A small number of TiFlash replica cannot be created

[CabinfeverB]

Bug Report

What did you do?

Enable shuffle-region-scheduler
create some new tables and create TiFlash replica for these tables.

What did you expect to see?

All table's TiFlash replicas will be AVAILABLE normally.

What did you see instead?

A TiFlash replica is UNAVAILABLE.

And found region info

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

v6.2.0-alpha-146-g99e38eb7

[CabinfeverB]

I found the root cause of this problem. When the region is in the joint state due to executing operator such as move-peer, at this time, if the placement rule is created and the key range of the rule is smaller than that of the region, The Joint Checker fails to match the rule during building Opearator, while the Split Checker skips the region because it is in joint state, resulting in a deadlock like situation.

[CabinfeverB]

following code return > 0 when region doesn't cross multi range , otherwise return -1:

pd/server/schedule/rangelist/range_list.go

Lines 56 to 64 in 750aada

	func (l List) GetData(start, end []byte) (index int, data []interface{}) {
	i := sort.Search(len(l.segments), func(i int) bool {
	return bytes.Compare(l.segments[i].startKey, start) > 0
	})
	if i == 0 || i != len(l.segments) && (len(end) == 0 || bytes.Compare(end, l.segments[i].startKey) > 0) {
	return -1, nil
	}
	return i - 1, l.segments[i-1].data
	}