[SPARK-17279][SQL] better error message for exceptions during ScalaUDF execution #14850

cloud-fan · 2016-08-27T08:12:44Z

What changes were proposed in this pull request?

If ScalaUDF throws exceptions during executing user code, sometimes it's hard for users to figure out what's wrong, especially when they use Spark shell. An example

org.apache.spark.SparkException: Job aborted due to stage failure: Task 12 in stage 325.0 failed 4 times, most recent failure: Lost task 12.3 in stage 325.0 (TID 35622, 10.0.207.202): java.lang.NullPointerException
    at line8414e872fb8b42aba390efc153d1611a12.$read$$iwC$$iwC$$iwC$$iwC$$anonfun$2.apply(<console>:40)
    at line8414e872fb8b42aba390efc153d1611a12.$read$$iwC$$iwC$$iwC$$iwC$$anonfun$2.apply(<console>:40)
    at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIterator.processNext(Unknown Source)
...

We should catch these exceptions and rethrow them with better error message, to say that the exception is happened in scala udf.

This PR also does some clean up for ScalaUDF and add a unit test suite for it.

How was this patch tested?

the new test suite

cloud-fan · 2016-08-27T08:13:41Z

cc @yhuai

SparkQA · 2016-08-27T10:17:21Z

Test build #64528 has finished for PR 14850 at commit 3c7bdff.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

yhuai · 2016-08-27T16:13:11Z

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/ScalaUDF.scala

+      f(input)
+    } catch {
+      case e: NullPointerException =>
+        throw new RuntimeException(npeErrorMessage, e)


Can we still use NullPointerException? NullPointerException can have a specified message. Then, you can use setStackTrace to set the original stacktrace.

SparkQA · 2016-08-29T17:37:32Z

Test build #64562 has finished for PR 14850 at commit 1bd8382.

This patch fails Spark unit tests.
This patch merges cleanly.
This patch adds no public classes.

SparkQA · 2016-08-30T13:40:47Z

Test build #64652 has finished for PR 14850 at commit 9a82d67.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

clockfly · 2016-09-01T13:53:15Z

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/ScalaUDF.scala

+    val result = try {
+      f(input)
+    } catch {
+      case e: NullPointerException =>


It is a bit hacky to set stack trace like this.
npe.setStackTrace(e.getStackTrace)

If user search the code line reported in the stack trace, user may not able to find the code that matches the error message.

For this code branch eval(input: InternalRow), the existing NPE message should be clear enough if there is a full stacktrace, and the stack contains method of the UDF.

The error message you provided can be totally wrong.
"Given UDF throws NPE during execution, please check the UDF to make sure it handles null parameters correctly".

What if NPE is not caused by null parameter? prompting this message is misleading.

clockfly · 2016-09-01T14:33:59Z

@cloud-fan

There are two branches to execute an UDF.

Call override def eval(input: InternalRow): Any directly. For this code branch, I don't think there is any barrier for an user to know it is an UDF problem, as the ScalaUDF is in the stack trace.
Another branch is in codegen. This code branch is confusing, in my opinion. Since every different code piece from different expression is fused into one code block. When an exception is thrown, we don't know who owns that code block.

clockfly · 2016-09-02T04:53:42Z

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/ScalaUDF.scala

-        s".apply($funcTerm.apply(${funcArguments.mkString(", ")}));"
+    val getFuncResult = s"$funcTerm.apply(${funcArguments.mkString(", ")})"
+    val rethrowException = "throw new org.apache.spark.SparkException" +
+      """("Exception happens when execute user code in Scala UDF.", e);"""


Exception happens when executing user defined function (className: input argument type => output argument type) .
Or
`Failed to execute user defined function (className: input argument type => output argument type)

SparkQA · 2016-09-02T04:59:20Z

Test build #64829 has finished for PR 14850 at commit b7c459b.

This patch fails Spark unit tests.
This patch merges cleanly.
This patch adds no public classes.

SparkQA · 2016-09-02T06:36:33Z

Test build #64833 has finished for PR 14850 at commit 4efb6fc.

This patch fails Spark unit tests.
This patch merges cleanly.
This patch adds no public classes.

SparkQA · 2016-09-02T07:33:07Z

Test build #64840 has finished for PR 14850 at commit c9cd5e0.

This patch fails Spark unit tests.
This patch merges cleanly.
This patch adds no public classes.

SparkQA · 2016-09-02T08:24:45Z

Test build #64845 has finished for PR 14850 at commit c6284bd.

This patch fails Spark unit tests.
This patch merges cleanly.
This patch adds no public classes.

SparkQA · 2016-09-03T06:06:42Z

Test build #64889 has finished for PR 14850 at commit bf786e6.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

clockfly · 2016-09-05T21:49:01Z

+1

cloud-fan · 2016-09-06T02:36:24Z

thanks for the review, merging to master!

cloud-fan · 2016-09-07T02:12:05Z

also backport it to 2.0

…F execution ## What changes were proposed in this pull request? If `ScalaUDF` throws exceptions during executing user code, sometimes it's hard for users to figure out what's wrong, especially when they use Spark shell. An example ``` org.apache.spark.SparkException: Job aborted due to stage failure: Task 12 in stage 325.0 failed 4 times, most recent failure: Lost task 12.3 in stage 325.0 (TID 35622, 10.0.207.202): java.lang.NullPointerException at line8414e872fb8b42aba390efc153d1611a12.$read$$iwC$$iwC$$iwC$$iwC$$anonfun$2.apply(<console>:40) at line8414e872fb8b42aba390efc153d1611a12.$read$$iwC$$iwC$$iwC$$iwC$$anonfun$2.apply(<console>:40) at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIterator.processNext(Unknown Source) ... ``` We should catch these exceptions and rethrow them with better error message, to say that the exception is happened in scala udf. This PR also does some clean up for `ScalaUDF` and add a unit test suite for it. ## How was this patch tested? the new test suite Author: Wenchen Fan <[email protected]> Closes #14850 from cloud-fan/npe. (cherry picked from commit 8d08f43) Signed-off-by: Wenchen Fan <[email protected]>

better error message for NPE during ScalaUDF execution

3c7bdff

yhuai reviewed Aug 27, 2016
View reviewed changes

address comments

9a82d67

cloud-fan force-pushed the npe branch from 1bd8382 to 9a82d67 Compare August 30, 2016 11:29

clockfly reviewed Sep 1, 2016
View reviewed changes

address comments

b7c459b

cloud-fan changed the title ~~[SPARK-17279][SQL] better error message for NPE during ScalaUDF execution~~ [SPARK-17279][SQL] better error message for exceptions during ScalaUDF execution Sep 2, 2016

clockfly reviewed Sep 2, 2016
View reviewed changes

address comments

420bb32

cloud-fan force-pushed the npe branch from 4efb6fc to c9cd5e0 Compare September 2, 2016 06:49

cloud-fan force-pushed the npe branch from c9cd5e0 to c6284bd Compare September 2, 2016 07:36

fix ml

bf786e6

cloud-fan force-pushed the npe branch from c6284bd to bf786e6 Compare September 3, 2016 03:57

asfgit closed this in 8d08f43 Sep 6, 2016

cloud-fan deleted the npe branch December 14, 2016 12:33

wangmiao1981 mentioned this pull request Jul 8, 2017

[SparkR][SPARK-20307]:SparkR: pass on setHandleInvalid to spark.mllib functions that use StringIndexer #18496

Closed

[SPARK-17279][SQL] better error message for exceptions during ScalaUDF execution #14850

[SPARK-17279][SQL] better error message for exceptions during ScalaUDF execution #14850

Uh oh!

Conversation

cloud-fan commented Aug 27, 2016 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What changes were proposed in this pull request?

How was this patch tested?

Uh oh!

cloud-fan commented Aug 27, 2016

Uh oh!

SparkQA commented Aug 27, 2016

Uh oh!

yhuai Aug 27, 2016

Choose a reason for hiding this comment

Uh oh!

SparkQA commented Aug 29, 2016

Uh oh!

SparkQA commented Aug 30, 2016

Uh oh!

clockfly Sep 1, 2016

Choose a reason for hiding this comment

Uh oh!

clockfly Sep 1, 2016 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

clockfly commented Sep 1, 2016

Uh oh!

clockfly Sep 2, 2016

Choose a reason for hiding this comment

Uh oh!

SparkQA commented Sep 2, 2016

Uh oh!

SparkQA commented Sep 2, 2016

Uh oh!

SparkQA commented Sep 2, 2016

Uh oh!

SparkQA commented Sep 2, 2016

Uh oh!

SparkQA commented Sep 3, 2016

Uh oh!

clockfly commented Sep 5, 2016

Uh oh!

cloud-fan commented Sep 6, 2016

Uh oh!

cloud-fan commented Sep 7, 2016

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

cloud-fan commented Aug 27, 2016 •

edited

Loading

clockfly Sep 1, 2016 •

edited

Loading