What problem does it solve?
This Skill solves StarRocks performance instability in high-concurrency BI workloads, where p99 latency spikes and CPU saturation happen due to excessive parallelism, connection bottlenecks, and missing isolation controls.
Core Features & Use Cases
- Pipeline DOP tuning: Reduce or cap
pipeline_dop globally, per-session, or per-query to prevent thread oversubscription during bursts of small dashboard queries.
- Workload isolation with resource groups: Create and size
short_query resource groups to protect interactive BI traffic from ETL/ad-hoc contention using CPU and concurrency limits.
- Queueing and connection handling: Enable StarRocks query queueing and properly size/route FE connections (including pool sizing and load balancing across FE nodes) to avoid thundering-herd overload.
- Scan and point-lookup optimization: Increase scan parallelism via tablet/bucket distribution for concurrency, and enable Primary Key table direct lookup with persistent index for sub-millisecond key-value access.
- Cache for repeated dashboard patterns: Enable query cache and explain cache eligibility/invalidation so repeated historical dashboard queries return faster.
Quick Start
Apply this skill to StarRocks by setting a lower global pipeline DOP, creating a short_query resource group for BI users, enabling the appropriate query queue settings, and turning on query cache and point-query detail cache for your dashboard/lookup patterns.
