[SPARK-49308][CONNECT] Support UserDefinedAggregateFunction in Spark Connect Scala Client #49785
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
HyukjinKwon
approved these changes
Feb 4, 2025
|
Merging to master/4.0 |
asfgit
pushed a commit
that referenced
this pull request
Feb 4, 2025
…Connect Scala Client ### What changes were proposed in this pull request? This PR adds support for `UserDefinedAggregateFunction` to the Spark Connect Scala Client. While this is a deprecated feature, we still believe it is useful to support it to ensure we reduce incompatibilities between classic and connect. Implementation wise I opted to convert the `UserDefinedAggregateFunction` to an `Aggregator`, and use that code path for execution. This is probably not as fast as the original implementation (more allocations). ### Why are the changes needed? This reduces friction between the classic and connect implementations. ### Does this PR introduce _any_ user-facing change? Yes. It enabled Spark Connect Scala Client users to use `UserDefinedAggregateFunction`s. ### How was this patch tested? Added tests to ### Was this patch authored or co-authored using generative AI tooling? No. Closes #49785 from hvanhovell/SPARK-49308. Authored-by: Herman van Hovell <herman@databricks.com> Signed-off-by: Herman van Hovell <herman@databricks.com> (cherry picked from commit 4953a9c) Signed-off-by: Herman van Hovell <herman@databricks.com>
eejbyfeldt
reviewed
May 16, 2025
| NewInstance(cls, arguments, Nil, propagateNull = false, dt, outerPointerGetter)) | ||
|
|
||
| case AgnosticEncoders.RowEncoder(fields) => | ||
| val isExternalRow = !path.dataType.isInstanceOf[StructType] |
There was a problem hiding this comment.
Is it really safe to call dataType here? The path expression might not be resolved and then this will throw an exception.
There was a problem hiding this comment.
It should be. If you don't know the dataType at this point, then you can't build a deserializer.
There was a problem hiding this comment.
The problem comes up if you have a RowEncoder being used inside a ProductEncoder. The the path in the recursion will come from
and then
UnresolvedExtractValue and the .dataType will throw
org.apache.spark.sql.catalyst.analysis.UnresolvedException: [INTERNAL_ERROR] Invalid call to dataType on unresolved object SQLSTATE: XX000
at org.apache.spark.sql.catalyst.analysis.UnresolvedExtractValue.dataType(unresolved.scala:939)
at org.apache.spark.sql.catalyst.DeserializerBuildHelper$.createDeserializer(DeserializerBuildHelper.scala:411)
Is there some assumption somewhere that the encoders should not be fully composable and RowEncoder can only be used it certain cases?
There was a problem hiding this comment.
@hvanhovell Created this PR #51319 that fixes the issue.
asf-gitbox-commits
pushed a commit
that referenced
this pull request
Oct 1, 2025
### What changes were proposed in this pull request? This fixes support for using a RowEncoder inside a ProductEncoder. ### Why are the changes needed? The current does a dataType check on a path when contructing the RowEncoder deserializer. But this is not safe and if the RowEncoder is used inside a ProductEncoder, it will throw because the path Expression is unresolved. The check was introduced in #49785 ### Does this PR introduce _any_ user-facing change? Yes, it makes it possible to use RowEncoder in more cases. ### How was this patch tested? Existing and new unit tests. ### Was this patch authored or co-authored using generative AI tooling? No Closes #51319 from eejbyfeldt/SPARK-52614. Authored-by: Emil Ejbyfeldt <emil.ejbyfeldt@choreograph.com> Signed-off-by: Herman van Hovell <herman@databricks.com>
eejbyfeldt
added a commit
to eejbyfeldt/spark
that referenced
this pull request
Oct 2, 2025
This fixes support for using a RowEncoder inside a ProductEncoder. The current does a dataType check on a path when contructing the RowEncoder deserializer. But this is not safe and if the RowEncoder is used inside a ProductEncoder, it will throw because the path Expression is unresolved. The check was introduced in apache#49785 Yes, it makes it possible to use RowEncoder in more cases. Existing and new unit tests. No Closes apache#51319 from eejbyfeldt/SPARK-52614. Authored-by: Emil Ejbyfeldt <emil.ejbyfeldt@choreograph.com> Signed-off-by: Herman van Hovell <herman@databricks.com>
asf-gitbox-commits
pushed a commit
that referenced
this pull request
Oct 6, 2025
This is backport of SPARK-52614 #51319 to branch-4.0 ### What changes were proposed in this pull request? This fixes support for using a RowEncoder inside a ProductEncoder. ### Why are the changes needed? The current does a dataType check on a path when contructing the RowEncoder deserializer. But this is not safe and if the RowEncoder is used inside a ProductEncoder, it will throw because the path Expression is unresolved. The check was introduced in #49785 ### Does this PR introduce _any_ user-facing change? Yes, it makes it possible to use RowEncoder in more cases. ### How was this patch tested? Existing and new unit tests. ### Was this patch authored or co-authored using generative AI tooling? No Closes #52503 from eejbyfeldt/SPARK-52614-4.0. Authored-by: Emil Ejbyfeldt <emil.ejbyfeldt@choreograph.com> Signed-off-by: Herman van Hovell <herman@databricks.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What changes were proposed in this pull request?
This PR adds support for
UserDefinedAggregateFunctionto the Spark Connect Scala Client. While this is a deprecated feature, we still believe it is useful to support it to ensure we reduce incompatibilities between classic and connect.Implementation wise I opted to convert the
UserDefinedAggregateFunctionto anAggregator, and use that code path for execution. This is probably not as fast as the original implementation (more allocations).Why are the changes needed?
This reduces friction between the classic and connect implementations.
Does this PR introduce any user-facing change?
Yes. It enabled Spark Connect Scala Client users to use
UserDefinedAggregateFunctions.How was this patch tested?
Added tests to
Was this patch authored or co-authored using generative AI tooling?
No.