NotImplementedError raised when using FeatureStore.push()

[singsinghai]

Expected Behavior

I'm currently starting to try simple usecase of Feast v0.40 for local development, I'm ingesting data from Airflow job to Postgres.

def process():
        ...
        store = FeatureStore(repo_path=f"{mother_path}/helpers/feast/")

        driver = Entity(name="driver_id", join_keys=["driver_id"], description="driver ID")

        table_name = 'feast_driver_hourly_stats'
        create_postgres_table(
            table_name=table_name, 
            schema="""
                driver_id SERIAL PRIMARY KEY,
                event_timestamp TIMESTAMP,
                surge_score FLOAT,
                trip_cancel INT,
                number_trip INT,
                latest_completed INT
            """
        )
        

        driver_stats_source = PostgreSQLSource(
            name=table_name,
            query=f"SELECT * FROM airflow.feast.{table_name}",
            timestamp_field="event_timestamp",
        )

        push_source = PushSource(
            name="push_source",
            batch_source=driver_stats_source
        )

        # Define the feature view (without specifying a source since we'll use a DataFrame)
        driver_feature_view = FeatureView(
            name="driver_feature_view",
            entities=[driver],
            ttl=None,  # Set TTL if needed
            schema=[
                Field(name="surge_score", dtype=Float32),
                Field(name="trip_cancel", dtype=Int64),
                Field(name="number_trip", dtype=Int64),
                Field(name="latest_completed", dtype=Int64),
            ],
            # online=True,
            stream_source=push_source
        )

        # Apply the feature store (register entities and feature views)
        store.apply([driver_feature_view])
        result = store.push("push_source", data, to=PushMode.OFFLINE)

I have ensured that the database, the schema, and the table exists. I expect after calling push(), the data will then be ingested (written) into the destined table.

Current Behavior

Suddenly, I got the follow error:

File "/opt/airflow/dags/test_feast.py", line 112, in process
    result = store.push("push_source", data, to=PushMode.OFFLINE)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/airflow/.local/lib/python3.11/site-packages/feast/feature_store.py", line 1427, in push
    self.write_to_offline_store(
  File "/home/airflow/.local/lib/python3.11/site-packages/feast/feature_store.py", line 1511, in write_to_offline_store
    provider.ingest_df_to_offline_store(feature_view, table)
  File "/home/airflow/.local/lib/python3.11/site-packages/feast/infra/passthrough_provider.py", line 303, in ingest_df_to_offline_store
    self.offline_write_batch(self.repo_config, feature_view, table, None)
  File "/home/airflow/.local/lib/python3.11/site-packages/feast/infra/passthrough_provider.py", line 180, in offline_write_batch
    self.offline_store.__class__.offline_write_batch(
  File "/home/airflow/.local/lib/python3.11/site-packages/feast/infra/offline_stores/offline_store.py", line 353, in offline_write_batch
    raise NotImplementedError

The write_to_offline_store() method might have called directly to an interface, causing the bug. I'm currently trying to review my configs/codes to spot any mistake and also read the push() method. However, I think I might have to raise an issue here.

Specifications

• Version: v0.40
• Platform: Airflow, Postgres
• Subsystem: Python 3.11

[tokoko]

postgres offline store doesn't implement offline_write_batch yet. There was a PR #3767 a while back, but it was abandoned. The relevant ticket is #3759.

[singsinghai]

postgres offline store doesn't implement offline_write_batch yet. There was a PR #3767 a while back, but it was abandoned. The relevant ticket is #3759.

Hi @tokoko , so currently with Postgres, Feast supports offline retrieval but not offline ingestion? Is there currently any available offline ingestion source yet?

[tokoko]

@singsinghai sorry, forgot to reply here. yes, that's right, you can't ingest data with feast, your best option is to populate postgres tables w/o feast and use feast only for retrieval.

Is there currently any available offline ingestion source yet?

Not sure I get the question tbh, if you're using postgres offline store, then no. there's no way to ingest data to offline store source tables with feast.

[tokoko]

Closing this as we already have another feature request ticket.