Too many empty regions during br-restore

[cosven]

Question

TiDB version is v4.0.11.

1. Restoring 2T tpcc data with 3pd 2db 4kv.

The br version is v4.0.11.

First time

Second time

2. Restoring 2T tpcc data with 1pd 1db 3kv.

The br version is v4.0.12(14d55a7)

Restore finished at 22:47:31.

3. Restoring 2T tpcc data with 1pd 1db 3kv.

The tikv config is

coprocessor.region-split-size: "48M"

The br version is v4.0.11
Restore finished at 03:11:30.

[cosven]

Re-run the first scene (3pd 2db 4kv + br v4.0.11 2t) with customized tikv config

coprocessor.region-split-size: "48M"

Note: I run the go-tpc tpcc run right after br-restore finished. I dont know if it affects the test result.
I think I also run the go-tpc tpcc run in the first scene, so I update the metrics screenshot for the first scene.

[YuJuncen]

After tikv/tikv#9924, this problem is mainly solved:

[cyliu0]

@cosven Is there anything bad happened yet because of this?
@YuJuncen @3pointer Could you please estimate the impact of this bug? Will this one lead to a performance decrease or service unavailable? Feel free to change the severity I just labeled.

[cosven]

Is there anything bad happened yet because of this?

Currently, no.

I think too many empty region has potential influence on tikv CPU usage and pd scheduling efficiency.