datafusion-comet icon indicating copy to clipboard operation
datafusion-comet copied to clipboard

[WIP] Add spark 35

Open holdenk opened this issue 1 year ago • 5 comments

Which issue does this PR close?

Adds Spark 3.5

Rationale for this change

Spark 3.5 is the latest Spark release.

What changes are included in this PR?

Adds Spark 3.5 support, this introduces a new per-spark version of shims as part of adding interfacing to the 3.5 APIs. This has the advantage of static compile checks over the existing shim approach which does reflection for everything, but I can switch to the runtime reflection model for consistency if that's preferred.

How are these changes tested?

WIP

holdenk avatar Feb 28 '24 00:02 holdenk

Some class and function has changed in spark 3.5, maybe we need some more shim classes to support both spark 3.5 and the versions before.

I has a similar commit to do this: https://github.com/apache/arrow-datafusion-comet/commit/b038267eb5e6ddafccf84b0147760f6992f36077

wankunde avatar Feb 29 '24 02:02 wankunde

Yup this adds some shims (there are existing shims too for 3.3 etc but they use reflection heavily).

holdenk avatar Feb 29 '24 04:02 holdenk

FWIW I created https://github.com/apache/arrow-datafusion-comet/issues/140 to improve the current Spark shim handling.

sunchao avatar Feb 29 '24 18:02 sunchao

I love it :)

On Thu, Feb 29, 2024 at 10:45 AM Chao Sun @.***> wrote:

FWIW I created #140 https://github.com/apache/arrow-datafusion-comet/issues/140 to improve the current Spark shim handling.

— Reply to this email directly, view it on GitHub https://github.com/apache/arrow-datafusion-comet/pull/127#issuecomment-1971741645, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAAOT5NRC355APXEYYRFIADYV53MJAVCNFSM6AAAAABD5DERESVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSNZRG42DCNRUGU . You are receiving this because you authored the thread.Message ID: @.***>

holdenk avatar Feb 29 '24 18:02 holdenk

FYI https://github.com/apache/datafusion-comet/pull/318 is just merged

kazuyukitanimura avatar Apr 25 '24 06:04 kazuyukitanimura

Thanks for getting this started @holdenk. I have now created https://github.com/apache/datafusion-comet/pull/587 to add experimental support for Spark 3.5.1 (there will be follow up PRs to fix some test failures), so will close this one.

andygrove avatar Jun 18 '24 22:06 andygrove