Welcome

Icebreaker is a serverless query platform for the data lakehouse. It has two parts: a managed control plane and a private data plane.

The control plane is a managed service that runs in Icebreaker’s environment, freeing your team from application management. The data plane is where all sensitive information stays and queries are executed; the data plane runs EC2 instances of your choosing completely inside your own VPC.

Icebreaker executes SQL workloads against Iceberg tables, using one Kubernetes pod per query, with cost and performance controls applied to every query before it runs.

These docs are for platform engineers, data platform engineers, or staff SREs responsible for activating Icebreaker in your enviroment.

This is an early-adopter release. Icebreaker today supports:

• Cloud: Amazon Web Services (EC2-based EKS clusters).
• Storage: Apache Iceberg tables (Parquet files) in S3.
• Catalog: Any Iceberg-compatible catalog (BYO — see Catalog integration).
• Protocol: Arrow Flight SQL. Connect with any JDBC-compatible SQL client.
• Workloads: SQL queries against Iceberg.

Support for additional clouds, catalogs, query engines, and AI model integrations is planned. The terminology and structure in these docs are intentionally cloud- and engine-neutral so that future scope fits cleanly.

• Concepts — Read these first to build the right mental model of Icebreaker. Start with How Icebreaker is delivered, then Architecture.
• Get Started — A guided install path from prerequisites to running your first query.
• How-to guides — Targeted instructions for specific operational tasks.
• Reference — Configuration flags, supported surface, and limits.

If you’re approving a pilot rather than running one, the Concepts section should be enough to evaluate fit.