After the recovery of a TiKV failure, other TiKV node was detected as slow stores #18605
Description
Bug Report
What version of TiKV are you using?
v8.5.2
What operating system and CPU are you using?
TiDB Cloud
16 vCPU, 64 GiB
2000 GiB Storage × 18 Nodes
Standard storage
Steps to reproduce
1、Simulate random tikv failure, then recover after 10 min
2、Consistently run simulated workload during & after failure
3、After the failure recovery of db-tikv-5, QPS dropped by 50%. db-tikv-2 was detected as a slow store and had its leaders evicted.
What did you expect?
- stable workload qps after failure recovery
- no unexpected evicting slow store
What did happened?
- qps drop 50% for minutes
