Describe the situation
There is still a performance drop between 22.7.7.24 -> 22.8.1.2097 (and up to the latest 22.12)

I'm running a trivial query "select count() from ... where counter_id = ... and ts > ... and ts < .. format Null settings local_filesystem_read_method='read'." (typical "hits" table).

The query pipeline is identical between the versions (22.7.7.24 and 22.8.1.2097). The slowness comes from "AggregatingTransform" step that's processing rows x3-4 times slower.

Trace logs:
22.7 - https://pastila.nl/?02aa875b/fd42b27162d53ca2a79b137e217b9ddd
22.8 - https://pastila.nl/?02aa875b/d0c917e0b0166c5f8dd3af2d1cc96035

The QPS drop is x2-3 times regardless of the concurrency level used in clickhouse-benchmark tool (same results for -c 1, 8, 16)

How to reproduce
I was not able to reproduce the issue when creating new tables and importing fake data. On the real machine I can upgrade/downgrade between the versions and clearly see the performance difference. I went to the point where one replica in the cluster is on 22.7.7.24 and another is 22.8.1.2097. Practically all queries are slower on 22.8 about x1.5-3 times.

Additional context
I checked all versions after 22.8 up to the latest 22.12, the issue persists.

Perf of 22.7

Perf of 22.8

The query in this case is x1.5 times slower.

Also, the code difference: v22.8.1.2097-lts...v22.7.7.24-stable