Today, we are excited to announce Starburst Enterprise 480-e LTS. This release introduces our commitment to multi-cluster data architecture, deepens our Iceberg leadership, and advances AI readiness while delivering steady gains in performance and usability.
Collectively, 480-e introduces significant improvements in the following areas:
- Introducing Coordinator HA and Cluster Routing afforded by the Starburst Control Plane
- Data Product Sharing
- Iceberg Data Products
- Enhanced Starburst AI Agent capabilities
- Iceberg v3 and Iceberg branching
- Significant performance gains
Starburst helps you manage your multi-cluster deployments
For the release, we are happy to announce a number of improvements to our features. These include.
Coordinator High Availability (HA)
Coordinator High Availability (HA) is introduced as a Public Preview capability. With Coordinator HA, administrators can now deploy a Starburst Enterprise cluster with an additional active-standby coordinator node. If the active coordinator node becomes unavailable for any reason, the Starburst Enterprise Control Plane will automatically detect the condition and orchestrate failover to the standby coordinator, thereby minimizing unexpected downtime.
Load Balancing
Starburst Enterprise’s built-in Load Balancing capability is now in Public Preview. Starburst Enterprise Load Balancing gives administrators an easy, out-of-the-box way to horizontally scale across multiple homogeneous Starburst Enterprise clusters to handle any workload size. Starburst Load Balancing requires deployment of the Starburst Enterprise Control Plane, which is included with Starburst Enterprise as part of the Elite license.
Workload isolation
Starburst Enterprise’s built-in Workload Isolation feature is now in Public Preview. With this feature, inbound queries are automatically routed to different physical Starburst Enterprise clusters based on the source application name, client tags, or other business-relevant attributes, and according to rules predefined by an administrator in the Starburst Enterprise Control Plane UI. This allows for the dedicated allocation of critical resources to prevent disruption. It also prevents the so-called “noisy neighbor” scenario, where one resource steals the resources of another.
Performance Improvements
CTE Reuse
Previously in Public Preview, CTE reuse is now GA. It reduces CPU and I/O costs by avoiding repeated table scans and common subquery computations. As a result, in standard performance benchmarking, the query latency improves up to 20% on average.
Read the official documentation for CTE Reuse.
Supporting Queries Exceeding RAM
Previously, to avoid Out of Memory Errors for large queries, customers needed to either overprovision their clusters or resort to Fault Tolerant Execution (FTE) mode. The former approach was expensive, while the latter was causing noticeable increases in query runtimes. We are now introducing an optimized execution mode that allows queries to exceed available memory resources without incurring much query latency penalty. This feature is currently in Private Preview.
Starburst ODBC Driver
The ODBC v3 driver, first introduced for Windows in our 480-e LTS, is now available in Public Preview for Linux and macOS. Delivering over 400% performance improvements compared to v2, it is a drop-in replacement for BI tools like Tableau and Power BI, as well as for any application connecting via a standard ODBC interface. If you’re running Starburst on Linux or connecting from a macOS workstation, now is a great time to evaluate the upgrade.
Arrow in Spooling Protocol
The spooling protocol can now be configured to use Arrow serialization. That allows Python clients to fetch the outputs of Starburst queries in an optimized, data-parallel format and to pass Arrow chunks for further processing without decoding, which makes the process up to 7x faster. This feature is currently in Private Preview.
Enhanced Starburst AI Capabilities
The 480-e release builds on the AI momentum started in 479-e, specifically regarding the Starburst AI agent. This AI Agent is designed to make your data more discoverable, understandable, and actionable, helping organizations translate raw data into real business outcomes.
Generate insights with natural language
To improve natural-language-to-SQL accuracy, the Starburst AI Agent now supports automated data profiling. When the ai.agent.data-profile.enabled property is activated on the coordinator, the agent automatically collects sample rows and column statistics from data product datasets and embeds this metadata directly into its language model prompts.
This contextual grounding ensures the agent generates highly precise SQL, capturing real data structures for both standard views and materialized views. Platform administrators can further optimize performance and security by configuring specific query timeouts, sample row limits, and cache expiry durations directly in their cluster configuration files.
Collaborate across clusters
To scale data collaboration without forcing consolidation, the Starburst AI Agent and Starburst Control Plane now support seamless data product sharing across clusters. This capability enables data teams to publish a curated, governed data product from a single cluster and make it instantly discoverable and queryable as a read-only asset across other distinct clusters in the global data architecture.
By allowing the AI Agent and distributed teams to interact with remote data products without duplicating underlying tables, organizations eliminate unnecessary data movement. This ensures that business metadata, schema definitions, and a unified source of truth remain consistent across every department, regardless of where the physical compute resources reside.
Iceberg v3 and table branching
The 480-e LTS release doubles down on our open lakehouse architecture by bringing advanced write isolation workflows and native type optimizations to our Apache Iceberg integration.
Snapshot tagging and table branching
Now available in public preview, users can manage and audit complex table branch operations directly through SQL syntax or the Starburst Enterprise REST API. This allows data teams to isolate experimental write pipelines, run historical analytical queries on explicit snapshot tags, and implement a Git-like branching workflow across their data lakehouse.
Iceberg format version 3 interoperability
This release updates the default Trino type mapping for the Iceberg Variant data type from JSON to native variant for Iceberg format version 3 tables. This underlying optimization ensures true cross-engine interoperability and execution compliance when sharing, writing, and processing highly nested semi-structured data profiles across your entire toolchain.
Additional features and performance gains
Multi-statement SQL Jobs
Now generally available for both Starburst Galaxy and Starburst Enterprise, Multi-Statement SQL Jobs allow users to schedule and execute sequential SQL workflows natively within the platform. This updates the previous single-statement model, allowing teams to combine multiple queries into a single automated job where Starburst guarantees each statement completes before the next begins.
This native scheduling removes the infrastructure overhead and fragility of relying on external orchestration tools like Airflow for simple data pipelines. If a statement fails, the job records the exact stopping point for simplified troubleshooting and troubleshooting retries, allowing business analysts and data engineers alike to manage orderly, multi-step workflows without external dependencies.
Data Products as Code
Now in public preview, Data Products as Code allows teams to manage, version, and deploy Starburst data products as self-contained YAML files. This replaces manual UI steps with an engineering-first approach, enabling teams to scaffold, lint, and validate data products offline via a new CLI tool before pushing them to Git-tracked repositories.
For enterprise data teams, this brings software engineering rigor to data asset management. By integrating code reviews, rollbacks, and automated CI/CD pipelines, organizations gain a complete compliance audit trail without disrupting existing UI workflows.
Write to Unity Catalog
Now generally available, our advanced write integration with Databricks Unity Catalog provides a full write capability for Unity Catalog-owned tables across both Delta Lake and Iceberg formats. This expands upon previous read-only limitations, allowing data engineers to execute direct write operations and seamlessly read advanced table configurations such as liquid clustering from within Starburst.
This deep integration eliminates the need to build complex, disjointed pipelines to move data between disparate environments. By serving as a robust bridge between compute engines, Starburst enables organizations to run analytical workloads using their preferred engine while maintaining Unity Catalog as a centralized source of truth.
OpenAPI Connector
Now available in private preview for Starburst Enterprise, the new OpenAPI Connector allows data teams to seamlessly query any OpenAPI-compliant REST API endpoint using standard SQL. This universal gateway replaces the need to build and maintain brittle, custom-coded API integrations for disparate internal operational systems or SaaS applications like Jira, GitHub, and Datadog.
By pointing the connector to an OpenAPI specification file, Starburst automatically interprets the endpoints and exposes them as queryable table functions, mapping JSON schemas directly to SQL data types. This enables organizations to immediately unlock previously siloed data and performantly join live API data with massive existing datasets sitting within their data lakes and warehouses.
Critical and breaking changes
This Starburst Enterprise release includes the following critical and breaking changes, summarized here. Please view the full documentation for more information, as well as the complete list of breaking changes.
Starburst Control Plane deployment update
With the 480-e LTS release, the Starburst Portal has been officially renamed to the Starburst Control Plane. This foundational update goes beyond a new naming convention, introducing core platform resilience capabilities like high availability, load balancing, and auto tag-based routing in public preview.
To support these advanced multi-cluster orchestration features, the underlying admin configuration framework has been modernized. Administrators migrating from legacy Gateway setups should note that old authorization and preset user blocks are now ignored. They must transition to the new explicit access control properties to ensure seamless cluster operations.
Starburst Admin deployment update
The 480-e LTS release updates underlying system requirements and deprecates several legacy infrastructure components managed via Starburst Admin. Notably, the Starburst-packaged Hive Metastore Service (HMS) is supported, but users will need to bring their own image. Additionally, some legacy Hadoop-based file system implementations for cloud object storage have officially reached the end of support.
Administrators must migrate to the Starburst Data Catalog and native file system support before upgrading beyond 480-e. These updates ensure that deployment infrastructure matches modern security standards and leverages parallel processing optimizations for accelerated catalog migrations.
Configuration changes
Several configuration properties have been renamed or removed in 480-e LTS. You must update your configuration files before upgrading to ensure your clusters start successfully:
- Query Management:
query.max-written-data-sizeis renamed toquery.max-write-physical-size. - Hive Connector:
hive.s3.storage-class-filteris renamed tohive.s3-glacier-filter, andhive.file-status-cache-tables.excludedis renamed tohive.file-status-cache.excluded-tables. - Kudu Connector:
kudu.schema-emulation.enabledis removed and replaced bykudu.schema-emulation.type. - Kafka Event Listener: The
kafka-event-listener.client-config-overridesproperty has been removed. - AI Agent Storage: The properties
ai.agent.persona-directory-path,ai.agent.session-storage,ai.agent.max-sessions, andai.agent.session-inactivity-timeouthave been removed. AI sessions are now exclusively handled via the Insights database, and custom personas are managed via the UI. - Snowflake Connector: The legacy
snowflake_jdbcconnectorhas reached end of support. You must migrate your catalog configurations to thesnowflake_parallelconnector before upgrading.
Please view the full documentation for more information.
Upgrade to the 480-e LTS Today
The 480-e LTS release marks a defining shift in how modern enterprises run, govern, and scale their data lakehouses. By embedding mission-critical high availability and load balancing directly into the Starburst Control Plane, platform teams are finally equipped to run multi-cluster architectures with absolute resilience.
Simultaneously, features such as native multi-statement scheduling, Data Products as Code, and deeper Databricks Unity Catalog write integrations remove the heavy data engineering friction that has traditionally slowed innovation. Combined with automated data profiling to ground Starburst’s AIDA in real structural context, 480-e LTS provides the exact performance, automation, and cross-cluster collaboration required to fuel the next generation of agentic enterprise AI workloads.
- View the full release notes here.
- Review our AI documentation here.
- Read more about AIDA.
- Read our recent keynote blog post from Justin Borgman.
