scheduler: region score are different between log and metrics

[lhy1024]

Bug Report

What did you do?

add a new node

What did you expect to see?

the cluster is balance

What did you see instead?

the cluster is no balance and their region score differ a lot, but pd doesn't create operator.

but log is different with metrics
[2022/09/30 06:54:42.576 +00:00] [DEBUG] [utils.go:124] ["skip balance region"] [scheduler=balance-region-scheduler] [region-id=193894476] [source-store=115121122] [target-store=717892230] [source-size=3230518] [source-score=3590635.5235656626] [source-influence=0] [target-size=68663] [target-score=1193743611.955962] [target-influence=0] [average-region-size=73] [tolerant-resource=16028]

Because, the score of metrics is RegionScore(s.opt.GetRegionScoreFormulaVersion(), s.opt.GetHighSpaceRatio(), s.opt.GetLowSpaceRatio(), 0)
the score of log is RegionScore(opts.GetRegionScoreFormulaVersion(), opts.GetHighSpaceRatio(), opts.GetLowSpaceRatio(), targetDelta)

this makes us cannot know why pd doesn't create operator by metrics.

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

master

[ChenPeng2013]

/severity critical

[matchge-ca]

/assign

[lhy1024]

/severity critical

I don't think it's a critical problem, which makes us cannot know why PD doesn't create operators only by metrics. It has no effect on the scheduler.

When delta is nearly with source or date, there will be the similar case, in another word, the capacity of tikv is small. We need more suitable metrics to diagnose schedule in this case.

[mayjiang0203]

/remove-severity critical
/severity moderate