Exception: instance_type should not be a pipeline variable (<class 'sagemaker.workflow.parameters.ParameterString'>)

[qidewenwhen]

Describe the bug
After the release https://github.com/aws/sagemaker-python-sdk/releases/tag/v2.92.0, pipelines which was working before started failing with this exception, which relates to this commit

Exception: instance_type should not be a pipeline variable (<class 'sagemaker.workflow.parameters.ParameterString'>)

Note
This is a breaking change which can break some existing customer codes, so open this issue, post the explanation and fix options here, in case any one searches for it.

[qidewenwhen]

The reason for the breaking change
The error is thrown when we pass in any kind of pipeline variables e.g. Parameter, Properties etc. to the image_uris.retrieve()
The pipeline variables are placeholders which are parsed in execution time but the image_uris.retrieve()is retrieving image_uris in compile time (PySDK stage). In other words, the parameterization here won't work at all. And we have received several internal/external customers complaining on this e.g. GitHub issue.
Thus we decided to explicitly forbid passing in any pipeline variables to image_uris.retrieve().

To resolve this issue

• If you directly put any pipeline variables to image_uris.retrieve(), please replace them with python primitives e.g. plain string
• If you see this error thrown from any estimator/processor, that's because image_uri is not provided (i.e. None) here so that the instance_type is used to retrieve image_uri under the hood. The quick fix can be either:
  • Provide image_uri to the estimator/processor
  • Or Just simply update the instance_type to be a plain string (This can be an easier fix)