Cloudera Data Engineering on premises 1.5.5 SP3 CHF3

Review the features, fixes, and known issues in the Cloudera Data Engineering 1.5.5 Service Pack 3 Cumulative hotfix 3 release.

What's new

There are no new features or enhancements introduced in this release.

Fixed issues

DEX-22841: Livy pod thread leak causing job failures
Previously, in Cloudera Data Services on premises 1.5.5 SP3, Livy pods silently became unresponsive (HTTP 500 errors) while appearing healthy in Kubernetes, requiring a manual restart. This was caused by a token refresh routine that leaked Kubernetes client threads until the container hit its process limit. This issue is now resolved by updating the token refresh logic to reuse the existing Kubernetes client, which eliminates the thread leak.

Known issues

The following are the known issues identified in 1.5.5 SP3 CHF3:

DEX-23208: Spark jobs fail to interact with Ozone after Cloudera Base on Premises upgrade due to stale ozone-parcel-image
When upgrading a Cloudera Base on premises cluster (example, from 7.1.9 to 7.3.2 SP1) while running Cloudera Data Engineering, the upgrade orchestration fails to automatically update the Cloudera Data Engineering ozone-parcel-image. As a result, the environment continues to use a stale image that contains a classpath incompatibility (Protobuf 2.x vs 3.x). This causes Spark jobs or sessions attempting to read from or write to Ozone to fail immediately with a java.lang.NoClassDefFoundError: com/google/protobuf/LiteralByteString.
Make sure that the Cloudera Data Services on premises and Cloudera Data Engineering services are upgraded alongside the Cloudera Base on premises upgrade to trigger the correct image selection logic.