feat: pass spark.comet.datafusion.* configs through to DataFusion session#3455
feat: pass spark.comet.datafusion.* configs through to DataFusion session#3455andygrove merged 5 commits intoapache:mainfrom
Conversation
…sion Spark configs with the prefix spark.comet.datafusion.* are now passed through to DataFusion's SessionConfig. The prefix spark.comet. is stripped so that e.g. spark.comet.datafusion.sql_parser.parse_float_as_decimal becomes datafusion.sql_parser.parse_float_as_decimal in DataFusion. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
comphead
left a comment
There was a problem hiding this comment.
Thanks @andygrove I'm happy we put back sending those user defined DF configs to the native execution
|
Are there any configs that we can think of that would absolutely break Comet correctness? Either adding to a disallow list to filter out, or warn users about? |
Good point. We could put this functionality behind a new config flag with appropriate documentation. Something like |
At a minimum I think that's a good start. It's a huge config space, so I think guarding it behind another config that says something to the effect of "DataFusion configs are exposed for development purposes and possible workload-specific optimizations. Changing DataFusion configs could affect Comet correctness and/or stability." |
|
Thanks for the review @mbutrovich. I added the config. |
mbutrovich
left a comment
There was a problem hiding this comment.
LGTM, thanks @andygrove!
Summary
spark.comet.datafusion.*are now passed through to DataFusion'sSessionConfigspark.comet.prefix is stripped so that e.g.spark.comet.datafusion.sql_parser.parse_float_as_decimal=truebecomesdatafusion.sql_parser.parse_float_as_decimal=truein DataFusion--confwithout code changes🤖 Generated with Claude Code