How Databricks Lakehouse Replay reruns your read-only serverless queries on unreleased runtimes to catch regressions before they ship.
Lakehouse Replay, now in Public Preview, moves part of that burden to Databricks. Instead of you testing the new runtime, Databricks tests it against your queries, in your workspace, before the version is released to anyone.
Databricks pipeline unit testing hits Beta: pytest-style tests for Lakeflow Declarative Pipelines, with mock data and Auto CDC coverage.
Unit testing a plain Python function has never been the hard part: write the function, feed it inputs, assert on the output, done. Testing a pipeline transformation has always been a different story, and that gap is exactly what Databricks pipeline unit testing was built to close. Read any long-running Databricks community thread on testing Lakeflow Declarative Pipelines (or Delta Live Tables, as it was previously known), and the advice hasn’t changed in years: extract your transformation logic into a plain function that takes a DataFrame and returns one, unit test that in isolation, and hope the wrapper around it, the actual dp.table decorator, the Auto CDC flow, the expectation, behaves the same way once deployed. Or skip testing the logic directly, stand up a whole separate dev pipeline pointed at a test schema, and eyeball the output table by hand.
If you build proposals, RFPs (requests for proposal), or sizing estimates for Databricks workloads, this is worth a look. Lakemeter turns workload assumptions into structured, exportable cost estimates in minutes instead of days, with full transparency into every pricing assumption behind the number.
How the new ai_mask() function redacts names, emails, and more from free text, and what you need to run it.