ClickBench-like query fails on S3 storage

[scanhex12]

Company or project name

ClickHouse

Describe what's wrong

https://pastila.nl/?00cc5530/ea24c5a593228f4164fe86c58ed9c5b3#WVNtZoqrw69/5YpUrjXe9A==GCM

Does it reproduce on the most recent release?

Yes

How to reproduce

https://pastila.nl/?00cc5530/ea24c5a593228f4164fe86c58ed9c5b3#WVNtZoqrw69/5YpUrjXe9A==GCM

Expected behavior

No response

Error message and/or stacktrace

No response

Additional context

No response

[alexey-milovidov]

I checked why there was a wrong column name.

ClickBench uses the right Parquet files with IsRefresh column. They are named "compatible", created specifically for ClickBench by normalizing data types (so we have only signed integers), renaming columns, validating UTF-8.
This dataset is located at:
https://clickhouse-public-datasets.s3.eu-central-1.amazonaws.com/hits_compatible/hits.parquet

While I was preparing something to run in tests, I took another dataset, because it is loaded in us-east-1, where our tests are run:
https://clickhouse-datasets.s3.us-east-1.amazonaws.com/hits/hits.parquet
This dataset is an unedited variant, which existed before I published a newer version of ClickBench in 2022. That's why it has Refresh column, not IsRefresh . It was renamed to IsRefresh in ClickBench, because some databases couldn't accept the Refresh name.

So the queries with missing columns are incorrect. But the problem is in the LOGICAL_ERROR. This error is used as an assert in code and should not happen. This is what we still have to fix.