[SPARK-31312][SQL][2.4] Cache Class instance for the UDF instance in HiveFunctionWrapper #28086
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
…unctionWrapper This patch proposes to cache Class instance for the UDF instance in HiveFunctionWrapper to fix the case where Hive simple UDF is somehow transformed (expression is copied) and evaluated later with another classloader (for the case current thread context classloader is somehow changed). In this case, Spark throws CNFE as of now. It's only occurred for Hive simple UDF, as HiveFunctionWrapper caches the UDF instance whereas it doesn't do for `UDF` type. The comment says Spark has to create instance every time for UDF, so we cannot simply do the same. This patch caches Class instance instead, and switch current thread context classloader to which loads the Class instance. This patch extends the test boundary as well. We only tested with GenericUDTF for SPARK-26560, and this patch actually requires only UDF. But to avoid regression for other types as well, this patch adds all available types (UDF, GenericUDF, AbstractGenericUDAFResolver, UDAF, GenericUDTF) into the boundary of tests. Credit to cloud-fan as he discovered the problem and proposed the solution. Above section describes why it's a bug and how it's fixed. No. New UTs added. Closes apache#28079 from HeartSaVioR/SPARK-31312. Authored-by: Jungtaek Lim (HeartSaVioR) <[email protected]> Signed-off-by: Wenchen Fan <[email protected]>
|
Test build #120655 has finished for PR 28086 at commit
|
Contributor
Author
|
To reduce the code divergence I just enabled language feature (as compiler guided on error message). There seems to be the way we don't enable it (simply use AnyRef instead of wildcard) so that's up to our preference. |
maropu
approved these changes
Apr 1, 2020
|
Test build #120656 has finished for PR 28086 at commit
|
cloud-fan
pushed a commit
that referenced
this pull request
Apr 1, 2020
…HiveFunctionWrapper ### What changes were proposed in this pull request? This patch proposes to cache Class instance for the UDF instance in HiveFunctionWrapper to fix the case where Hive simple UDF is somehow transformed (expression is copied) and evaluated later with another classloader (for the case current thread context classloader is somehow changed). In this case, Spark throws CNFE as of now. It's only occurred for Hive simple UDF, as HiveFunctionWrapper caches the UDF instance whereas it doesn't do for `UDF` type. The comment says Spark has to create instance every time for UDF, so we cannot simply do the same. This patch caches Class instance instead, and switch current thread context classloader to which loads the Class instance. This patch extends the test boundary as well. We only tested with GenericUDTF for SPARK-26560, and this patch actually requires only UDF. But to avoid regression for other types as well, this patch adds all available types (UDF, GenericUDF, AbstractGenericUDAFResolver, UDAF, GenericUDTF) into the boundary of tests. Credit to cloud-fan as he discovered the problem and proposed the solution. ### Why are the changes needed? Above section describes why it's a bug and how it's fixed. ### Does this PR introduce any user-facing change? No. ### How was this patch tested? New UTs added. Closes #28086 from HeartSaVioR/SPARK-31312-branch-2.4. Authored-by: Jungtaek Lim (HeartSaVioR) <[email protected]> Signed-off-by: Wenchen Fan <[email protected]>
Contributor
|
thanks, merging to 2.4! |
Contributor
Author
|
Thanks for the quick review and merge! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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 patch proposes to cache Class instance for the UDF instance in HiveFunctionWrapper to fix the case where Hive simple UDF is somehow transformed (expression is copied) and evaluated later with another classloader (for the case current thread context classloader is somehow changed). In this case, Spark throws CNFE as of now.
It's only occurred for Hive simple UDF, as HiveFunctionWrapper caches the UDF instance whereas it doesn't do for
UDFtype. The comment says Spark has to create instance every time for UDF, so we cannot simply do the same. This patch caches Class instance instead, and switch current thread context classloader to which loads the Class instance.This patch extends the test boundary as well. We only tested with GenericUDTF for SPARK-26560, and this patch actually requires only UDF. But to avoid regression for other types as well, this patch adds all available types (UDF, GenericUDF, AbstractGenericUDAFResolver, UDAF, GenericUDTF) into the boundary of tests.
Credit to @cloud-fan as he discovered the problem and proposed the solution.
Why are the changes needed?
Above section describes why it's a bug and how it's fixed.
Does this PR introduce any user-facing change?
No.
How was this patch tested?
New UTs added.