✦TL;DR

Learn how to crawl Databricks assets and publish them to Atlan for discovery.

||

Your AI can read this via Docs MCPInstall MCP →Connect

Databricks assets app

The Databricks assets app crawls Unity Catalog assets from Databricks and publishes them to Atlan. Build it with the DatabricksCrawler builder.

Creating an app creates a new connection

Each create mints a new connection. To re-crawl, re-run the existing workflow (see Re-run an existing app).

Personal access token

To crawl Databricks using a personal access token:

Python

Databricks crawling with a personal access token

from pyatlan.client.atlan import AtlanClient
from pyatlan.model.apps import DatabricksCrawler

client = AtlanClient()

response = (
    DatabricksCrawler(client)
    .basic( # (1)
        password="dapi...", # (2)
        http_path="/sql/1.0/warehouses/abc123", # (3)
        host="dbc-1234.cloud.databricks.com", # (4)
    )
    .connection(
        name="production-databricks",
        admin_roles=[client.role_cache.get_id_for_name("$admin")],
    )
    .extraction_strategy("system-tables") # (5)
    .asset_selection( # (6)
        include_hierarchy={"my_catalog": ["sales", "marketing"]},
    )
    .nested_columns("true") # (7)
    .run(name="databricks-prod")
)
print(response.slug, response.run_id)

1. Step 1—Credential. Personal-access-token auth; the token is vaulted.
2. The Databricks personal access token.
3. The SQL warehouse HTTP path.
4. The Databricks workspace host.
5. Step 3—Metadata. system-tables (recommended) queries Unity Catalog's system.information_schema; rest-api uses the Databricks SDK.
6. Asset selection—see Asset selection below.
7. Parse STRUCT/ARRAY/MAP columns into nested child columns.

Service principal (AWS / Azure)

To authenticate with a service principal instead of a token:

Python

Databricks crawling with a service principal

# AWS
DatabricksCrawler(client).aws_service(
    client_id="...", client_secret="...", host="dbc-1234.cloud.databricks.com",
)

# Azure
DatabricksCrawler(client).azure_service(
    client_id="...", client_secret="...", tenant_id="...",
    host="adb-1234.azuredatabricks.net",
)

Asset selection

The asset_selection(...) method mirrors the UI's multi-mode picker. Combine any of the four modes (all optional):

Python

Asset selection—hierarchy and regex

(
    DatabricksCrawler(client)
    .basic(password="dapi...", http_path="...", host="...")
    .connection(name="production-databricks", admin_roles=[...])
    .asset_selection(
        include_hierarchy={"my_catalog": ["sales"]}, # (1)
        exclude_hierarchy={"my_catalog": ["staging"]}, # (2)
        include_regex={"schema": "prod_.*"}, # (3)
        exclude_regex={"table": ".*_tmp$"}, # (4)
    )
    .run(name="databricks-prod")
)

1. Include by hierarchy—{catalog: [schema, ...]} (sent as an object).
2. Exclude by hierarchy—same shape; exclude takes priority over include.
3. Include by regex—{asset_type: regex} for schema or table.
4. Exclude by regex—same; {"table": ...} excludes matching tables/views.

Other metadata options

Python

Additional Databricks configuration

(
    DatabricksCrawler(client)
    .basic(password="dapi...", http_path="...", host="...")
    .connection(name="production-databricks", admin_roles=[...])
    .import_tags("true") # (1)
    .enable_view_lineage("true") # (2)
    .import_ai_models("true") # (3)
    .incremental_extraction("true") # (4)
    .run(name="databricks-prod")
)

1. Sync Unity Catalog tags to Atlan.
2. Build column-level lineage for views from their definitions.
3. Import Databricks ML model and model-version metadata.
4. Only extract assets changed since the last successful run.