I am creating a spark submit through oozie and how do I ensure that my spark jar has priority over all others jars that will be run through oozie lib path ?
[Loaded org.apache.poi.xssf.model.ThemesTable$ThemeElement from file:/usr/local/lib/radoop/rapidminer_libs-7.6.1.jar]
I tried setting oozie.use.system.libpath = false , did not work then I tried modifying HADOOP_CLASSPATH variable did not work
I also tried modifying spark submit adding #--conf "spark.driver.userClassPathFirst=true" --conf "spark.executor.userClassPathFirst=true" - Also did not work
Then I tried all kinds of tricks with maven that did not work trying to relocate the jar.
Similar question here https://community.hortonworks.com/questions/6610/can-i-ensure-that-my-own-jars-have-classpath-prior.... is there a way to do it without java action or only with java action?
I am using spark 2.3,scala 2.11
So your findings lead you to think there is a jar in oozie share lib directory that is taking precedence over yours?
Did you find which exact jar in oozie share lib directory is causing the problem?
I think the options you have tried are all good attempts to avoid jar conflict if that is coming from oozie. But perhaps the problem could come from MR configs as an example. Hence I suggest you try to find which jar could be causing the problem for you in this case.
It is a rapid miner jar it contains poi version 3.14 but I am using java poi library version 3.17 in my spark jar.