Deploy the app

Deploy the first-party data-profiling Databricks App into any workspace by changing declared inputs only — never by editing tracked application source.

The app is an optional operator companion: guidebook browsing, profiling, comparison, load results. The library, CLI, and committed artifacts remain the product core.

What you need

• A Databricks workspace where you can create Apps and use a SQL warehouse
• Permission to create a catalog schema and volume (or an admin who can)
• A tablespec install path for the app (pip install . from the repo root, or the git pin in apps/data-profiling/requirements.txt)

1. Choose the metadata home

Pick a (catalog, schema, volume) triple where this deployment’s governance tables and run artifacts will live. The app is told that address; it does not scan for a naming convention.

Two deployments may share one metadata home when you want that.

2. Dry-run provision, then provision

From a machine with Databricks CLI/auth:

cd apps/data-profiling
export DATABRICKS_CONFIG_PROFILE=<profile>   # or DATABRICKS_HOST + TOKEN

python scripts/provision.py --dry-run \
  --catalog <catalog> \
  --schema <schema> \
  --volume <volume> \
  --warehouse-id <sql-warehouse-id>

Confirm the printed target, then re-run without --dry-run. Provisioning is idempotent and additive: a second run is a no-op; it never drops columns.

3. Set deployment inputs

In app.yaml (or the Apps UI env form), declare at least:

Optional (absent-tolerant — they hide a surface, they do not crash the app):

Precedence is fixed:

deployment environment  >  connections.yaml metadata:  >  built-in default

Built-in defaults are intentionally generic (main.tablespec_profiler). Do not put a real environment address in tracked source.

4. Grant the app service principal

Find the identity under Compute → Apps → your app → Authorization (also shown in the app sidebar after start).

Missing grants should fail at startup with a message that names the setting and the grant — not mid-click on first query.

5. Deploy

From the repository root (recommended when you want pip install . for tablespec):

databricks sync --profile <profile> . /Workspace/Users/<you>/tablespec
databricks apps deploy <app-name> --profile <profile> \
  --source-code-path /Workspace/Users/<you>/tablespec

The root app.yaml installs tablespec from the repo, then chdir’s into apps/data-profiling so connections.yaml and assets resolve correctly.

From apps/data-profiling/ alone if you prefer the subdirectory requirements.txt git pin for tablespec.

6. Confirm in the UI

1. App process starts without a configuration stack trace.
2. Sidebar shows the resolved metadata location (the declared triple).
3. Guidebook tab can reflect a catalog table or load UMF from a volume path.
4. Profile or Compare against a table the app identity can SELECT.

If startup fails, fix the named setting/grant and redeploy — do not patch literals into application code.

Local smoke (no workspace)

cd apps/data-profiling
PROFILER_RUNTIME=mock \
  PROFILER_METADATA_CATALOG=main \
  PROFILER_METADATA_SCHEMA=tablespec_profiler \
  uv run pytest tests/test_fr23_stack.py tests/test_config.py \
    tests/test_provision.py tests/test_diagnostics.py -q

Or from the repo root: make app-smoke.

Next