high-concurrency
Installation
SKILL.md
High-Concurrency Best Practices
Investigation and tuning guide for high-QPS workloads: data modeling, primary-key optimization, query cache, pipeline parallelism, connection pooling, and emergency load disabling.
Five root causes account for most high-concurrency performance problems:
- Cause A — Connection pool exhausted (too many connections, no client pool)
- Cause B — FE planning CPU saturation (complex queries, no plan cache)
- Cause C —
pipeline_doptoo high for short queries (scheduling overhead) - Cause D — Session-level timeout override causing memory volatility
- Cause E — Query cache not effective (low hit ratio, wrong workload type)
Metric Taxonomy — Read This First
Before tuning, establish a baseline measurement of the current state.