Data Architecture · Migration & Modernization
Databricks Migration Assessment Checklist
A migration fails in inventory, not in Spark. If you cannot name the workloads, owners, and contracts, you are not ready to convert them.
9 min · Independent technical note
Moving a warehouse, Hadoop estate, or legacy Spark farm onto Databricks is a conversion program. The assessment is the part that prevents you from copying every stored procedure into a more expensive runtime.
1. Inventory what actually runs
- Jobs, stored procedures, ETL packages, notebooks, and SaaS extracts—with last-run evidence, not wiki pages.
- Tables and files with row counts, grain, and a suspected owner.
- Reports and applications that consume them, including the unofficial ones.
2. Classify, do not just list
Every workload should land in one of: convert as-is, redesign, retire, or park. “Park” is for things you cannot yet prove are unused. “Convert as-is” should be rare. Legacy logic often encodes business rules that should be made explicit in silver or gold rather than translated line-for-line.
3. Map dependencies and cutover risk
A table is easy. The overnight process that waits on it, the vendor extract that lands beside it, and the finance file that still joins on an old key are the migration. Draw those edges before you schedule wave one.
4. Define the target contracts
Name the Unity Catalog layout, environment strategy, and the bronze/silver/gold (or equivalent) contracts the converted workloads must obey. If the target is “a workspace with Delta tables,” you have a landing zone, not an architecture.
5. Agree validation before conversion
- Row counts and amounts at agreed grains.
- Late-arriving and restatement behavior.
- NULL, duplicate, and timezone handling.
- Who signs that a domain can cut over—and who can roll back.
6. Plan retirement explicitly
Dual-running forever is how you pay for two platforms. Each wave should include the date the source job is turned off, the owner of that decision, and the evidence required.
If nobody will own a report after cutover, it is a retirement candidate. Migrating orphaned content is how lakehouses inherit the last decade's clutter.
Technical statements in this article follow Databricks public documentation on lakehouse architecture, Unity Catalog, and platform capabilities. Product names belong to Databricks, Inc.
Continue
Databricks Architecture
Databricks vs Traditional Data Warehouse Architecture
The useful comparison is not brand versus brand. It is whether one architecture can support engineering, warehousing, governance, and AI without copying data into a second estate.
9 min
Databricks Architecture
Medallion Architecture: When Bronze, Silver, and Gold Make Sense
Medallion is a quality contract between layers. It is not a mandatory folder taxonomy, and it is not automatically correct because the names are famous.
8 min
Start with the business case
Find the first data or AI opportunity worth proving.
We evaluate the business problem, systems, data, architecture, and economics behind it—then identify the smallest production engagement capable of proving whether the opportunity is real.
Business case first · Architecture-led · Production-focused