Lakebase Streamlit App Accelerator¶

The Lakebase Streamlit App accelerator scaffolds a Streamlit analytics app hosted directly on Databricks, connected to the TPCH sample dataset via a SQL warehouse and a Lakebase (managed Postgres) instance for master data management. No external hosting or infrastructure needed — the app runs inside your Databricks workspace and inherits workspace authentication automatically.

What is a Databricks App?¶

Databricks Apps is a serverless hosting environment for Python web applications inside your workspace. You deploy code, Databricks handles compute, scaling, and authentication. Users access the app through a workspace URL and are authenticated via their Databricks identity — no separate auth system required.

Streamlit is the Python library used here. It lets you build interactive dashboards with plain Python: write a script, and Streamlit renders charts, tables, filters, and metrics as a web UI.

What gets generated¶

lakebase-streamlit-app/
├── databricks.yml              # Asset Bundle root config
├── .gitignore
├── app/
│   ├── app.py                  # Streamlit application
│   ├── app.yaml                # Databricks App entrypoint config
│   └── requirements.txt        # Python dependencies
└── resources/
    ├── apps/
    │   └── app.yml             # Databricks App resource definition
    ├── postgres_projects/
    │   └── lakebase.yml        # Lakebase project (managed Postgres)
    ├── postgres_branches/
    │   └── production.yml      # Production branch (protected, no expiry)
    └── postgres_endpoints/
        └── primary.yml         # Read/write endpoint (autoscaling)

Two-step deploy¶

Lakebase resources must exist before the app can connect to them. The first deploy provisions the Postgres project, branch, and endpoint. Then you copy the connection details from the Databricks UI and redeploy.

Step 1 — provision Lakebase:

dpa init lakebase-streamlit-app
cd lakebase-streamlit-app
databricks bundle deploy --target dev

After the deploy completes, open the Databricks UI under Lakebase and copy:

• PGHOST — the endpoint hostname (from the Connect modal)
• ENDPOINT_NAME — the endpoint resource path (from the Computes tab, e.g. projects/.../branches/.../endpoints/primary)

Step 2 — deploy the app with connection details:

databricks bundle deploy --target dev \
  --var="lakebase_pghost=<host>" \
  --var="lakebase_endpoint=<endpoint-path>"

The app will appear under Apps in the Databricks UI.

What the app shows¶

The app has two tabs.

Analytics — five views backed by queries against samples.tpch via a SQL warehouse:

• KPI row — Total Revenue, Total Orders, Avg Order Value, Unique Customers
• Monthly Revenue — area chart of revenue by month
• Revenue by Market Segment — bar chart across the five TPCH market segments
• Orders by Status — donut chart of Open / Fulfilled / Pending orders
• Revenue & Discount by Ship Mode — grouped bar chart of gross vs. net revenue per shipping mode
• Top 20 Customers — table with customer name, segment, nation, revenue, orders, avg order value

Results are cached for one hour (@st.cache_data(ttl=3600)), so repeated page views don't re-query the warehouse.

Master Data — an editable table backed by Lakebase (Postgres):

• Loads the top 50 customers from samples.tpch.customer and merges with any overrides stored in Lakebase.
• Users can edit Display Name and Market Segment inline via st.data_editor.
• Saving upserts changes into a customer_master table in Lakebase. The table is created automatically on first load.

This demonstrates using Lakebase as a low-latency transactional layer alongside the analytical SQL warehouse — a pattern common in master data management and operational apps.

Authentication¶

The app connects to Lakebase using OAuth tokens issued by the Databricks SDK. Tokens are rotated automatically on each connection (tokens expire after 1 hour) using a custom psycopg3 connection class:

class _OAuthConn(psycopg.Connection):
    @classmethod
    def connect(cls, conninfo="", **kwargs):
        cred = WorkspaceClient().postgres.generate_database_credential(endpoint=ENDPOINT_NAME)
        kwargs["password"] = cred.token
        return super().connect(conninfo, **kwargs)

The DATABRICKS_CLIENT_ID env var is auto-injected by Databricks Apps as the service principal identity and is used as the Postgres username.

Variables¶

Customising the app¶

The app lives in app/app.py. Add Python packages to app/requirements.txt and redeploy — the runtime installs them on app startup.