Example of migrated code can be found in the odap-framework-demo repository

Steps

1. Config

   feature_factory:
     catalog: "hive_metastore"
   	database: "odap_features_{entity}"
   	features:
   	    latest_table: "features_latest"
   	    dir_path: "dbfs:/odap_features/{entity}/features"
   

2. Metadata now have a column table instead of location

   ALTER TABLE odap_features.metadata_customer RENAME COLUMN location to table
   

3. Feature store table needs to be recreated using timestamp_keys instead of partition_keys

   Option 1

   • delete feature store completely a calculate again

   Option 2

   • delete databricks feature store in UI by hand
   • create a new table with timestamp_key using Databricks Feature store Python API

4. Feature notebook metadata now have to include a global key - table with the name of the table where those features should be saved

   • copy partial tables to different table
   • drop columns from original table

5. Refactor all segments to the new latest table name

6. optional update target store to the new PySpark target store + add a new timeshift widget

7. optional

   split job into multiple jobs for each feature store table