Company or project name

No response

Use case

I noticed that CH Q23 is slower than duckdb and doris in ClickBench: https://benchmark.clickhouse.com/#system=+lks|ahi|DcD&type=-&machine=+atl&cluster_size=-&opensource=-&tuned=+n&metric=hot&queries=-

Q23 is a typical topn scenario.

Describe the solution you'd like

When we run order by limit N in the pipeline with a small N, we can generate a filter at runtime and push the filter down to the source operator, thereby reducing the amount of data processed in the pipeline and improving performance

Describe alternatives you've considered

No response

Additional context

No response