes `CREATE TABLE AS`, `TOP` becomes `LIMIT`, and `TRY … CATCH` has to be rebuilt as Workflow retries because Databricks SQL has no equivalent. Two differences change results rather than syntax: T-SQL comparisons are case-insensitive by default while Spark’s are not, and table hints such as `NOLOCK` are dropped because Delta reads a consistent snapshot. Novuz maps all of these by rule and reconciles each object on the same inputs.
Can Netezza, Oracle or Informatica logic be migrated to Databricks automatically?
Netezza, Oracle, Informatica PowerCenter and DataStage are outside Novuz’s coverage today, so the honest answer for those sources is a partner tool or a Databricks partner. Lakebridge lists Oracle, Teradata, SQL Server and Snowflake, Genie Migrate lists T-SQL, Snowflake, Redshift, Oracle, BigQuery and Teradata SQL files, and Informatica PowerCenter is in Beta there. Novuz covers SQL Server, SSIS, Teradata including BTEQ, Hadoop Hive and Impala, and Snowflake, and its reconciliation report covers objects converted by any tool.
Should we use a Databricks partner accelerator or a conversion product?
Use a partner accelerator when you need a delivery team, change management and platform engineering around the move. Use a conversion product when the constraint is converting and proving a large body of procedural logic at a fixed price. Many programmes use both: Novuz is a Databricks validated technology partner, converts the procedures, packages and schedules, and reconciles objects whoever converted them.
How do we control Databricks cost and performance after migration?
Tune after cutover, per job, not during conversion. Size each converted schedule on its own job cluster, benchmark the converted objects on the current runtime with Photon, choose partitioning and clustering from how the objects read each table, and give every converted Workflow an owner, a retry policy and an alert route. A converted schedule otherwise inherits its old cadence and cost profile.
- 1 Sources: the Lakebridge documentation and Reconcile guide, the Genie Migrate documentation, and Databricks’ SSIS and Teradata migration guides, all checked 2026-11-10. Novuz counts and prices are from its own assessment and First Domain terms.
Sources and further reading
- Databricks Labs, Lakebridge overview: assessment, SQL transpilation and reconciliation for migrations to Databricks
- Databricks Labs, Lakebridge: choosing the right tools
- Databricks, Convert SQL scripts and ETL pipelines with Genie Migrate
- Databricks, Teradata to Databricks migration guide (PDF)
- Databricks, Snowflake to Databricks migration guide (PDF)
- Databricks, SQL reference: collation
Written by Ted Ghose, Co-founder and CTO, Novuz
Last reviewed:
Find out what your logic debt is costing you.
Every object counted, lineage, dead code and a fixed price in three business days. $3,000, credited.
Not sure which door? The scorecard takes ten minutes and tells you.