Release 482-e STS (18 Aug 2026)#

Starburst Enterprise platform (SEP) 482-e is the follow up SEP release to 481-e. It includes all improvements from the following Trino releases:

This release is a short term support (STS) release.

Breaking changes#

  • Removed support for Alluxio-backed exchange storage for fault-tolerant execution. If you configured Alluxio as the exchange storage provider for fault-tolerant execution, you must migrate to a supported alternative before upgrading.

  • Removed support for the Alluxio file system in the Delta Lake, Hive, Hudi, and Iceberg connectors. This does not affect support for the Alluxio file system cache. If you use the Alluxio file system with any of these connectors, you must migrate to a supported file system before upgrading.

  • Removed the snowflake_jdbc connector. The snowflake_parallel connector is now the only implementation of the Snowflake connector. To avoid disruption, migrate to the snowflake_parallel connector before upgrading.

  • Removed support for creating external Hive tables with the Unity catalog with Hive.

  • The hive.max-initial-splits and hive.max-initial-split-size configuration properties have been removed from the Hive connector. Remove these from your connector configuration before upgrading.

  • The implicit coercion between char and varchar has been reversed. A char value now coerces to varchar with trailing spaces removed, and comparisons between char and varchar now follow varchar semantics with no blank padding, making trailing spaces significant. Review any queries or applications that rely on the previous behavior before upgrading.

  • The lower_bounds and upper_bounds columns of the Iceberg $files metadata table have changed from map(integer, varchar) to typed row values. Update any queries that read these columns before upgrading.

  • The target_max_file_size and parquet_writer_row_group_size Iceberg session properties have been removed, including the deprecated parquet_writer_block_size alias. Remove any references to these session properties before upgrading.

  • Filtering varchar columns in the Synapse connector could return incorrect results where stored values differ only in trailing spaces. Inequality and range predicates are no longer pushed down. A DELETE or UPDATE that relied on pushing down these predicates now fails. Review affected queries before upgrading.

  • The DB2 connector no longer pushes down inequality and range predicates on VARCHAR columns to prevent incorrect results caused by trailing space differences. Equality and IN predicates on VARCHAR columns are now re-evaluated by the engine after retrieval. Review any queries that filter VARCHAR columns in the DB2 connector for potential performance impact before upgrading.

  • File system caching is no longer configured at the catalog level. It is now configured once per node via cache-manager.config-files in config.properties. Move all fs.cache.* settings from catalog properties to the node-level. See File system cache for more information.

  • The DynamicFilter parameter of ConnectorSplitManager.getSplits was removed from the Service Provider Interface (SPI). Replace it with the set of dynamic filter columns, and pass the current predicate to ConnectorSplitSource.getNextBatch as a DynamicFilterSnapshot.

  • The predicate API was removed from the public SPI class io.trino.spi.connector.Constraint. Migrate to the ConnectorExpressionEvaluator interface for evaluating expressions during partition and split pruning.

  • Removed deprecated methods without the ConnectorTableCredentials parameter from ConnectorPageSourceProvider and ConnectorPageSinkProvider in SPI.

  • Removed a deprecated method from TableFunctionProcessorProvider in SPI.

General#

  • Data Products as Code is now generally available.

  • Added public preview support for a local disk tier in the embedded buffer service.

  • Added public preview support for the Iceberg REST API in Starburst Metastore.

  • Added support for AWS IAM authentication for PostgreSQL as a backend service database.

  • Added optional Soft memory limit, Hard CPU limit, and Soft CPU limit configuration fields to built-in resource groups in the Workload management UI.

  • Added support for configuring the embedded buffer service to use the Trino native file system as the spooling storage driver.

  • Added the hive.metastore.thrift.client.max-message-size configuration property, which controls the maximum size of a single Thrift response from the Hive metastore. This property is also available to the HMS to Starburst Metastore migration tool through its properties file.

  • Added optional CSV parsing parameters to the starburst.io.load table function: field_separator, quote_char, escape_char, multiline, and line_separator.

  • Extended Insights data retention to include the data_product_completed_query and data_product_completed_query_relation data product query history tables.

  • Added a client tags filter to the Insights query overview, allowing queries to be filtered by client tag substring.

  • Added Okta and Ping Identity support for Iceberg REST API authentication in Starburst Metastore.

  • Added support for dynamic configuration passthrough for REST catalogs that contain Iceberg tables.

  • Updated the default value of the gateway.routing.type Starburst Control Plane configuration property from CLIENT_TAGS to AUTO_TAGS.

  • Fixed an issue where concurrent modifications could cause a deadlock in Starburst Metastore when using an Oracle database.

  • Fixed an authentication failure where OAuth2 login returned Unauthorized when tag-based routing (gateway.routing.type=AUTO_TAGS) was enabled in the Starburst Control Plane.

  • Fixed the Query Editor so that clicking Query details opens the details view in the same browser tab instead of a new one.

  • Starburst data catalog was renamed to Starburst Metastore. As a result, the following properties have been renamed. These properties are now deprecated and will be removed in a future release:

    • catalog.enabled has been deprecated and renamed to metastore.enabled

    • catalog.default.catalog.id has been deprecated and renamed to metastore.default.catalog.id

    • catalog.default.location.uri has been deprecated and renamed to metastore.default.location.uri

Security#

  • Updated org.eclipse.parsson:parsson to 1.1.8 in the secrets provider plugin to mitigate CVE-2026-9563.

  • Fixed an issue where the Add permissions catalog dropdown in the built-in access control (BIAC) UI did not list any catalogs for users with the sysadmin role, even though SHOW CATALOGS returned catalogs correctly.

  • Fixed an issue in BIAC that forced access control checks to reload the entire biac_role_grant table after role grants or changes, causing the coordinator to stall.

Starburst AI#

  • Added public preview support for downloading tables created by AIDA as CSV files.

  • Added support for returning the businessContext field when retrieving data product details using the Starburst MCP server.

  • Added support for the Google Gemini Enterprise Agent Platform (formerly Vertex AI) AI model provider.

  • Extended the built-in access control audit log to capture AIDA interactions.

  • Added the ai.agent.natural-language-analytics.docs-search-enabled configuration property to include or skip documentation search during SQL generation. Defaults to true.

  • When accessing AIDA through your coordinator URL, AIDA now defaults to the ALL role when you have no previously selected role, rather than public, removing the need to manually select a role to reach data products and AI models. Opening AIDA through the Starburst Enterprise web UI is unaffected.

  • AI model connection details are now managed as trusted connections, and administrating them is restricted to the sysadmin role. Users with the Create privilege on AI models can still create and manage AI models, but can no longer view or edit the connection details, such as endpoints and credentials. On upgrade, the AI storage database is automatically migrated: existing model connection data moves into the Trusted connections tab and the connection columns are removed from AI model storage. This migration cannot be reversed.

BigQuery connector#

  • Added support for partitioned_by table property. Partition columns are now shown in SHOW CREATE TABLE output for partitioned tables.

Great Lakes connector#

Hive connector#

  • Added the table_metastore_statistics system table, exposing whether HMS column statistics are collected for each table.

  • Added support for configuring HTTP connect and read timeouts with the hive.metastore.unity.connect-timeout and hive.metastore.unity.read-timeout configuration properties when using Databricks Unity Catalog as a metastore.

Iceberg connector#

  • Added support for configurable DROP TABLE behavior, using the iceberg.drop-table-mode catalog configuration property.

  • Added support for the system.register_view procedure for registering existing Iceberg views in JDBC and REST catalogs when iceberg.register-view-procedure.enabled=true.

  • Added support for the create_changelog_view procedure, which creates a view that classifies row-level changes between two snapshots as insert, delete, update_before, or update_after.

  • Fixed a query failure where selecting from an Iceberg v2 table with equality deletes on nested fields returned an ICEBERG_BAD_DATA: Multiple entries with same key error.

  • Fixed an issue that caused a loss of materialized view data after running CREATE [OR REPLACE] MATERIALIZED VIEW for a view with a fixed storage location.

  • Starburst data catalog was renamed to Starburst Metastore. As a result, the following properties have been renamed:

    • catalog.iceberg.authentication.type to metastore.iceberg.authentication.type

    • catalog.iceberg.oauth2.issuer to metastore.iceberg.oauth2.issuer

    • catalog.iceberg.oauth2.audience to metastore.iceberg.oauth2.audience

    • catalog.iceberg.oauth2.jwk-url to metastore.iceberg.oauth2.jwk-url The following properties have been both renamed and deprecated, and will be removed in a future release:

    • catalog.iceberg.static-catalog.enabled to metastore.iceberg.static-catalog.enabled

    • catalog.iceberg.catalog-name to metastore.iceberg.catalog-name

    • catalog.iceberg.location to metastore.iceberg.location

MaxCompute connector#

  • Added maxcompute.external-table-fallback-byte-array-decoding configuration property to the MaxCompute connector to control byte array decoding for external tables when the SerDe class cannot be determined from table metadata.

OpenAPI connector#

OpenSearch connector#

  • Added support for multi-field pushdown.

  • Fixed an issue where schemas were not merged when querying wildcard tables with inconsistent field mappings.

Oracle connector#

  • Added ORA_HASH and AUTO values for the oracle.parallelism-type configuration property. ORA_HASH enables hash-based parallel reads on any regular table regardless of partitioning. AUTO automatically selects the best available strategy.

  • The oracle.parallel.max-splits-per-scan property now requires a value of   1 or greater.

Snowflake connector#

  • Added the snowflakeQueryExecutionTime metric to the parallel connector, reporting the time Snowflake spends executing a query. This metric is visible in EXPLAIN ANALYZE VERBOSE and in query stats.

  • Added the snowflake.max-chunk-retries configuration property. This property controls the number of retries for transient HTTP errors that occur when fetching Arrow query result chunks.

  • Fixed an issue where table statistics from one user’s session could be used to plan queries for a different user when using snowflake.impersonation-type=dynamic_connection.

SQL Server connector#

  • Added the sqlserver.transaction-isolation-level configuration property, which sets the JDBC transaction isolation level for connections to SQL Server. The sqlserver.snapshot-isolation.disabled property is now deprecated. Setting both properties in the same catalog causes validation to fail at startup.

Delta Lake connector#