TiFlash table scan performance gain after mpp column prune enhancement

[JaySon-Huang]

General Question

Ref pingcap/tidb#52143

After pingcap/tidb#52143, the tpc-ds 50 query-72 TableScan cost changed. The total query time of query-72 changed from 5.71s to 3.72s, and the TableScan of catalog_sales changed from 3.21s to 1.72s.

However, the data bytes read from Storage layer does not affected. We need to investigate why the cost of "TableScan" is changed.

[JaySon-Huang]

The reason is cpu_pending_time is shortened in the new execution plan. However, it is added to the time cost of TableScan.

# old execution plan
[2024/04/25 14:57:19.520 +08:00] [DEBUG] [PipelineExecutor.cpp:111] ["query finish with {\"cpu_execute_time_ns\":13909293292,\"cpu_pending_time_ns\":218585609025,\"io_execute_time_ns\":0,\"io_pending_time_ns\":0,\"await_time_ns\":113212387136,\"wait_for_notify_time_ns\":1879039619}"] [source="MPP<gather_id:1, query_ts:1714028234464527835, local_query_id:3, server_id:1877, start_ts:449322217487466499,task_id:5>"] [thread_id=765]

# new execution plan
[2024/04/25 15:01:04.085 +08:00] [DEBUG] [PipelineExecutor.cpp:111] ["query finish with {\"cpu_execute_time_ns\":13648277584,\"cpu_pending_time_ns\":109346223979,\"io_execute_time_ns\":0,\"io_pending_time_ns\":0,\"await_time_ns\":89383817806,\"wait_for_notify_time_ns\":2883058636}"] [source="MPP<gather_id:1, query_ts:1714028460927203570, local_query_id:2, server_id:275, start_ts:449322276849975299,task_id:5>"] [thread_id=779]