ML Private Cloud 1.5.4 CHF1

Review the fixed issues in the Cloudera Machine Learning (CML) 1.5.4 Cumulative hotfix 1 release.

DSE-37209: Upgrade of CML workspace with external NFS fails

For setups with an external network file system (NFS), the Cloudera Machine Learning (CML) workspace upgrade flow was different from setups with internal NFSs. There was an issue with the persistent volume claim (PVC) size in external NFS flows which caused failures during the upgrade path to 1.5.4. Now, the issue has been fixed. For more details, see Cloudera Customer Advisory 2024-778: CML Private Cloud Data Service upgrade to 1.5.4 is failing with external NFS Technical Service Bulletin.

DSE-36078: Spark workload histories are not logged in Spark history server

The Spark history server did not log Spark workload history so far. This issue has been fixed and the Spark workload history is now logged in the base cluster Spark history server.

DSE-37192: Time zone discrepancy for usage API

Some data with respect to the workload run were stored in the local time zone, as a result of which the API returned empty results. By setting all the columns of the dashboard database to UTC time zone, there is no longer time zone discrepancy for the usage API. Note that the data stored prior to the upgrade remains in the previously set time zone.

DSE-36489: Opening existing JupyterLab session redirected to project page

After having logged out from the CML UI and having logged back, the opening of an existing JupyterLab session led to the project page instead of opening the JupyterLab UI. This issue has been fixed and after a new login the JupyterLab UI can be opened.

DSE-32204: Improved error messaging during failed Kerberos authentication in session startup

If there was an issue with Kerberos credential authentication, CML session startup hung until the regular session timed out and no warning message was displayed for the user. Now, a warning message with additional details on the status of the Kerberos credential authentication is displayed for the user along with a startup failure message after 1 minute.

DSE-36666: The migration readiness check is disrupted when engine readiness check is run

The shell command executor on Cloudera Data Science Workbench (CDSW) could not handle the CDSW to CML migration readiness timeout. This issue has been fixed, the timeout has been handled. Additionally, any SQL query-based readiness checks that remain unhandled will be displayed for manual validation.

DSE-36759: AMPs and Feature Announcement sections do not work in NTP setups

Cloudera Machine Learning (CML) Private Cloud setups with non-transparent proxy (NTP) did not function properly, affecting Accelerators for ML Projects (AMPs) and feature announcements. The home page froze, the feature announcement displayed an error message, and the AMPs did not load. This issue has been fixed with the enablement of non-transparent proxy in a Private Cloud setup.

DSE-37454: Possible CML terminal disconnections

The Cloudera Machine Learning (CML) terminal tended to disconnect on Embedded Container Service (ECS) because the proxy_read_timeout value on workload pods was set to a default value of 60 seconds. The proxy_read_timeout value has been increased to 121 seconds and the issue of the CML terminal disconnecting connections has been resolved.

DSE-37610: Due to version mismatch of the jackson-module-scala_3 the mlops governance pod image failed

Earlier, the version of the jackson-module-scala_3 was updated in the mlopsgovernance/pom.xml file, but it was not updated in the parent/pom.xml file. This issue has been solved and there is no longer version mismatch for the jackson-module-scala_3.

DSE-36748: File upload issue fixed using APIv2

Earlier, when uploading a file to a sub-folder in a project, the file was added to the root folder. This issue has been fixed and now when uploading a file to a subfolder it is added to the project's sub folder.

Review the known issues in the Cloudera Machine Learning (CML) 1.5.4 Cumulative hotfix 1 release.

DSE-36967: Namespace Termination issue when using Portworx storage

There is an issue with Portworx version lower than 3.1.1, as the namespace deletion gets stuck in terminating state. Portworx is not able to cleanly unmount and clean up the underlying resources.

Workaround: The issues is fixed with Portworx version 3.1.1. Upgrade to Portworx version 3.1.1 or to later versions.

DSE-37002: Download API using v2 python swagger client does not work

The download_project_file method does not function as expected. The v2 python swagger only allows content to be downloaded as a string and not in bytes, as described in Binary string response encoding with Python 3.

Workaround: Use download_project_file_v2 as an alternate method.