Open singhpk234 opened 3 months ago
@viirya can i pick this up ?
Sure. I've not begun working on this yet. Which one you will work, NestedLoop or Cross Join? In Spark, they are two difference join operators, I think we should have different tickets for them instead of one.
I can start with BNLJ :) !
still working on it, will publish a pr by early next week, apologies for the delay.
No problem. Thank you for working on this.
Note that I found several bugs in current broadcast implementation when trying to enable broadcast by default in #213. Since BroadcastNestedLoopJoinExec uses broadcast, I suggest that you can work on the operator after #213 is merged.
I suggest that you can work on the operator after https://github.com/apache/arrow-datafusion-comet/pull/213 is merged
Ack, let me rebase with this branch meanwhile
What is the problem the feature request solves?
Datafusion supports Cross and NestedLoop joins as well : https://docs.rs/datafusion-physical-plan/36.0.0/datafusion_physical_plan/joins/index.html
It will really nice if we can add support for it like Hash and SortMergeJoin.
Describe the potential solution
No response
Additional context
SHJ pr