Fixed Issues
This section lists issues fixed in this release of Cloudera Machine Learning (CML) on Private Cloud.
- DSE-24333: Embedded Spark UI fails to load in Spark pushdown mode for Spark 3
-
When Spark is run on yarn, the configuration attempts to disable proxying of the Spark UI through the Yarn Resource Manager. This setting was not respected by Spark 3 binaries and resulted in the Could not determine the proxy server for redirection error message when attempting to reach the embedded Spark UI on the driver.
This behavior works again in Spark 3.2.3 with CDH 7.1.7 and Spark 3.3.2 with CDH 7.1.9.
- DSE-32887: Runtime updates cannot be enabled from the Cloudera Manager UI Site Administration dashboard
- Runtime updates can now be enabled from the CML UI Site Administration dashboard.
- DSE-29976: Job Arguments fail in PBJ runtimes
- PBJ (Powered by Jupyter) Runtimes were not able to access Job arguments as standard command line arguments. To mitigate this, a new environment variable was added to Jobs, called "JOB_ARGUMENTS". This environment variable can be used to access arguments specified in the Job Creation or Job Update UI.
- DSE-27285: The calculation of user resources was incorrect
- The User Resources tab in the workspace UI calculated with the used and available resources for a cluster that also included the resources for the master node. As the workload cannot be scheduled in the master nodes, the master node resources shall not be considered in these calculations. This issue has been solved, the resources for the master are no longer included in the calculations for the User Resources.