Visorgood commented on PR #6110:
URL:
https://github.com/apache/datafusion-comet/pull/6110#issuecomment-6725297579
@sunchao both fixed.
The two `s"..."` strings were already fixed locally when you looked – I had
not pushed, so the CI log you linked was from the October 24 head. That fix and
the DataSource V2 test update are both in this push.
The `checkPlans` gap is real. It asked separately whether any exchange was
columnar and whether any exchange had a rejected key, which the aggregate query
could satisfy with two different nodes. It now filters to the exchanges whose
key the removed check rejected and asserts routing on those. Re-ran the
benchmark with the stricter check, so these numbers are now backed by an
assertion that the measured exchange is the gated one:
range, Comet scan -7.2%
range, row-based Spark scan -6.8%
hash, Comet scan -17.5%
hash, row-based Spark scan -14.3%
hash, Comet operator above, Comet -45.2%
hash, Comet operator above, row-based -32.0%
Also merged main, which conflicted in `understanding-comet-plans.md` where
the wording moved to "built-in shuffle". 580 tests passing across the shuffle,
collation, expression, codegen and plan-stability suites; goldens unchanged.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
