Reliable Support for Custom Hardware Specifications for Python Functions

Background and Problem Statement: In the current production environment of watsonx.ai (WML on CPD/Software Hub, tested with versions 5.2.2 and 5.3.0), the use of custom hardware specifications for Python Function deployments is not reliably possible. Especially when using custom images in combination with a custom hardware specification, even a simple update of the deployment (without any actual changes) leads to an inconsistent and unusable state:

• The deployment disappears from the UI and cannot be found, but still exists as an asset in the system.

• This issue does not occur when only IBM standard hardware specs are used.

• The problem affects both online and batch deployments.

Reproducibility: This behavior is clearly reproducible (see Case TS021972721):

1. Create a custom hardware specification via API.

2. Create a custom image.

3. Create a runtime definition for the image.

4. Deploy using the custom image and custom hardware specification.

5. Update the deployment (no changes required).

Result: After the update, the deployment is no longer usable or visible, although it still exists in the asset repository. This issue does not occur when using IBM standard hardware specifications.

Operational Impact:

• Resource Waste: Without custom hardware specs, over-provisioned standard values must be used, leading to massive over-allocation of CPU and memory.

• Lack of Flexibility: Workarounds (e.g., switching to custom specs after initial deployment) are error-prone and unreliable.

• Operational Risk: Deployments can become unusable through simple updates, leading to outages and increased operational effort.

• Scalability Issues: The inability to assign hardware resources granularly limits efficient infrastructure usage and complicates cost control.

Urgency and Benefits of the Enhancement: Stable and documented support for custom hardware specifications for Python Functions is urgently required to:

• Assign resources efficiently and as needed,

• Ensure operational security and maintainability of deployments,

• Improve scalability and cost control,

• And to actually use the functionality promised in the API documentation.

Summary: Without the ability to reliably use custom hardware specifications for Python Functions, significant operational risks and unnecessary costs arise. The current implementation is error-prone and does not match the documented capabilities. Timely and stable support is therefore of high priority for all customers who rely on flexible and efficient resource usage.