Release Notes

This document provides a comprehensive overview of the latest updates and enhancements in HPE Ezmeral Unified Analytics Software (version 1.2.0), including new features, improvements, bug fixes, and known issues.

HPE Ezmeral Unified Analytics Software provides software foundations for enterprises to develop and deploy end-to-end data and advanced analytics solutions from data engineering to data science and machine learning across hybrid cloud infrastructures – delivered as a consumption-based software-as-a-service model.

New Features

GPU Monitoring and Resource Management
The platform has been enhanced with advanced GPU monitoring capabilities that provide real-time insights and control over GPU utilization. The system intelligently reclaims expensive GPU resources from idle workloads and proactively schedules and assigns those to higher priority jobs. Administrators can configure the reclaim policy and idle timing, aligning workload priorities with business needs. Administrators gain control over resource prioritization, while users experience prompt and efficient processing of their GPU requests.
For additional information, see GPU Support.

Tools and Frameworks User Isolation
HPE Ezmeral Unified Analytics Software extends user isolation beyond Kubeflow notebooks, pipelines, and models to include Airflow DAGs, MLflow metadata, Spark apps, and Superset dashboards to enhance security and control over workloads while allowing administrators to maintain comprehensive control and visibility into platform resources. This marks a significant stride toward establishing a more secure, robust, and workspace-centric platform, enriched with advanced role-based access control mechanisms.
For additional information, see User Isolation.

Monitoring Production Machine Learning Model
This release introduces a preview of Model Monitoring in HPE Ezmeral Unified Analytics Software, providing real-time model quality assurance, early detection of model and concept drift, comprehensive metric tracking, and seamless integration. With this HPE Ezmeral Unified Analytics Software capability, you can maintain and manage ML models at scale – delivering high-quality results in your production environments.
For additional information, see Model Monitoring.

Remote Object Store Data Sources
HPE Ezmeral Unified Analytics Software integrates with S3-compliant external object stores such as HPE Ezmeral Data Fabric Object Store, AWS S3, and MinIO. After a simple administrative setup to connect HPE Ezmeral Unified Analytics Software to remote object storage, users can establish connections with their identity tokens; secret access key management complexities are eliminated.
For additional information, see Connecting to External S3 Object Stores and Accessing Data in External S3 Object Stores.

Enhancements

External Data Fabric Volumes
The external data fabric volume connection mechanism has shifted from an NFS to a CSI-based approach for future readiness to provide additional capabilities, including dynamic provisioning, scalability, and improved performance and security. This change also broadens our compatibility with various block storage providers that adhere to the CSI Kubernetes open standard.
For additional information, see Connecting to External HPE Ezmeral Data Fabric Clusters.

User Workloads Identity Propagation
An enhanced platform security policy requires all user workloads to run under a user account that aligns with the user's profile and identity permissions assigned to the profile. This enhancement secures user workloads and ensures that workloads run according to the user's privileges.

Framework Upgrades
With this HPE Ezmeral Unified Analytics Software release, you can orchestrate and automate upgrades for the tools and frameworks through the HPE-managed SaaS control plane. This feature simplifies the enhancement process, improves operational efficiency, and reduces downtime. You can conveniently configure and manage scheduled upgrades via the HPE Ezmeral Unified Analytics Software UI, tailoring them to your specific setup and requirements.
For additional information, see Configuring Included Applications and Upgrading Included Frameworks.

Notebook Experience
HPE Ezmeral Unified Analytics Software offers new notebook tools to seamlessly query connected databases and object stores. These include new magic commands and pre-configured Python clients that boost productivity, integrate with your notebook, and simplify data operations, enhancing your overall experience.
For additional information, see Notebook Magic Functions.

Other Application Upgrades
For a list of updated applications, including Airflow, Feast, Livy, MLflow, Ray, Spark, and Superset, see Support Matrix.

Resolved Issues

This release introduces the following fixes:
HPE Ezmeral Data Fabric Connectivity
Access delays when connecting to HPE Ezmeral Data Fabric have been resolved.

Spark Application Navigation
Browsing nested folders while creating Spark Applications now functions as expected.

Custom Python-pip Package Installation via Ray init API
You can now install custom pip packages using Ray init API.

EzPresto Memory Allocation
Memory allocation for queries now considers the entire cluster's resources, preventing query failures due to insufficient memory.

Known Issues

The following sections describe known issues with workarounds where applicable:
Warning Banner Persists After Updating Activation Key
On the Activation Key tab, the activation key expiration warning banner remains on the screen after you click the Here link in the banner to upload a new activation key. The license updates as expected, but you must refresh the page to make the warning banner disappear.

Cluster Inaccessible After GPU Node Reboot
If you reboot the GPU nodes in an HPE Ezmeral Unified Analytics Software cluster, various platform service pods that run on those nodes may not resume after the reboot, making HPE Ezmeral Unified Analytics Software inaccessible.
Workaround
To resolve this issue, complete the following steps on each GPU host:
  1. Edit the /usr/lib/systemd/system/kubelet.service file and remove the following line: 
    "Requires=containerd.service"
  2. Save the kubelet.service file.
  3. Run the following command:
    systemctl daemon-reload

GPU Hosts Must Run RHEL 8.8
HPE Ezmeral Unified Analytics Software only supports RHEL 8.8 on GPU hosts. Any other version of RHEL causes installation and cluster expansion on GPU hosts to fail.
Installation Fails due to Insufficient GPU/CPU
The installation process fails if the installation requests more GPU/CPU resources than are available. To prevent installation failures, ensure that the resource requests match the available GPU/CPU capacity.

Installer UI (Cannot access Infrastructure Services log files)
The system returns the following error message when you try to download the Infrastructure Services log files after the Infrastructure Services phase of the installation fails:
{"error":"Internal error"}
Workaround
To get the Infrastructure Services log files, change type=workloaddeploy in the browser URL to type=addonproc, as shown in the following example:
http://10.10.100.200:8080/api/v1/install/logs?type=addonproc

Hosts Not Ready After Reboot
If the HPE Ezmeral Unified Analytics Software hosts are not ready after a reboot, restore the system to a ready state. For details, see Host (Node) Management.

Spark Magic
To leverage Spark magic (%manage_spark) on Jupyter Notebooks for interactive Livy on GPU, users must manually configure GPU settings (%config_spark). For details, see Enabling GPU Support for Livy Sessions.

NVIDIA GPU Cannot Enforce SELinux
Due to a known NVIDIA GPU issue (https://github.com/NVIDIA/gpu-operator/issues/553), SELinux cannot be enforced for GPU deployments.
Workaround
Set GPU hosts to either disabled or permissive mode until this issue is resolved.

Ray Dashboard UI
A known Ray issue (https://github.com/ray-project/ray/issues/14664) prevents the Ray Dashboard UI from displaying the GPU worker group details correctly.

EzPresto does not release memory when a query completes
EzPresto does not release the memory it uses to run queries, but will re-use this memory in subsequent queries. For example, if a query consumes 100GB of memory, EzPresto does not release the 100GB of memory when the query completes. EzPresto uses this memory to run the next query. If the next query requires more memory, for example 120GB, EzPresto takes an additional 20GB of memory and does not release the additional 20GB of memory after the query completes. The amount of unreleased memory becomes 120GB.

Failed Workload Cluster Deployment Due to Cluster Name Length
Workload cluster deployment fails when the cluster name has more than 21 characters.
Workaround
Rename the workload cluster to a name that is 21 characters or less.

Installation

Before you install or upgrade, HPE recommends that you back up your data.  If you encounter any issues during or after the installation process, please contact HPE support. We appreciate your feedback and strive to continually enhance your product experience. 

Additional Resources

Thank you for choosing HPE Ezmeral Unified Analytics Software. Enjoy the new features and improvements introduced in this release.