not able to merge spark scala code in spark pipeline

[AshishM]

we are merging our spark scala code with spark pipeline, however if i run the code step individually it runs fine (in both function mode / free mode) but when it is been merged in spark pipeline it give error

 

java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624) at java.lang.Thread.run(Thread.java:748) Driver stacktrace:, caused by: ClassNotFoundException: CustomScalaRecipe$$anonfun$1

scala code:

 

import org.apache.spark.sql.functions._

// Recipe inputs
val iups_prepared_joined_by_time_cell_joined = inputDatasets("iups_prepared_joined_by_time_cell_joined")


def redaction(data:String, d_cnt: Int): String = {
     
     var data_trim = "" ;
     if (data.isEmpty || data =="" || data == null){
         return data ;
     }else {   
         data_trim = data.trim()  ;
     }
     
     if (d_cnt == 1){
         var data_length = data_trim.length();
         return data_trim.replaceAll(data_trim, "*"*data_length)
     }
     else if (d_cnt> 5){
         return data_trim
     }
     else {
         val redac_length = 6-d_cnt
         val length_r = Seq(redac_length, data_trim.length()).min
         return data_trim.slice(0,data_trim.length()-length_r) + "*"*(length_r+3)
    }
     
 }

val redaction_udf =  udf((data: String,d_cnt:Int) => redaction(data,d_cnt));

val iups_pre_anonymized = iups_prepared_joined_by_time_cell_joined.withColumn("price_plan", when(col("price_plan").isNull, lit(null)).otherwise(redaction_udf(col("price_plan"),col("price_plan_distinct"))))
// Recipe outputs

Map("iups_pre_anonymized" -> iups_pre_anonymized)

[AdrienL · Accepted solution]

Hello,

Thanks for reporting this, it seems to be a bug in the way spark pipelines and spark serialization work, I'll check to see if we can fix that in a future DSS version.

In the meantime, you can workaround it by either:

• (Simplest) disabling pipelining for this recipe, in the recipe's Advanced tab


• Converting the UDF to sparkSQL (a bit difficult for your use case)


• Compiling your UDF outside of DSS and place the resulting jar in <DATA_DIR>/lib/java, and call the UDF from your code

Best

[AshishM]

Thanks for the answer Andrien, Hopefully it will be fixed in future releases.

Meanwhile i have one more question, why my spark pipeline job runs only 8 tasks in parallel even though i can run 1200 tasks in parallel. Any clue about this and how can improve it

[AdrienL]

Are talking about DSS activities or Spark tasks in a single spark job? Where did you get those 8 / 1200 numbers?

[AshishM]

hi Adrien,

i have created another question for this since its completely different issue, please use below link for the question

https://answers.dataiku.com/5883/pipeline-running-spark-tasks-whereas-spark-recipe-culster

[bkostya]

Hello Dataiker! 
Are any solutions available since 2019?  I'm using dataiku 7.0.2 and have that problem of ClassNotFoundException: CustomScalaRecipe$$anonfun$1

[CoreyS]

Hi @bkostya I can confirm that this original issue was was fixed in version 5.1.5. Do you the same distinction between "it runs fine individually" but "it fails in a pipeline"?

[bkostya]

It fails in a recipe (pipeline). The CustomScalaRecipe caught my eye. Please look at that ticket.