Wayne Gibson
10/07/2026, 5:31 PMWayne Gibson
10/07/2026, 6:21 PMRocky
10/07/2026, 6:21 PMWayne Gibson
10/07/2026, 6:22 PMRocky
10/07/2026, 6:22 PM:::beta warning in the official documentation, meaning it is not yet recommended for critical production workloads without thorough testing.
* Availability: It is currently only supported in Shared-nothing (classic) clusters. Support for Shared-data (cloud-native) clusters is still listed as a "to-be-supported" feature in future releases.
* Refinement: Recent updates (like v3.2.4) added the ability to use ALTER TABLE on hybrid tables and expanded support for complex types (JSON, ARRAY, MAP, STRUCT), showing that the development team is still actively polishing the feature.
Known Limitations
To move to GA, StarRocks typically addresses major architectural gaps. Currently, hybrid storage has these constraints:
* Storage Overhead: It essentially stores data twice (once in columns and once in a hidden __row column), which significantly increases disk usage.
* Writing Performance: Data writing can occasionally block the "short-circuit" point queries during the apply stage, impacting the very low latency it is designed to provide.
* Feature Support: It does not yet support features that rely strictly on columnar mode, such as certain types of partial updates.
Recommendation
If you are looking for high-concurrency point lookups (thousands of QPS) or fast partial updates on Primary Key tables, you can use it today by setting enable_experimental_rowstore = true in your FE configuration and specifying "store_type" = "column_with_row" during table creation.
For updates on the GA timeline, I recommend following the StarRocks GitHub Roadmap or the #starrocks-user channel in the community Slack, where maintainers often share upcoming release milestones.
References
• table_design/hybrid_table.md
• release_notes/release-3.3.mdWayne Gibson
10/09/2026, 3:46 PMKevin Cai
10/09/2026, 3:52 PMWayne Gibson
10/09/2026, 5:08 PM