Tableau

All integration guides. Tableau Desktop Free Edition 2026.3 with Flight SQL JDBC 19.0.0 passed connection, metadata discovery and a three-row preview. The acceptance scope is read-only; worksheets, aggregates, extracts, joins and dashboards remain untested. See the manual validation record.

Install the driver

Install Tableau Desktop for your platform, then install the JDBC driver below. The recorded run used Free Edition 2026.3.

Start the server and create the sample SQL table. From the repository root:

make bin/flight-sql-jdbc-driver-19.0.0.jar

Copy the JAR to Tableau's driver directory and restart Tableau. The validated macOS setup used ~/Library/Tableau/Drivers/. For other platforms, follow Tableau's JDBC driver locations. The recorded macOS test needed no Tableau JVM changes, TDC overrides or custom connector.

Connect and preview

  1. Select To a Server → More → Other Databases (JDBC).
  2. Enter jdbc:arrow-flight-sql://localhost:8815/?useEncryption=false.
  3. Keep the SQL92 dialect. Leave credentials blank for this local unauthenticated setup.
  4. Sign in and select/type database redis, then schema public.
  5. Add flight_tool_sample and click Update Now in the data preview.

Expect three columns and three rows, including the NULL label on row 2. The historical validation used a different fixture, bi_jdbc_validation, and Flight port 18825; its exact commands and expected values are in the linked record. The guide's shared sample uses the normal port 8815.

Scope and cleanup

Preview success establishes that read path only. Test worksheet-generated SQL, filters and aggregates before depending on them. Custom SQL may be wrapped as a SELECT subquery, so it is not a general DDL/DML editor. No write support is claimed. Authentication and TLS were not covered by the recorded run.

Remove the sample through the fixture cleanup. JDBC sees SQL tables; named Arrow datasets need the native Flight API.

Write workflow and limits

Tableau has no validated Flight SQL write-back route in this repository. Use an external ADBC ingestion job, then refresh the configured read integration. This writes through Python, not through Tableau. Install the Python client environment and create the shared sample first. Ensure id 99 is unused in that disposable sample, then run:

import adbc_driver_flightsql.dbapi as sql
with sql.connect("grpc://localhost:8815", autocommit=True) as conn, conn.cursor() as cur:
    cur.execute("SELECT id FROM flight_tool_sample WHERE id = 99")
    assert cur.fetch_arrow_table().num_rows == 0, "choose a different unused id"
    cur.execute("INSERT INTO flight_tool_sample (id, label, amount) VALUES (99, 'guide', 9.5)")
    cur.fetch_arrow_table()
    cur.execute("SELECT label, amount FROM flight_tool_sample WHERE id = 99")
    assert cur.fetch_arrow_table().to_pylist() == [{"label": "guide", "amount": 9.5}]

Refresh the Tableau read preview after configuring its connection; expect the new row 99 / guide / 9.5. Do not use preview or refresh queries as DML editors. After the read check, remove only the row you inserted through the same Python environment:

import adbc_driver_flightsql.dbapi as sql
with sql.connect("grpc://localhost:8815", autocommit=True) as conn, conn.cursor() as cur:
    cur.execute("DELETE FROM flight_tool_sample WHERE id = 99")
    cur.fetch_arrow_table()
    cur.execute("SELECT id FROM flight_tool_sample WHERE id = 99")
    assert cur.fetch_arrow_table().num_rows == 0

The live guide run passed the external Python write/read-back and cleanup. A fresh native Tableau connection previewed all four rows, including 99 / guide / 9.50000. This validates external ingestion followed by Tableau reading; it does not establish native Tableau writes.