spark3 | scala 2.12 | jackson ScalaObjectMapper | NoSuchMethodError

[kennydataml]

Describe the bug

Loading dat file produces a NoSuchMethodError

To Reproduce

sc.addFile(copybook)
sc.addFile(dat)
df = spark.read.option.format("cobol")\
  .option("copybook", SparkFiles.get(copybook))\
  .load(SparkFiles.get(dat))

Expected behaviour

Should load successfully

Screenshots

20/04/28 19:09:53 INFO DefaultSource: Cobrix 'spark-cobol' build 2.0.7 (2020-04-14T12:00:03)
Traceback (most recent call last):
  File "/tmp/spark-0a0a39a6-a906-4b6c-af92-2ece5fcb2776/k8s_spark_cobol_adls.py", line 77, in <module>
    .load(SparkFiles.get(datafile1))
  File "/opt/spark/python/lib/pyspark.zip/pyspark/sql/readwriter.py", line 160, in load
  File "/opt/spark/python/lib/py4j-0.10.8.1-src.zip/py4j/java_gateway.py", line 1286, in __call__
  File "/opt/spark/python/lib/pyspark.zip/pyspark/sql/utils.py", line 98, in deco
  File "/opt/spark/python/lib/py4j-0.10.8.1-src.zip/py4j/protocol.py", line 328, in get_return_value
py4j.protocol.Py4JJavaError: An error occurred while calling o58.load.
: java.lang.NoSuchMethodError: com.fasterxml.jackson.module.scala.experimental.ScalaObjectMapper.$init$(Lcom/fasterxml/jackson/module/scala/experimental/ScalaObjectMapper;)V
        at za.co.absa.cobrix.spark.cobol.source.parameters.CobolParametersParser$$anon$1.<init>(CobolParametersParser.scala:499)
        at za.co.absa.cobrix.spark.cobol.source.parameters.CobolParametersParser$.getOccursMappings(CobolParametersParser.scala:499)
        at za.co.absa.cobrix.spark.cobol.source.parameters.CobolParametersParser$.parse(CobolParametersParser.scala:215)
        at za.co.absa.cobrix.spark.cobol.source.DefaultSource.createRelation(DefaultSource.scala:56)
        at za.co.absa.cobrix.spark.cobol.source.DefaultSource.createRelation(DefaultSource.scala:48)
        at org.apache.spark.sql.execution.datasources.DataSource.resolveRelation(DataSource.scala:339)
        at org.apache.spark.sql.DataFrameReader.loadV1Source(DataFrameReader.scala:240)
        at org.apache.spark.sql.DataFrameReader.$anonfun$load$2(DataFrameReader.scala:229)
        at scala.Option.getOrElse(Option.scala:189)
        at org.apache.spark.sql.DataFrameReader.load(DataFrameReader.scala:229)
        at org.apache.spark.sql.DataFrameReader.load(DataFrameReader.scala:190)
        at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
        at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
        at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
        at java.lang.reflect.Method.invoke(Method.java:498)
        at py4j.reflection.MethodInvoker.invoke(MethodInvoker.java:244)
        at py4j.reflection.ReflectionEngine.invoke(ReflectionEngine.java:357)
        at py4j.Gateway.invoke(Gateway.java:282)
        at py4j.commands.AbstractCommand.invokeMethod(AbstractCommand.java:132)
        at py4j.commands.CallCommand.execute(CallCommand.java:79)
        at py4j.GatewayConnection.run(GatewayConnection.java:238)
        at java.lang.Thread.run(Thread.java:748)

20/04/28 19:09:53 INFO SparkContext: Invoking stop() from shutdown hook

Additional context

Using spark3 pyspark on Kubernetes. Require to use sc.addFile and SparkFiles.get due to limitations of kubernetes with ADLS2

[kennydataml]

error is due to conflicting jackson versions with Spark3. Had to pull this Git repo and build the jar with sbt clean package. changed the dependencies.scala to match with spark 3 before building.

[yruslan]

Hi,

Thanks for the bug report!

I'm going to reopen it since spark-cobol should support Spark 3 out of the box.

[tr11]

@kenny-bui-slalom, which version did you use to make it work?

[kennydataml]

spark 3.0.0-preview2 currently uses jackson version 2.10.0

[tr11]

Yeah, tried with 3.0.0-preview2 and had similar issues.
Just changing the scala version to 2.12 fixed it for me even with the jackson version 2.10.3.

I'll add a PR to specifically list the dependency as an override.

[yruslan]

@tr11 , Thanks for the fix!

[yruslan]

@tr11 , after the explicit jackson dependency is added, spark-cobol is failing to build.

  java.lang.NoSuchMethodError: com.fasterxml.jackson.databind.JsonMappingException.<init>(Ljava/io/Closeable;Ljava/lang/String;)V
  at com.fasterxml.jackson.module.scala.JacksonModule$class.setupModule(JacksonModule.scala:61)
  at com.fasterxml.jackson.module.scala.DefaultScalaModule.setupModule(DefaultScalaModule.scala:17)
  at com.fasterxml.jackson.databind.ObjectMapper.registerModule(ObjectMapper.java:718)
  at za.co.absa.cobrix.spark.cobol.parameters.CobolParametersParser$.getOccursMappings(CobolParametersParser.scala:498)
  at za.co.absa.cobrix.spark.cobol.parameters.CobolParametersParser$.parse(CobolParametersParser.scala:213)

[yruslan]

I've played with shading our own version of Jackson in cobol-parser. It works well, but cobol-parser becomes a 3M artifact (from original 700K). And shading disrupts Maven's dependency management somewhat. So I think I'll just going to write a tiny JSON parser for the OCCURS parameters. It will be tied to the specific structure of the parameters JSON.

[tr11]

Makes sense. Since we already import ANTLR, we can use the grammar at https://github.com/antlr/grammars-v4/blob/master/json/JSON.g4. I can very easily build the visitor for that.

[yruslan]

Cool, thanks for the link!