[C++][Acero] Engine fails on long joins

[vkhodygo]

Describe the bug, including details regarding any error messages, version, and platform.

Tried to run some R code, and it fails with the following error:

Error in `compute.arrow_dplyr_query()`:
! Invalid: There are more than 2^32 bytes of key data. Acero cannot process a join of this magnitude

I suppose this problem is related to the engine itself, not to R in particular. Is it possible to get a fix somehow?

Component(s)

C++

[thisisnic]

Thanks for reporting this @vkhodygo. It looks like the sheer size of the join means it's not supported.

There's more information in the comment on this issue here which shows what determines the amount of key data (number of rows, number of join key columns, and data types of the join key columns).

[vkhodygo]

@thisisnic Thanks for getting back to me. I did some reading, and it looks like a common problem. Do you think it'll be resolved any time soon?

[thisisnic]

I think it would depend on #31769 being resolved, and I don't think that anyone is working on this at the moment, so I'm afraid the answer is probably "no".

[vkhodygo]

Well, it looks like a workaround by splitting the data into batches manually is the only solid solution at the moment.

I'll keep my eye on #31769, if you think this issue should be closed please do so.

[thisisnic]

I'll close this as it'll be covered in that other ticket, but if you have encounter any other problems around this, feel free to open another issue!