PR#56093 may cause 8.8% performance regression in benchmarksql

[owlsome2501]

Bug Report

1. Minimal reproduce step (Required)

1. Deploy a cluster with 3 TiDB and 3 Tikv.
2. Run benchmark benchmarksql (benchmarksql-1k).

2. What did you expect to see? (Required)

No performance regression.

3. What did you see instead (Required)

#56093 caused QPS regression 8.83% in benchmark benchmarksql (benchmarksql-1k) compared to baseline 60193.12.

description	tidb	tikv	pd	QPS
last success	a21d9ea	tikv/tikv@7c61278	tikv/pd@098b802	57369.65
first regression	4c2ca6f	tikv/tikv@7c61278	tikv/pd@098b802	54875.54

4. What is your TiDB version? (Required)

4c2ca6f

[MyonKeminta]

The PR #56093 introduces tikv/pd#8483 to TiDB.

It's reasonable that this PR does have risk to introduce performance regression in non-current mode (which would be default after finishing tikv/pd#8432). However:

• After I rerun the failed bisect task twice, both of these reruns failed to reproduce the regression.
• According to the previously done microbench (mentioned client: Make tsoStream receives asynchronously tikv/pd#8483 (comment)), the additional overhead introduced by that PR should not be that much.

So I think this report looks quite doubtful.

But anyway, I prepared this PR as a fix to this problem: tikv/pd#8636

[MyonKeminta]

Failed to reproduce the regression after 4 retries in total. Regard this as a misreport for now.