I’m using 1.3.
I noticed that Hive-related error occurs whenever I try to create a online feature group with 5 string columns.
This problem appears whenever online=True regardless off offline is False or True.
After checked out the syslog, i noticed that a Hive table with that name had already been created, and it was indeed listed in DataSets/featurestore as iris_1 but not listed in the Feature Store list.
@Duy_Quang_Nguy_n - I’ve seen this error happening when a previous feature group creation failed. There might be some directories left over.
Can you check in the Datasets (on the left sidebar of Hopsworks) >> Featurestore if you already have a directory called iris_1 and, if so remove it?
Otherwise you can try to increment the feature group version to 2.
Another suggestion is to set proper types for your Pandas dataframe. If you cast your features to double then they will be created as such in the feature store. As currently they are of type object Spark is going to cast them automatically to String.
I did notice what you said and have tried clean the dangling folder inside DataSets/featurestore as well as using a whole new name.
This is just a dummy datasets (iris dataset pull using sklearn) so im casting them to string just for the sake of illustration for this potential bug.
