[SPARK-18118] SpecificSafeProjection.apply of Java Object from Dataset to JavaRDD Grows Beyond 64 KB

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.1.0
• Fix Version/s: 2.0.3, 2.1.0
• Component/s: SQL
• Target Version/s: 2.1.0
• Assignee: Kazuaki Ishizaki
• Reporter: Aleksander Eskilson
• Created: 26/Oct/16 18:09
• Updated: 28/Nov/16 12:20
• Resolved: 28/Nov/16 12:20

Description

For sufficiently wide or nested Java Objects, when SpecificSafeProjection attempts to recreate the object from an InternalRow, the generated SpecificSafeProjection.apply method is larger than allowed:

Caused by: org.codehaus.janino.JaninoRuntimeException: Code of method "apply(Ljava/lang/Object;)Ljava/lang/Object;" of class "org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificSafeProjection" grows beyond 64 KB

Although related, this issue appears not to have been resolved by SPARK-15285. Since there is only one top-level object when projecting, splitExpressions finds no additional Expressions to split. The result is a single large, nested Expression that forms the apply code.

See the reproducer for an example [1].

[1] - https://github.com/bdrillard/specific-safe-projection-error

Issue Links

• is related to: SPARK-15285 Generated SpecificSafeProjection.apply method grows beyond 64 KB
• links to: [Github] Pull Request #16032 (kiszk)