Feature Request

Describe your feature request related problem

In the test of issue #3050, it can be observed that the hot-write-move-peer basically disappeared after the PR #3200 was merged when rolling restart TiKV. But there are still many hot-read-move-peer. The latency during rolling restart may still can be improved.

QPS

• 
• Base (The data source is 600 sysbench logs within 10 minutes of the corresponding time)

  tps	17455.70
  tps-jitter-sd	7.66%
  tps-jitter-kmean(k=2)	27.88%
  tps-jitter-rand-percent(p=10.00%)	4.81%
  p99-lat	100.69ms
  p99-lat-jitter-sd	44.28%
  p99-lat-jitter-kmean(k=2)	146.49%
  p99-lat-jitter-rand-percent(p=10.00%)	29.66%

• Rolling restart after merging server: existing statistics will not be deleted when putting store && check when store status changing #3200

  tps	16864.37
  tps-jitter-sd	14.96%
  tps-jitter-kmean(k=2)	59.01%
  tps-jitter-rand-percent(p=10.00%)	8.36%
  p99-lat	103.37ms
  p99-lat-jitter-sd	352.09%
  p99-lat-jitter-kmean(k=2)	2240.71%
  p99-lat-jitter-rand-percent(p=10.00%)	27.73%

Describe the feature you'd like

Reduce the unnecessary scheduling.

Describe alternatives you've considered

Disable move-peer hotspot scheduling during rolling restart

Teachability, Documentation, Adoption, Migration Strategy

• Go