HarborSQL

HarborSQL is an MIT-licensed, self-hosted alternative to Databricks SQL Warehouse for supported read-only queries over Unity Catalog Delta tables.

It accepts Databricks SQL connector-style requests, forwards the caller's Databricks bearer token to Unity Catalog, vends short-lived table credentials, opens Delta data with delta-rs, executes read-only SQL with DataFusion, and returns Databricks-compatible result sets.

Capabilities

• Read-only SELECT queries
• Unity Catalog Delta table discovery and authorization
• AWS S3-backed temporary table credentials vended by Unity Catalog
• Databricks SQL connector and JDBC driver compatibility for a focused Thrift-over-HTTP surface
• DataFusion query execution with Delta Lake table providers
• Bounded in-memory result materialization
• Structured tracing, request IDs, and Prometheus metrics
• Stable client-facing error codes with centrally redacted internal logs

Pre-1.0: HarborSQL is pre-1.0 software. The compatibility surface is intentionally defined and still growing. Cloud Fetch, durable or streaming result storage, and broader Databricks SQL protocol coverage are not yet part of the default runtime.

Start Here

• Databricks SQL Warehouse Alternative compares HarborSQL with managed warehouse compute and identifies the workloads that fit.
• Getting Started covers the main migration path.
• SQL Warehouse Migration compares the same Unity Catalog table through both endpoints, including values and metadata.
• Databricks JDBC Driver covers Java/JDBC clients, driver versions, and auth options.
• Advanced Usage covers one-off query mode.
• How It Works explains the request path through Unity Catalog and Delta.
• Configuration lists environment variables and defaults.
• Result Types documents the current Arrow to Thrift mapping.