Similarity solutions with an Update Frequency using the "Workflow Similarity" capability can throw a "Malformed URL" in the prediction result when testing the solution - Known Error

Similarity solutions with an Update Frequency using the "Workflow Similarity" capability can throw a "Malformed URL" in the prediction result when testing the solution Description When you upgrade to Washington , it will add new system properties to use the new Workflow based solution training by default for all Predictive Intelligence capabilities. However, updates on Similarity solutions which have a set Update Frequency, when a prediction is invoked on the solution, it throws an error when attempting to update the model on the Prediction Server before making the prediction, and it will eventually return "Malformed URL" in the prediction result.

Steps to Reproduce

In a Washington instance, the system property [glide.platform_ml.api.enable_workflow_similarity] will be set to "true" by default to train Workflow Similarity-based solutions using the "Workflow Similarity" capability. When you train a Similarity solution which has a set Update Frequency set, the solution training will complete successfully, but after a number of updates on the solution, it will eventually return a "Malformed URL" on all predictions triggered on this Similarity solution by using the Test Solution tab on the [ml_solution] record.

Workaround There are two options to overcome this issue until we release a permanent fix for it: Switch to the Similarity-based solution or train the Workflow Similarity-based solution daily until the issue has been fixed.

Option-1:

Turn on the retrain option daily. Turn off the update from the solution definition, by setting the Update Frequency to "Do not update" Result: The model is updated daily by solution training, and predictions will not be delayed by updates, so it won't throw the Malformed URL error during the prediction.

Option-2:

Set the system property [glide.platform_ml.api.enable_workflow_similarity] to "false". Make sure the system property [glide.platform_ml.update.enable_async_similarity_update] is set to "true". This system property is used to enable async updates. If the system property does not exist, please create it in the "global" scope. Re-train the solution, ensuring that the similarity definition capability is not "Workflow Similarity", but instead it is just "Similarity". Check the [ml_solution] record, and add the Capability field to determine the ML capability. Result: The updates will now happen asynchronously via the Similarity-pipeline where there is no defect, and it will no longer throw a "Malformed URL" when making a prediction on the impacted Similarity solution, whilst still maintaining the updates as per the Update Frequency field on the solution definition.