Skip to content

Creating New Services

Optional Exercise: Updating a Service

If we want to make changes to the Feature Repository - for example, creating a new Feature Package, updating an existing Feature Service, or creating a new Feature Service - that is done locally. Here we'll walk through the approach that is recommended to make safe updates for new Feature Services.

Instructions for Updating a Feature Service

  1. View your local Feature Store Configuration in your terminal, and in particularly the Feature Service we just reviewed. This is located within the /shared folder in your Feature Store Configuration. Open it by calling vi feature_service.py
  2. Reviewing the ctr_prediction_service object, you can see that online_serving has been enabled and the feature packages listed here are the same as the Web UI.
  3. Attempt to add a new Feature Package to this Service, ad_group_ctr_performance.
  4. Once the change has been made, call tecton plan to test the changes. It is making a comparison to the Configuration that exists in the Tecton environment.
  5. Notice that adding a new Feature Package to an existing Feature Service causes the Feature Service to be recreated. Since this model could be used in production, it's important to consider the negative implications of changing this object.
  6. Reset the object ctr_prediction_service to its original state and create a challenger Feature Service for building a different model that includes ad_group_ctr_performance. Name this new object ctr_prediction_service_challenger.
  7. Call tecton plan and note that no object is being recreated. Call tecton apply to apply your changes to the Tecton environment and see that the update has been made via your Web UI.