Top SQL is not reporting correctly in the Sysbench PointSelect workload

[breezewish]

Bug Report

Please answer these questions before submitting your issue. Thanks!

1. Minimal reproduce step (Required)

sysbench oltp_point_select prepare --mysql-host=127.0.0.1 --mysql-port=4000 --mysql-user=root --db-driver=mysql --mysql-db=test --threads=400 --time=600 --report-interval=10
sysbench oltp_point_select run --mysql-host=127.0.0.1 --mysql-port=4000 --mysql-user=root --db-driver=mysql --mysql-db=test --threads=400 --time=300 --report-interval=10

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

Top SQL shows exactly the same Top N sql statements for each run.

3. What did you see instead (Required)

The Second Top sql statements changes each time when the cluster is re-deployed. It even shows that CREATE TABLE statement is taking a lot of CPU, which is definitely incorrect.

4. What is your TiDB version? (Required)

master.