[WIP] Don't review: E2e

[GideonPotok]

What changes were proposed in this pull request?

Why are the changes needed?

Does this PR introduce any user-facing change?

How was this patch tested?

Was this patch authored or co-authored using generative AI tooling?

[GideonPotok]

@uros-db adding the map binaryValues to Mode doesn't persist the mappings as desired. It looks to me like the general pattern for accumulating anything in spark aggregation functions is by putting it into an InternalRow and having merge and update take internal row as parameter.

Is there another way you advise? Do I use a class that extends from InternalRow, which would include both the OpenHashMap[AnyRef, Long] as well as our Map[String, String] ? Let me know what the right pattern is because obviously having that element of state be a member of Mode implementation is not the standard way/ doesn't seem to work.

[uros-db]

Yes, aggregation buffers are InternalRows in Spark, however - I would not suggest extending stuff as you described in order to get this working. This approach turned out to not be as straight-forward as we'd have hoped, and after all - this is yet another approach with an aux structure for proper collation-aware aggregation in Mode, so since we don't really have guarantees of better performance, I'd suggest sticking with GroupMapReduce for now (#46597). Sorry for leading you the wrong way with this one, I hoped this one would turn out better / easier to implement!