Repository navigation
moved SparkOptions to datafusion-spark - #27371
rizzabhdwaj wants to merge 2 commits into
Conversation
|
Thanks for working on this, @rizzabhdwaj. #27337 removed That would be step 1 of #27342: add There is some urgency: I'd like to add more options to that extension, including Spark-version behavior for the |
|
Yeah, absolutely . I'll open a separate PR in apache/datafusion-spark shortly for the changes that belong there. I've also reached out to @AnuragRaut08 about collaborating on this. Hopefully, we can work together and get this issue resolved soon. Thanks for the guidance! |
|
@rizzabhdwaj Can you resolve the conflicts? Rebase if needed. |
3d81900 to
535851e
Compare
535851e to
115838b
Compare
|
I've resolved all the conflicts. |
Which issue does this PR close?
issue - #27342
Rationale for this change
Spark-specific configuration was stored in DataFusion core even though it is only consumed by Spark-compatible functions. This caused Spark options to appear in every DataFusion session and required new Spark options to wait for a DataFusion release.
What changes are included in this PR?
SparkOptionsandMapKeyDedupPolicyfrom DataFusion core.ConfigExtensionusing thesparknamespace.SparkOptionsfrom configuration extensions.datafusion.spark.map_key_dedup_policytospark.map_key_dedup_policy.EXCEPTIONandLAST_WINduplicate-key behavior.enable_ansi_modeunchanged because its ownership between DataFusion core and Spark remains unresolved.What is the testing strategy for this PR?
cargo fmt --all -- --check
git diff --check
Are there any user-facing changes?
Yes. The map deduplication option is no longer a built-in DataFusion option.
The old key:
datafusion.spark.map_key_dedup_policy
is replaced by:
spark.map_key_dedup_policy
The new key is available when Spark features are enabled. The default behavior remains EXCEPTION, and LAST_WIN remains available for preserving the previous permissive behavior.