LEARN. PRACTICE. GROW.Concept and mechanism
Git folders supports notebook and file development with branches, commits, diffs, and synchronization with a repository. A local commit does not mean the change reached the remote; pushing and review belong to the collaboration workflow. Avoid making an editable shared folder the implicit production reference. Current documentation recommends Declarative Automation Bundles, formerly Databricks Asset Bundles, and versioned artifacts for production deployments. A bundle describes resources and configuration, including targets for different environments. Keep shared code and make catalog, identity, parameter, and resource differences explicit. Resolve credentials through suitable mechanisms instead of storing them in the repository.
Guided application
In an exercise, the bundle validates, but a query contains a hard-coded development catalog. bundle validate checks configuration against schemas; it does not prove that logic meets its contract or that results belong to the intended environment. Review resolved configuration and perform integration and reconciliation checks in a controlled environment. The deploy command creates or updates resources; run starts execution. Record the commit, artifact version, target, identity, and outcome. If returning to a previous version becomes necessary, assess data changes and external effects too, because reverting code does not automatically undo writes. Promotion should connect the version to concrete evidence of operation and recovery.
Valid configuration can execute wrong logic against the wrong catalog.
Common pitfalls
Commit without push; ignored target; validate treated as functional testing; code rollback treated as data rollback.
Related topics: Platform, compute, and data contracts · Incremental ingestion, state, and schema · Transformation, grain, and quality
Promote an identifiable version with reviewed configuration and reconciled results.
Reference: Declarative Automation Bundles lifecycle · 2026-05-04