@Rocky any update on multi-warehouse support ?

Multi-warehouse support (often referred to as Compute Pools or Sub-clusters) has seen significant updates recently, particularly with the release of StarRocks 4.x. This feature is a core component of the shared-data architecture (compute-storage separation). Here is the current status and recent updates: 1. First-Class Warehouse Support (v4.1+) In the latest versions, "Warehouse" has become a more explicit concept in the SQL layer. You can now manage and switch between different compute clusters directly via SQL: *

SET WAREHOUSE = 'warehouse_name'

: You can explicitly set which compute cluster a session should use. *

SELECT current_warehouse()

: A new utility function to check the effective warehouse for your current session. * Resource Group Integration: Resource groups now support a

warehouse

attribute, allowing you to bind specific users or query types to a dedicated compute pool for physical isolation. 2. Multi-Tenant Data Management To support a robust multi-warehouse environment, StarRocks 4.1 introduced: * Automatic Tablet Splitting/Merging: In shared-data mode, tablets can now split automatically when they become too large or become "hotspots." This ensures that different warehouses can scale and access data more efficiently without manual re-sharding. * Enhanced CPU Isolation: New resource group attributes like

cpu_weight_percent

and

exclusive_cpu_weight

allow for stricter resource guarantees when multiple workloads share the same underlying compute nodes within a warehouse. 3. Architecture Context * Shared-Data Mode required: Multi-warehouse functionality is designed for the shared-data architecture where data lives in remote storage (S3, HDFS, etc.). * Compute Nodes (CN): Warehouses are composed of Compute Nodes (CN). You can scale these nodes up or down independently of your storage, and you can create multiple distinct groups (warehouses) that point to the same dataset. 4. Roadmap & Community Based on the 2025/2026 Roadmap, the focus is now on: * Auto-scaling: Improving the elasticity of these warehouses based on query load. * Unified Governance: Enhancing how different warehouses interact with external catalogs (like Iceberg or Delta Lake) to ensure consistent security and performance across the "Lakehouse." How to get started: If you are on a shared-data cluster (v4.x), you can begin by creating resource groups associated with specific compute pools and using the

SET WAREHOUSE

command to isolate your ELT workloads from your ad-hoc analytics.hoc queries. References • administration/management/resource_management/resource_group.md • quick_start/shared-data.md • sql-reference/sql-statements/cluster-management/resource_group/CREATE_RESOURCE_GROUP.md • data_source/catalog/maxcompute_catalog.md