# Expected Diagnosis: latency-surge-with-tikv-cpu-imbalance

The expected result should converge on the case-specific mechanism below while
keeping conclusions bounded by the collected evidence.

## Expected conclusion

A high-QPS query on the talk_list table used IndexLookUp and generated heavy back-to-table seeks that saturated the unified read pool. Region leaders for the hot table were concentrated on one TiKV node, so the read amplification produced a node-local CPU spike and latency errors. Leader scattering and a better query/index path are the direct mitigations.

## Required evidence chain

- Identify the affected TiKV node and correlate its CPU saturation with the latency/error window.
- Use TopSQL or execution-plan evidence to show IndexLookUp back-to-table seek amplification on the hot table.
- Show that region leaders for the workload were concentrated on the same node.
- Recommend leader scattering and query/index optimization; a later CPU-aware scheduler can reduce recurrence risk.