A compute group (warehouse) is a set of compute nodes—compute nodes (CNs) or backend nodes (BEs)—within a StarRocks instance. It provides the compute resources required to run queries, data import jobs, and data processing tasks. Each compute group is an independent resource pool, and compute groups are physically isolated from each other.
Limitations
Default compute group:
A default compute group is created automatically when you create an instance and cannot be deleted.
The default compute group requires at least three running nodes and cannot be scaled to zero. Other compute groups can be scaled to zero nodes.
Multiple compute groups:
Only shared-data instances of StarRocks 3.3 and later support multiple compute groups.
The maximum number of compute groups per instance is 50.
Features
Resource isolation: Divide compute resources (CN nodes only) across compute groups to give each business unit or department its own independent resource pool. This prevents resource contention and ensures operational stability and query reliability.
Shared storage: All compute groups share the same storage layer while keeping compute resources isolated. Different teams and business units access the same underlying data without duplicating it, reducing data redundancy and the risk of inconsistency.
Flexible scaling: Add or remove nodes within a compute group at any time without reallocating data. Scale specific compute groups to meet demand spikes without affecting other compute groups or existing operations.
-
Elastic scaling: Configure scaling rules to automatically scale a compute group up or down based on load. This keeps resources matched to demand during both peak and off-peak periods.
NoteOnly non-default compute groups can be scaled to zero nodes. The default compute group requires a minimum of three running nodes at all times. For more information, see the Limitations section.
Multi-compute-group scalability: Create a new compute group for each new business unit or use case. Every compute group has its own resource pool, so new workloads run independently of existing ones—simplifying operations and maintenance (O&M) in multi-business environments.
Use cases
Workload isolation: Assign different compute groups to different task types for physical resource isolation. For example, use one compute group for ad hoc queries and analytics and another for ETL processing. Each task type uses its own isolated compute resources without interfering with the others.
Cross-department collaboration: Let multiple business departments share the same data assets while each uses a dedicated compute group for queries and analytics. Department workloads remain isolated from each other.
Centralized management: Consolidate multiple instances into a single instance with multiple compute groups to reduce O&M overhead. Consider this approach only when the instances need to share data.