Databricks Bundle Inspector
Find Databricks bundle problems before you deploy, including ones databricks bundle validate misses. See them in your code, at the exact line, and explore how your jobs and tasks connect.
It checks your Declarative Automation Bundle (formerly Databricks Asset Bundle) for missing files, broken task dependencies, notebook parameters a task doesn't pass, and more. Findings go to the Problems panel. The graph shows each job's tasks and how they depend on each other. A pipeline appears as a task in the job that runs it; the graph doesn't show the tables inside a pipeline, or a pipeline no job runs.
Available on the VS Code Marketplace and on Open VSX for Cursor and Windsurf.
What it catches
Problems that pass databricks bundle validate and only fail when you deploy or run the job:
- Missing files and libraries that tasks and pipelines point at.
- Files that won't be deployed, because
.gitignore or sync.exclude skips them.
include mistakes: a pattern with a typo, or a resource file no pattern covers.- Notebook and file mix-ups: a path without its extension, or a notebook where a file is expected.
- Widgets a notebook reads that the task doesn't pass.
- Task values read from a task that isn't upstream, or that never sets them.
- Secret scopes named by their resource key instead of their name.
- Unresolved variables for the selected target.
Where it can, it suggests the fix: "Did you mean resources/*.yml?" For a notebook path missing its extension or a mistyped include pattern, use the light bulb to apply a suggested Quick Fix. Click a finding's code in the Problems panel to read what it means and how to fix it.
The inspector also reports Not checked when a variable prevents it from checking a local file. Checks on code read Python and SQL. Scala and R are not read.
Every check, and what code is read
What it shows
- Job graph: tasks and what they depend on, with pan, zoom and search.
- Task details: type, source file, parameters, compute, linked pipelines and problems labelled Error, Warning or Note.
- Pipelines: pick a pipeline in the graph's dropdown to see the jobs in the bundle that run it, its default catalog and schema, each source in
libraries with whether it was found, and its problems, all from the YAML. It doesn't show the tables inside a pipeline. A bundle with only pipelines opens on its first one.
- Secrets and widgets found in each task's code.
- Targets: pick a target in the graph to see its variables and issues resolved for it. The Problems panel keeps showing the structural preview, which needs no login.
- Review report: copy a short summary for chat, or a full report for a pull request with a picture of the job's tasks, its errors and warnings, and tables of tasks, parameters and files.
- Sidebar: the Databricks Bundle Inspector icon in the Activity Bar lists every bundle in the folder with its result, the targets each can be checked against, and the Databricks CLI and sign-in profile the checks use.
Getting started
- Install the Databricks CLI,
v0.299.0 or newer.
- Open a folder that contains a
databricks.yml or databricks.yaml. The extension checks the bundle by itself, lists what it finds in the Problems panel, shows the count in the status bar, and checks again every time you save.
- For the graph, open the bundle file or a resource file it loads through
include, then click Open Graph (the cube and magnifier button) in the editor title bar. From a resource file, the graph opens on the job under your cursor, or the file's first job.
If the CLI can't be found, Open Graph says why and links to the install steps for your OS. Besides your PATH, the inspector looks for the CLI the Databricks extension ships and, on Windows, in the Path Windows has stored, so a WinGet install works without a restart. Otherwise restart VS Code so it sees the new PATH, or click Locate CLI… and pick the CLI file.
Setup and usage in detail
Safety
- It runs one program, the Databricks CLI you installed, to validate the bundle. To find it, Windows only, it may also read (never change) your stored
Path with reg.exe. Nothing is deployed and your code is never run, including scripts and Python a bundle asks the CLI to run.
- It writes nothing into your bundle folder, not even the
.databricks folder the CLI creates for a real target.
- It stays off until you tell VS Code you trust the folder.
- The checks that run by themselves need no login. Picking a real target in the graph uses your active CLI profile, or the one set in
databricksBundleInspector.profile, and may contact your workspace. With a login, the CLI's validate also creates the target's empty files folder in your workspace if it is missing; nothing is uploaded. A target that sets no workspace.host uses the workspace of your default profile. If the workspace doesn't answer within 15 seconds, the graph shows the structural preview instead and says why.
See the security policy.
Telemetry
The extension collects usage data, such as which features are used. Never file paths, names or contents. Nothing is sent until you have closed the notice about it and opened VS Code again.
To turn it off, use either setting:
databricksBundleInspector.telemetry.enabled: set to false, for this extension only.telemetry.telemetryLevel: anything other than all, for VS Code and every extension that respects it.
What is collected, and where it goes
Project status
Latest release: 0.6.2. This README also describes changes on main that are awaiting release; see Unreleased changes.
Active development. Feedback, bug reports, and feature requests are welcome on the issue tracker.
License
Apache-2.0. See LICENSE.
