spark_batch_config

Summary

Declare a tecton.SparkBatchConfig for configuring a Batch Source with a Data Source Function. The function takes in a SparkSession and an optional tecton.FilterContext, if supports_time_filtering=True. Returns a DataFrame.

Parameters

• data_delay (Optional[datetime.timedelta]) - By default, incremental materialization jobs run immediately at the end of the batch schedule period. This parameter configures how long they wait after the end of the period before starting, typically to ensure that all data has landed. For example, if a feature view has a batch_schedule of 1 day and one of the data source inputs has data_delay=timedelta(hours=1) set, then incremental materialization jobs will run at 01:00 UTC. Default: 0:00:00


• supports_time_filtering (Optional[bool]) - When set to True, the Data Source Function must take the filter_context parameter and implement time filtering logic. supports_time_filtering must be set to True if <data source>.get_dataframe() is called with start_time or end_time. supports_time_filtering must also be set to True if using tecton.declarative.FilteredSource with a Data Source when defining a FeatureView. The FeatureView will call the Data Source Function with the tecton.FilterContext, which has the start_time and end_time set. Default: false

Example

from tecton import spark_batch_config

@spark_batch_config(supports_time_filtering=True)
def redshift_data_source_function(spark, filter_context):
    spark_format = "com.databricks.spark.redshift"
    params = {
        "user": "<user name>",
        "password": os.environ["redshift_password"]
    }
    endpoint = "<redshift endpoint>"
    full_connection_string = f"jdbc:redshift://{endpoint};user={params['user']};password={params['password']}"
    df_reader = (
        spark.read.format(spark_format)
        .option("url", full_connection_string)
        .option("forward_spark_s3_credentials", "true")
    )
    df_reader = df_reader.option("dbtable", "your_table_name")
    df = df_reader_load()
    ts_column = "timestamp"
    df = df.withColumn(ts_column, col(ts_column).cast("timestamp"))
    # Handle time filtering
    if filter_context:
        if filter_context.start_time:
            df = df.where(col(ts_column) >= filter_context.start_time)
        if filter_context.end_time:
            df = df.where(col(ts_column) < filter_context.end_time)
    return df