RFC: Add support for user defined part/partition data skipping indices about clickhouse

[UnamedRus]

Duplicate

#33583

[josh-hildred]

@UnamedRus I think both solutions would solve similar problems, but there are differences in what I propose. #33583 proposes allowing the user to change the order that ClickHouse uses indices (including the primary index). Whereas I proposes adding a completely new index that is always defined at the part/partition level and is used at the same time as the MinMax index over the partitioning key (In MergeTreeDataSelectExecutor::filterPartsByPartition). This index would only be available when partitioning.

[CurtizJ]

Also see #64210. I think the idea is pretty similar.

[josh-hildred]

@CurtizJ Yes, I agree the idea is also similar. I think all three issues agree on the fact that ClickHouse performance could be improved by allowing user defined indices to be used earlier (I.e. to filter parts or partitions) when reading data. Some discussion on what a) makes the most sense for ClickHouse and b) would be most useful to users would be interesting.