Lakehouse Replay

Lakehouse Replay is a Databricks developer experience capability, introduced June 2026.

Automatically replays a sample of read-only serverless workloads against upcoming Databricks Runtime releases to catch regressions before they reach production.

• It quietly re-runs a sample of your read-only serverless workloads against the next Databricks Runtime before you upgrade, so a runtime regression shows up in a shadow run instead of in prod - no setup, your data never leaves the workspace, and the replay compute isn't billed. 🔗
• It is in Public Preview, enabled by a workspace admin from the Previews page, and the workloads it replays never show up in your job run history or query history - only in the audit log. 🔗

⚠ Limitations: Replays only read-only SQL and DataFrame workloads that run on serverless compute (serverless SQL warehouses, notebooks, and jobs) and read Unity Catalog Delta tables, so anything that writes is out of scope; for DataFrame workloads only the Spark plan submitted to production is replayed, not the preceding Python cells. It detects execution failures, not wrong results, and coverage is documented as expanding. 🔗

Category

Developer experience

Introduced

June 2026🔗

Verified

2026-08-20