[SPARK-39496][SQL] Handle null struct in Inline.eval
#36903
Closed
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
Jun 18, 2022
HyukjinKwon
pushed a commit
that referenced
this pull request
Jun 18, 2022
### What changes were proposed in this pull request?
Change `Inline.eval` to return a row of null values rather than a null row in the case of a null input struct.
### Why are the changes needed?
Consider the following query:
```
set spark.sql.codegen.wholeStage=false;
select inline(array(named_struct('a', 1, 'b', 2), null));
```
This query fails with a `NullPointerException`:
```
22/06/16 15:10:06 ERROR Executor: Exception in task 0.0 in stage 0.0 (TID 0)
java.lang.NullPointerException
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.apply(Unknown Source)
at org.apache.spark.sql.execution.GenerateExec.$anonfun$doExecute$11(GenerateExec.scala:122)
```
(In Spark 3.1.3, you don't need to set `spark.sql.codegen.wholeStage` to false to reproduce the error, since Spark 3.1.3 has no codegen path for `Inline`).
This query fails regardless of the setting of `spark.sql.codegen.wholeStage`:
```
val dfWide = (Seq((1))
.toDF("col0")
.selectExpr(Seq.tabulate(99)(x => s"$x as col${x + 1}"): _*))
val df = (dfWide
.selectExpr("*", "array(named_struct('a', 1, 'b', 2), null) as struct_array"))
df.selectExpr("*", "inline(struct_array)").collect
```
It fails with
```
22/06/16 15:18:55 ERROR Executor: Exception in task 0.0 in stage 0.0 (TID 0)/ 1]
java.lang.NullPointerException
at org.apache.spark.sql.catalyst.expressions.JoinedRow.isNullAt(JoinedRow.scala:80)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.writeFields_0_8$(Unknown Source)
```
When `Inline.eval` returns a null row in the collection, GenerateExec gets a NullPointerException either when joining the null row with required child output, or projecting the null row.
This PR avoids producing the null row and produces a row of null values instead:
```
spark-sql> set spark.sql.codegen.wholeStage=false;
spark.sql.codegen.wholeStage false
Time taken: 3.095 seconds, Fetched 1 row(s)
spark-sql> select inline(array(named_struct('a', 1, 'b', 2), null));
1 2
NULL NULL
Time taken: 1.214 seconds, Fetched 2 row(s)
spark-sql>
```
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
New unit test.
Closes #36903 from bersprockets/inline_eval_null_struct_issue.
Authored-by: Bruce Robbins <[email protected]>
Signed-off-by: Hyukjin Kwon <[email protected]>
(cherry picked from commit c4d5390)
Signed-off-by: Hyukjin Kwon <[email protected]>
HyukjinKwon
pushed a commit
that referenced
this pull request
Jun 18, 2022
Change `Inline.eval` to return a row of null values rather than a null row in the case of a null input struct.
Consider the following query:
```
set spark.sql.codegen.wholeStage=false;
select inline(array(named_struct('a', 1, 'b', 2), null));
```
This query fails with a `NullPointerException`:
```
22/06/16 15:10:06 ERROR Executor: Exception in task 0.0 in stage 0.0 (TID 0)
java.lang.NullPointerException
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.apply(Unknown Source)
at org.apache.spark.sql.execution.GenerateExec.$anonfun$doExecute$11(GenerateExec.scala:122)
```
(In Spark 3.1.3, you don't need to set `spark.sql.codegen.wholeStage` to false to reproduce the error, since Spark 3.1.3 has no codegen path for `Inline`).
This query fails regardless of the setting of `spark.sql.codegen.wholeStage`:
```
val dfWide = (Seq((1))
.toDF("col0")
.selectExpr(Seq.tabulate(99)(x => s"$x as col${x + 1}"): _*))
val df = (dfWide
.selectExpr("*", "array(named_struct('a', 1, 'b', 2), null) as struct_array"))
df.selectExpr("*", "inline(struct_array)").collect
```
It fails with
```
22/06/16 15:18:55 ERROR Executor: Exception in task 0.0 in stage 0.0 (TID 0)/ 1]
java.lang.NullPointerException
at org.apache.spark.sql.catalyst.expressions.JoinedRow.isNullAt(JoinedRow.scala:80)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.writeFields_0_8$(Unknown Source)
```
When `Inline.eval` returns a null row in the collection, GenerateExec gets a NullPointerException either when joining the null row with required child output, or projecting the null row.
This PR avoids producing the null row and produces a row of null values instead:
```
spark-sql> set spark.sql.codegen.wholeStage=false;
spark.sql.codegen.wholeStage false
Time taken: 3.095 seconds, Fetched 1 row(s)
spark-sql> select inline(array(named_struct('a', 1, 'b', 2), null));
1 2
NULL NULL
Time taken: 1.214 seconds, Fetched 2 row(s)
spark-sql>
```
No.
New unit test.
Closes #36903 from bersprockets/inline_eval_null_struct_issue.
Authored-by: Bruce Robbins <[email protected]>
Signed-off-by: Hyukjin Kwon <[email protected]>
(cherry picked from commit c4d5390)
Signed-off-by: Hyukjin Kwon <[email protected]>
Member
|
Merged to master, branch-3.3 and branch-3.2. @bersprockets do we need this for branch-3.1 too (assuming from the JIRA version, yes?) it has a conflict. feel free to create a backport PR. |
Contributor
Author
|
Thanks!
Will do. |
bersprockets
added a commit
to bersprockets/spark
that referenced
this pull request
Jun 18, 2022
Change `Inline.eval` to return a row of null values rather than a null row in the case of a null input struct.
Consider the following query:
```
set spark.sql.codegen.wholeStage=false;
select inline(array(named_struct('a', 1, 'b', 2), null));
```
This query fails with a `NullPointerException`:
```
22/06/16 15:10:06 ERROR Executor: Exception in task 0.0 in stage 0.0 (TID 0)
java.lang.NullPointerException
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.apply(Unknown Source)
at org.apache.spark.sql.execution.GenerateExec.$anonfun$doExecute$11(GenerateExec.scala:122)
```
(In Spark 3.1.3, you don't need to set `spark.sql.codegen.wholeStage` to false to reproduce the error, since Spark 3.1.3 has no codegen path for `Inline`).
This query fails regardless of the setting of `spark.sql.codegen.wholeStage`:
```
val dfWide = (Seq((1))
.toDF("col0")
.selectExpr(Seq.tabulate(99)(x => s"$x as col${x + 1}"): _*))
val df = (dfWide
.selectExpr("*", "array(named_struct('a', 1, 'b', 2), null) as struct_array"))
df.selectExpr("*", "inline(struct_array)").collect
```
It fails with
```
22/06/16 15:18:55 ERROR Executor: Exception in task 0.0 in stage 0.0 (TID 0)/ 1]
java.lang.NullPointerException
at org.apache.spark.sql.catalyst.expressions.JoinedRow.isNullAt(JoinedRow.scala:80)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.writeFields_0_8$(Unknown Source)
```
When `Inline.eval` returns a null row in the collection, GenerateExec gets a NullPointerException either when joining the null row with required child output, or projecting the null row.
This PR avoids producing the null row and produces a row of null values instead:
```
spark-sql> set spark.sql.codegen.wholeStage=false;
spark.sql.codegen.wholeStage false
Time taken: 3.095 seconds, Fetched 1 row(s)
spark-sql> select inline(array(named_struct('a', 1, 'b', 2), null));
1 2
NULL NULL
Time taken: 1.214 seconds, Fetched 2 row(s)
spark-sql>
```
No.
New unit test.
Closes apache#36903 from bersprockets/inline_eval_null_struct_issue.
Authored-by: Bruce Robbins <[email protected]>
Signed-off-by: Hyukjin Kwon <[email protected]>
HyukjinKwon
pushed a commit
that referenced
this pull request
Jun 21, 2022
### What changes were proposed in this pull request? Backport of #36903 Change `Inline.eval` to return a row of null values rather than a null row in the case of a null input struct. ### Why are the changes needed? Consider the following query: ``` set spark.sql.codegen.wholeStage=false; select inline(array(named_struct('a', 1, 'b', 2), null)); ``` This query fails with a `NullPointerException`: ``` 22/06/16 15:10:06 ERROR Executor: Exception in task 0.0 in stage 0.0 (TID 0) java.lang.NullPointerException at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.apply(Unknown Source) at org.apache.spark.sql.execution.GenerateExec.$anonfun$doExecute$11(GenerateExec.scala:122) ``` (In Spark 3.1.x, you don't need to set `spark.sql.codegen.wholeStage` to false to reproduce the error, since Spark 3.1.x has no codegen path for `Inline`). This query fails regardless of the setting of `spark.sql.codegen.wholeStage`: ``` val dfWide = (Seq((1)) .toDF("col0") .selectExpr(Seq.tabulate(99)(x => s"$x as col${x + 1}"): _*)) val df = (dfWide .selectExpr("*", "array(named_struct('a', 1, 'b', 2), null) as struct_array")) df.selectExpr("*", "inline(struct_array)").collect ``` It fails with ``` 22/06/16 15:18:55 ERROR Executor: Exception in task 0.0 in stage 0.0 (TID 0)/ 1] java.lang.NullPointerException at org.apache.spark.sql.catalyst.expressions.JoinedRow.isNullAt(JoinedRow.scala:80) at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.writeFields_0_8$(Unknown Source) ``` When `Inline.eval` returns a null row in the collection, GenerateExec gets a NullPointerException either when joining the null row with required child output, or projecting the null row. This PR avoids producing the null row and produces a row of null values instead: ``` spark-sql> set spark.sql.codegen.wholeStage=false; spark.sql.codegen.wholeStage false Time taken: 3.095 seconds, Fetched 1 row(s) spark-sql> select inline(array(named_struct('a', 1, 'b', 2), null)); 1 2 NULL NULL Time taken: 1.214 seconds, Fetched 2 row(s) spark-sql> ``` ### Does this PR introduce _any_ user-facing change? No. ### How was this patch tested? New unit test. Closes #36949 from bersprockets/inline_eval_null_struct_issue_31. Authored-by: Bruce Robbins <[email protected]> Signed-off-by: Hyukjin Kwon <[email protected]>
sunchao
pushed a commit
to sunchao/spark
that referenced
this pull request
Jun 2, 2023
Change `Inline.eval` to return a row of null values rather than a null row in the case of a null input struct.
Consider the following query:
```
set spark.sql.codegen.wholeStage=false;
select inline(array(named_struct('a', 1, 'b', 2), null));
```
This query fails with a `NullPointerException`:
```
22/06/16 15:10:06 ERROR Executor: Exception in task 0.0 in stage 0.0 (TID 0)
java.lang.NullPointerException
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.apply(Unknown Source)
at org.apache.spark.sql.execution.GenerateExec.$anonfun$doExecute$11(GenerateExec.scala:122)
```
(In Spark 3.1.3, you don't need to set `spark.sql.codegen.wholeStage` to false to reproduce the error, since Spark 3.1.3 has no codegen path for `Inline`).
This query fails regardless of the setting of `spark.sql.codegen.wholeStage`:
```
val dfWide = (Seq((1))
.toDF("col0")
.selectExpr(Seq.tabulate(99)(x => s"$x as col${x + 1}"): _*))
val df = (dfWide
.selectExpr("*", "array(named_struct('a', 1, 'b', 2), null) as struct_array"))
df.selectExpr("*", "inline(struct_array)").collect
```
It fails with
```
22/06/16 15:18:55 ERROR Executor: Exception in task 0.0 in stage 0.0 (TID 0)/ 1]
java.lang.NullPointerException
at org.apache.spark.sql.catalyst.expressions.JoinedRow.isNullAt(JoinedRow.scala:80)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.writeFields_0_8$(Unknown Source)
```
When `Inline.eval` returns a null row in the collection, GenerateExec gets a NullPointerException either when joining the null row with required child output, or projecting the null row.
This PR avoids producing the null row and produces a row of null values instead:
```
spark-sql> set spark.sql.codegen.wholeStage=false;
spark.sql.codegen.wholeStage false
Time taken: 3.095 seconds, Fetched 1 row(s)
spark-sql> select inline(array(named_struct('a', 1, 'b', 2), null));
1 2
NULL NULL
Time taken: 1.214 seconds, Fetched 2 row(s)
spark-sql>
```
No.
New unit test.
Closes apache#36903 from bersprockets/inline_eval_null_struct_issue.
Authored-by: Bruce Robbins <[email protected]>
Signed-off-by: Hyukjin Kwon <[email protected]>
(cherry picked from commit c4d5390)
Signed-off-by: Hyukjin Kwon <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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?
Change
Inline.evalto return a row of null values rather than a null row in the case of a null input struct.Why are the changes needed?
Consider the following query:
This query fails with a
NullPointerException:(In Spark 3.1.3, you don't need to set
spark.sql.codegen.wholeStageto false to reproduce the error, since Spark 3.1.3 has no codegen path forInline).This query fails regardless of the setting of
spark.sql.codegen.wholeStage:It fails with
When
Inline.evalreturns a null row in the collection, GenerateExec gets a NullPointerException either when joining the null row with required child output, or projecting the null row.This PR avoids producing the null row and produces a row of null values instead:
Does this PR introduce any user-facing change?
No.
How was this patch tested?
New unit test.