Connect outside tools

Outside tools reach a Sail cluster two ways: HiveServer2 (Thrift) over HTTPS for BI tools and JDBC/ODBC apps, and Spark Connect for PySpark. Every tool signs in with your lakehouse username and password, the same ones Trino uses, and sees only what your access rules allow.

Before you connect

1. Allow outside connections. In the cluster's Edit dialog, set Network access to Specific IPs (your address is filled in when you create the cluster) or Open. A Private cluster is reachable from inside Databasin only, and its connection details say so.
2. Find the connection details. On a running cluster, open Cluster Details on its card: the Thrift host, port 443, HTTP path cliservice, the JDBC URL, the ODBC settings and the Spark Connect URL. The Connection Credentials button (the key icon) in the SQL editor shows your username and password.

Tableau

Use the Spark SQL connector with the Simba Spark ODBC 2.9 driver. The newer Databricks-branded ODBC drivers (2.12 and later) only connect to Databricks.

• Type: SparkThriftServer. Transport: HTTP. SSL: on.
• Host and port: the Thrift host, port 443.
• HTTP path: cliservice. To land in one catalog, use cliservice/<catalog> (or cliservice/<catalog>/<schema>) instead of Initial SQL.
• Authentication: Username and Password, with your lakehouse credentials.

Power BI

Power BI connects with its Spark connector, using the HTTP protocol and the same host, port and HTTP path as Tableau.

Power BI's Spark connector doesn't wait for a sleeping cluster, so the cluster may need a moment to wake before Power BI can connect. Keeping the coordinator always on (the default) helps. If you hit problems connecting Power BI, tell us through support.

DBeaver and other JDBC apps

Use the Apache Kyuubi JDBC 1.12.0 driver. Download it from the cluster card; in DBeaver, pick the built-in Apache Kyuubi driver.

• Driver class: org.apache.kyuubi.jdbc.KyuubiHiveDriver
• URL: jdbc:hive2://<host>:443/default;transportMode=http;httpPath=cliservice;ssl=true
• One catalog only: set httpPath=cliservice/<catalog>.

If you connected with DBeaver's Apache Spark driver or the Hive JDBC driver, a schema can show tables from other catalogs with the same schema name, because those drivers don't send the catalog when they list tables. Switch to the Apache Kyuubi driver, or set httpPath=cliservice/<catalog>.

PySpark, Jupyter and VS Code

Copy the Spark Connect URL (sc://<host>:443/;use_ssl=true) from the cluster card, install PySpark 4.2, and add your credentials:

from pyspark.sql import SparkSession

spark = SparkSession.builder.remote(
    "sc://<host>:443/;use_ssl=true;user_id=<lakehouse username>;token=<lakehouse password>"
).getOrCreate()

Python and PySpark on Sail covers versions, UDFs from your own machine, and what's available.

When the cluster is asleep

• Catalog browsing answers at once while the coordinator is on: SHOW, DESCRIBE, information_schema and the metadata calls BI tools make don't wait for workers.
• Queries wait for workers after a quiet period. Workers take about 2–3 minutes to start; a dedicated coordinator that has scaled to zero takes about 2–5 minutes.
• If the coordinator is asleep too, a JDBC connection (Kyuubi or Hive) can fail while the cluster wakes: connect again once it has started (about 2–5 minutes for a dedicated coordinator). The Simba ODBC driver retries on its own.
• A tool that keeps polling keeps the workers awake.
• Scheduled starts wake the cluster before a dashboard refresh. Set them when you edit the cluster.

Where to go next