unbalanced bytes and keys rate between auto scaling node with others

[CheneyDing]

Bug Report

What did you do?

test hot region

TiDB 16 core 8GB 3 nodes
TiKV 8 core 4GB 3 nodes

go-ycsb config

workload=core
recordcount=2500000000
operationcount=100000000000
threadcount=800 
insertstart=0
fieldcount=2
fieldlength=512
readallfields=true
writeallfields=true
fieldlengthdistribution=constant
readproportion=1
updateproportion=0
insertproportion=0
readmodifywriteproportion=0
scanproportion=0
maxscanlength=200
scanlengthdistribution=uniform
requestdistribution=zipfian
table=usertable

What did you expect to see?

balanced read rate bytes and keys between auto scaling node with others

What did you see instead?

unbalanced read bytes and keys rate

What version of PD are you using (pd-server -V)?

ai-sche-4ea1991
todo: test on master
todo: check auto scaling node config

[lhy1024]

Maybe we should test it on the master branch

[CheneyDing]

test result on master-4252d7e , almost the same as before.

[nolouch]

We should improve the balance algorithm, and let it converge better.