Release Notes 2.14 (September 2023)

App Panel for Pipelines

Users can now create no-code App Panels for Pipelines allowing them to easily swap Data Assets and specify parameters for every step of the workflow. The Pipelines App Panel Build Mode allows you to add a subtitle and description to your Pipeline, choose specific results to display upon completion of a run, and change the order of Capsules as they appear in the App Panel.

Dedicated Machine Filtering

Administrators can now choose which dedicated machine types will be available for Capsule runs in their Code Ocean deployment from the Dedicated Machines tab of the Admin Panel. One use for this is to deactivate the most expensive instance types to aid in cost control.

From the Select Compute Resource menu in a Capsule, users will be able to see which resources have been deactivated and can easily copy the instance type to share with their administrator if they'd like to ask to have that instance type reactivated.

G5 Instance Types Available as Dedicated Machines

G5 GPU instances are now available as Dedicated Machines. These instances are designed to accelerate graphics-intensive applications and machine learning inference. They can also be used to train simple to moderately complex machine learning models.

Updated Python + R Starter Environment

The default repository for R packages has been changed from the standard CRAN repository to the Posit repository which allows R packages to be installed from binary resulting in significantly faster installation times. Python packages have also been updated to their latest versions.

Additional Information in the Capsule Timeline

Machine Type and Resources

Machine type (Flex or Dedicated) and size (GPU, CPU, and RAM) are now visible under Compute Resources when clicking the dropdown menu of a run in the Capsule Timeline, increasing the traceability and reproducibility of results.

Average and Peak Memory Usage Shown as Percentage

Previously, only the absolute memory usage was was shown under the Performance Metrics for each run in the Timeline. Now, when users click Performance Metrics from a run's dropdown menu in the Timeline, they’ll be able to see memory usage as a percentage of the memory available in the provisioned resource. This information combined with the Machine Type and Resources information newly available in the Timeline allows users to quickly determine if they have under-provisioned or over-provisioned compute resources and adjust accordingly.

5GB Capsule Root Volume Size

Users will now have 5GB of available root space in their Cloud Workstation sessions, regardless of their Capsule environment's size. This provides more flexibility when installing packages in a Cloud Workstation session.

The screenshots below show the available root space before and after version 2.14 in a Capsule using the TensorFlow (Python 3) (2.11, Python 3.10, CUDA 11.7.0, Ubuntu 22.04) Starter Environment with PyTorch installed in the Capsule environment.

Secrets Available at Build Time

User Secrets (AWS Cloud Credentials, Database Credentials, API Credentials, and Custom Keys) are now available during environment builds with the values accessible in the Dockerfile and PostInstall under their typical environment variable names.

Note: To make secrets available at build time, make sure to commit changes in your capsule before attaching the secrets.

Support for Runs on Multiple Availability Zones

When configuring a Code Ocean deployment, the administrator is able to select two AWS Availability Zones. In the past, this has solely been for redundancy in the case of an AWS service interruption. To provide a greater chance of successfully provisioning selected compute resources, resources will automatically be provisioned from across the two Availability Zones selected while configuring your deployment. This can be especially useful when working with powerful GPU instances or other highly-requested resources.