Fixed Issues in Apache Oozie

Review the list of Oozie issues that are resolved in Cloudera Runtime 7.2.16.

CDPD-40326: mapreduce.job.acl-view-job property in Oozie workflow.xml not taking full effect.

This issue is resolved.

CDPD-29098: Oozie - Replace log4j 1.x with reload4j.

This issue is resolved.

CDPD-43343: Oozie log streaming bug when log timestamps are the same on multiple Oozie servers.

This issue is resolved.

CDPD-43192: Added additional HBase Jars to sharelib to support proper HBase interaction.

CDPD-41425: The following issue was fixed: OOZIE-3254 LAST_ONLY and NONE execution modes: possible OutOfMemoryError when there are too many coordinator actions to materialize.

CDPD-41328: Oozie incorrectly deleted the action dir base instead of the action dir belonging to the Job and thus causing intermittent failures. This has been fixed.

CDPD-39134: OOZIE-3661: Oozie cannot handle environment variables with key=value content.

CDPD-34593: Oozie's Spark action was unable to send lineage information to Atlas. We've added a new authentication / credential type called KafkaCredentials which will obtain a delegation token from Kafka. For more information please see the "Action Authentication" documentation in the official Oozie documentation which is accessible from the Oozie UI.

CDPD-30246: There were Jars missing from Oozie's Sqoop sharelib when a user wanted to import from RDBMS to HDFS/Hive into ORC file format or into HBase. The missing Jars are added now.

CDPD-27164: Oozie will no longer use the LoadBalancer to issue a callback notification, but instead it will try all available Oozie instances one-by-one. If the callback succeeded against one of the Oozie instances, then we will not try the other ones. This way the LoadBalancer will not be used for such purposes.

OPSAPS-63816: Cloudera Manager will provide the address of all Oozie server instances as a configuration to all Oozie instances. This will be then used by Oozie's callback mechanism so that instead of making the callback through the LoadBalancer in HA mode, the callback will be attempted through each Oozie instance, and if one of them succeeds, then we stop. This way we'll no longer use the LoadBalancer, and make the callback mechanism safer by not having a middle-man.