April 26, 2023
Release notes and fixed issues for version 2.0.38.
New Features / Improvements
- Custom Data Connections - Site administrators can now configure access to external data sources with the new Custom Data Connections feature. Data Scientists can access the external data via the cml.data library and its 2-liner abstractions.
- Add Data - Data Scientists can now upload files to Hive and Impala Virtual Warehouse tables from the Data tab of any CML Project.
- Model Replicas - Site Administrators can now configure the maximum number of model replicas that users can select for their models via the Maximum Model Replicas field on the Administration → Settings → Model Deployment Settings page.
- Models - Users can now deploy large ML models. Model size is not limited to 50 MB anymore.
- Model Registry - Administrators can now find the Machine User Workload User Name that is needed for configuring their model registries to access their S3 or ADLS Gen2 bucket on the Workspace Details page.
- Install Workspaces (Tech Preview) - Auto-retriable workflow for Install Workspace for Azure Private Cluster. See the Feature Preview doc for more information.
- Usage Monitoring - Usage data for Spark Executors is now recorded in the CSV file that Site Administrators can download from the tab.
- API Keys - Improved security by storing Legacy API keys as hashes in the
database. Existing Legacy API keys are automatically rotated as part of the upgrade
process to ensure that previous keys cannot be used. API v1 keys will not be usable
after the upgrade. This does not impact the Models API, and Models authentication is not
affected. To manually rotate a Legacy API key, do the following:
- In Rotate to generate a new Legacy ApiKey and ApiKeyHash pair. , click
- Copy the Api Key that is shown after rotation and use it in future requests.
- Note - The API key will not be visible on the UI once you refresh the page. Make sure to copy it before leaving the page.
- Modify Instance Group Type - Administrators can easily change the CPU or GPU instance types of node groups for a CML workspace, without having to re-provision the workspace.
Fixed Issues
- Model Registry (DSE-25909) - Removed the 19 character limit for model names in Model Registry.
- Model Registry (DSE-25641) - Fixed an issue where redeploying models from model registry to a CML workspace may fail.
- Model Registry (DSE-24683) - Added workload name to Model Registry service account list.
- Model Registry (DSE-25906) - Model Registry is no longer limited to listing 10 models at a time.