def write(
self, feature_set: FeatureSet, dataframe: DataFrame, spark_client: SparkClient,
):
"""Loads the data from a feature set into the Historical Feature Store.
Args:
feature_set: object processed with feature_set informations.
dataframe: spark dataframe containing data from a feature set.
spark_client: client for spark connections with external services.
If the debug_mode is set to True, a temporary table with a name in the format:
historical_feature_store__{feature_set.name} will be created instead of writing
to the real historical feature store.
"""
dataframe = self._create_partitions(dataframe)
if self.debug_mode:
spark_client.create_temporary_view(
dataframe=dataframe,
name=f"historical_feature_store__{feature_set.name}",
)
return
s3_key = os.path.join("historical", feature_set.entity, feature_set.name)
spark_client.write_table(
dataframe=dataframe,
database=self.database,
table_name=feature_set.name,
partition_by=self.PARTITION_BY,
**self.db_config.get_options(s3_key),
)
def test_create_temporary_view(self, target_df: DataFrame,
spark_session: SparkSession) -> None:
# arrange
spark_client = SparkClient()
# act
spark_client.create_temporary_view(target_df, "temp_view")
result_df = spark_session.table("temp_view")
# assert
assert_dataframe_equality(target_df, result_df)
def write(
self,
feature_set: FeatureSet,
dataframe: DataFrame,
spark_client: SparkClient,
) -> None:
"""Loads the data from a feature set into the Historical Feature Store.
Args:
feature_set: object processed with feature_set informations.
dataframe: spark dataframe containing data from a feature set.
spark_client: client for spark connections with external services.
If the debug_mode is set to True, a temporary table with a name in the format:
historical_feature_store__{feature_set.name} will be created instead of writing
to the real historical feature store.
"""
dataframe = self._create_partitions(dataframe)
dataframe = self._apply_transformations(dataframe)
if self.interval_mode:
partition_overwrite_mode = spark_client.conn.conf.get(
"spark.sql.sources.partitionOverwriteMode").lower()
if partition_overwrite_mode != "dynamic":
raise RuntimeError(
"m=load_incremental_table, "
"spark.sql.sources.partitionOverwriteMode={}, "
"msg=partitionOverwriteMode have to "
"be configured to 'dynamic'".format(
partition_overwrite_mode))
if self.debug_mode:
spark_client.create_temporary_view(
dataframe=dataframe,
name=f"historical_feature_store__{feature_set.name}",
)
return
s3_key = os.path.join("historical", feature_set.entity,
feature_set.name)
spark_client.write_table(
dataframe=dataframe,
database=self.database,
table_name=feature_set.name,
partition_by=self.PARTITION_BY,
**self.db_config.get_options(s3_key),
)
def _write_in_debug_mode(
table_name: str, dataframe: DataFrame,
spark_client: SparkClient) -> Union[StreamingQuery, None]:
"""Creates a temporary table instead of writing to the real feature store."""
return spark_client.create_temporary_view(
dataframe=dataframe, name=f"online_feature_store__{table_name}")
