Spark jobs finishes but application takes time to close

Question

Running spark job using scala, as expected all jobs are finishing up on time , but somehow some INFO logs are printed for 20-25 minutes before job stops.

Posting few

Answer 1

As I put in a comment, I recommend using the spark-csv package instead of sc.saveAsTextFile and there are no problems with writing directly to s3 using that package :)

I don't know if you use s3 or s3n, but maybe try to switch. I have experienced problems with using s3a on Spark 1.5.2 (EMR-4.2) where writes timed out all the time and switching back to s3 solved the problem, so it's worth a try.

A couple of other things that should speed up writes to s3 is to use the DirectOutputCommiter

conf.set("spark.hadoop.mapred.output.committer.class","com.appsflyer.spark.DirectOutputCommitter")

and disabling generation of _SUCCESS files:

sc.hadoopConfiguration.set("mapreduce.fileoutputcommitter.marksuccessfuljobs", "false")

Note that disabling _SUCCESS files has to be set on the hadoop configuration of the SparkContext and not on the SparkConf.

I hope this helps.