PR #30288 results in 5.4%-10% performance regression in read-heavy workloads

[Yui-Song]

Bug Report

PR #30288 results in 5.4%-10% performance regression in read-heavy workloads

1. Minimal reproduce step (Required)

1. Deploy a cluster with nightly version
2. Turn on Prepare plan cache
3. Run sysbench oltp_read_only or other read-heavy workloads

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

3. What did you see instead (Required)

The performance of workloads drops 5.4%-10% compared with their baseline.
You can take commit 7555536 as baseline.

4. What is your TiDB version? (Required)

commit 4dc385c

[bb7133]

Interesting... we have found another performance regression(#30936) and thought it was resolved by #30937.

[Yui-Song]

The performance of sysbench read-heavy workloads like: oltp_read_write, oltp_read_only, select_random_points have been improved by #30937. However, the performance of shenma, a read-heavy bank workload, still drop 8.9%-10%. It's a CPU-consuming workload.

[Yui-Song]

Please refer to the internal OLTP Benchmark Overview page for more details.

[github-actions]

Please check whether the issue should be labeled with 'affects-x.y' or 'fixes-x.y.z', and then remove 'needs-more-info' label.