Probably a final mistake about clickhouse

[Algunenano]

There have been several bug (and bug fixes) lately around FINAL. Can you test with the latest 24.4 or provide a reproducer to verify it please?

[handgunman]

I can't update the version at the moment.
And I don't really understand how to reproduce this problem.

I tried to make the same table in fiddle and transfer these 2 records and one more to it, but the situation is not reproduced.

[nickitat]

there are few flags that could be tweaked:

split_intersecting_parts_ranges_into_layers_final
split_parts_ranges_into_intersecting_and_non_intersecting_final
enable_vertical_final

[handgunman]

at the moment new data has been added to the table and there is no discrepancy for the previously specified identifiers.
but there are new ones.
When selecting data with one of them with the split_intersecting_parts_ranges_into_layers_final setting, everything is correct

split_parts_ranges_into_intersecting_and_non_intersecting_final doesn't - change the result
enable_vertical_final - doesn't change the result

All of these settings are set to 1 and I changed them to 0

[nickitat]

could you pls share explain pipeline SELECT uniqExact(id) FROM cheks FINAL with default settings

[handgunman]

please see in file
check.txt

[nickitat]

looks like #62268

backport was merged on Apr 8

you're using a build from April 3
from https://github.com/ClickHouse/ClickHouse/releases/tag/v24.3.2.23-lts

[handgunman]

Ok, we'll have to wait until the update is available on the cloud service.

[nickitat]

there is also a subtle problem that

SELECT id
FROM cheks
WHERE id NOT IN (
SELECT id
FROM cheks
FINAL
)

isn't 100% reliable, because theoretically the subquery and outer query read from different snapshots.