Overly frequent storage stats reporting causing high tail latency when dataset is large #18955
Description
Bug Report
TiKV reports store size every 1 second, since it requires reading SST properties with RocksDB DB-wide mutex held, serving tail latency got impacted significantly when single-instance dataset is huge.
What version of TiKV are you using?
What operating system and CPU are you using?
Steps to reproduce
What did you expect?
Background activities have limited impact on online traffic.
What did happened?
Tail latency increase, QPS drop