StarRocks 3.0 Highlights: Storage‑Compute Separation, New RBAC, and Lakehouse Features

StarRocks 3.0 is a milestone release that builds on more than two years of development and over 80 prior versions. The 1.x series focused on performance through a vectorized execution engine, cost‑based optimizer (CBO), and global low‑cardinality dictionaries. The 2.x series added real‑time and data‑lake capabilities, including Primary‑Key models, support for external lake tables (Iceberg, Apache Hudi, Delta Lake), and a resource‑group engine for fine‑grained workload isolation.

Key Architectural Change

The most significant change in 3.0 is the adoption of a storage‑compute separation model. Data is persisted in remote object storage or HDFS, while local disks serve as caches. Users can dynamically add or remove compute nodes for second‑level scaling and manage cache lifecycles at the table level, achieving performance comparable to integrated architectures.

New Core Features

New RBAC Permission Framework – Provides role‑based access control that reduces authorization overhead. Supports 40+ permission items covering materialized views, resource groups, UDFs, and external catalogs. Default role activation follows the principle of least privilege.

Materialized View Enhancements – Adds offline pre‑computation, supports CTE, SELECT *, UNION operators, improves visibility of view construction, and enables query rewrite for cross join, outer join, and delta join. Automatic partition‑level refresh reacts to Hive catalog changes.

Trino Dialect Compatibility (Preview) – Allows Trino/Presto SQL to be rewritten to StarRocks syntax, leveraging the multi‑catalog feature for seamless lake‑analysis queries.

Richer Primary‑Key Update/Delete Syntax – Extends beyond simple upserts; users can employ CTEs and multi‑table references to perform generic UPDATE/DELETE operations, easing migration from traditional RDBMSes.

Automatic Partition Creation – Supports partition expressions with DATE/DATETIME columns and granularity (year, month, day, hour). Partitions are created on‑the‑fly during data ingestion, eliminating the need for pre‑creation.

Other Optimizations

Operator spill to disk to prevent OOM failures; supports aggregation, join, and sort operations.

AUTO_INCREMENT column for globally unique IDs.

JDBC catalog support (preview).

Query cache expanded to more join scenarios (bucket‑shuffle, broadcast, etc.).

Dynamic adaptive parallelism controlled by pipeline_dop.

CSV import now accepts skip_header, trim_space, enclose, and escape parameters.

Primary‑Key tables can specify separate sort keys for faster non‑PK queries.

Enhanced query‑monitoring via show proc '/current_queries/' and richer large‑query logs.

Improved SQL parsing error messages for clearer diagnostics.

Reference Links

Release Notes 3.0: https://docs.starrocks.io/zh-cn/main/release_notes/release-3.0

Storage‑Compute Separation: https://docs.starrocks.io/zh-cn/3.0/administration/deploy_shared_data

RBAC Overview: https://docs.starrocks.io/zh-cn/3.0/administration/privilege_overview

Materialized View Details: https://docs.starrocks.io/zh-cn/3.0/using_starrocks/Materialized_view

Primary‑Key Update: https://docs.starrocks.io/zh-cn/3.0/sql-reference/sql-statements/data-manipulation/UPDATE

Primary‑Key Delete: https://docs.starrocks.io/zh-cn/3.0/sql-reference/sql-statements/data-manipulation/DELETE

Automatic Partitioning: https://docs.starrocks.io/zh-cn/3.0/table_design/automatic_partitioning