Port remaining parallelizable aggregate functions to off-heap data structures

[puzpuzpuz]

Is your feature request related to a problem?

4097 ported min(str), max(str), as well as count_distinct() for long, int, and IPv4 types to parallel GROUP BY, but some functions remain unported. Namely:

• [x] count_distinct(uuid): requires a new long128 hash set, similar to the GroupByLongHashSet one
• [x] count_distinct(long256): requires a new long256 hash set, similar to the GroupByLongHashSet one
• [ ] approx_percentile(double): this one is tricky as we'll have to port HdrHistogram to become off-heap and flyweight
• [ ] string_agg(str): we already have GroupByCharSink which we can use here
• [x] all first/last and first_not_null/last_not_null functions: to port them, we'll have to access and store row ids in the group by map
• [ ] isOrdered(IPv4)/isOrdered(long) functions: again, we need to track row ids
• [x] ksum/nsum

There is also count_distinct(symbol), but we have early exit logic in that function (see #3974), so we don't want to port it, at least for now.

Describe the solution you'd like.

No response

Describe alternatives you've considered.

No response

Full Name:

Andrei Pechkurov

Affiliation:

QuestDB

Additional context

No response

[nwoolmer]

I would like to pick up the long128/long256 tasks, thank you!

[puzpuzpuz]

@nwoolmer that's awesome, thank you! Look forward to your contribution.

[nwoolmer]

Somebody is interested in HdrHistogram

https://github.com/questdb/questdb/issues/4566#issuecomment-2135704377