Skip to content

DataSource V2

DataSource V2 (DataSource API V2 or Data Source V2) is a new API for data sources in Spark SQL with the following abstractions:

DataSource V2 was tracked under SPARK-15689 DataSource V2 and was marked as fixed in Spark 2.3.0. This was subsequently rewamped alongside Catalog Plugin API and Multi-Catalog Support and released in Spark 3.0.0 (which could've safely been referred to as DataSource V3).

Query Planning and Execution

DataSource V2 relies on the DataSourceV2Strategy execution planning strategy for query planning.

Data Reading

DataSource V2 uses DataSourceV2Relation logical operator to represent data reading (aka data scan).

DataSourceV2Relation is planned (translated) to a ProjectExec with a DataSourceV2ScanExec physical operator (possibly under the FilterExec operator) when DataSourceV2Strategy execution planning strategy is executed.

When executed, DataSourceV2ScanExec physical operator creates a DataSourceRDD (or a ContinuousReader for Spark Structured Streaming).

DataSourceRDD uses InputPartitions for partitions, preferred locations, and computing partitions.

Data Writing

DataSource V2 uses WriteToDataSourceV2 and AppendData logical operators to represent data writing (over a DataSourceV2Relation logical operator). As of Spark SQL 2.4.0, WriteToDataSourceV2 operator was deprecated for the more specific AppendData operator (compare "data writing" to "data append" which is certainly more specific).

NOTE: One of the differences between WriteToDataSourceV2 and AppendData logical operators is that the former (WriteToDataSourceV2) uses DataSourceWriter directly while the latter (AppendData) uses DataSourceV2Relation to get the DataSourceWriter from.

WriteToDataSourceV2 and AppendData (with DataSourceV2Relation) logical operators are planned as (translated to) a WriteToDataSourceV2Exec physical operator.

When executed, WriteToDataSourceV2Exec physical operator...FIXME

Filter Pushdown Performance Optimization

DataSource V2 supports filter pushdown performance optimization for...FIXME

From Parquet Filter Pushdown in Apache Drill's documentation:

Filter pushdown is a performance optimization that prunes extraneous data while reading from a data source to reduce the amount of data to scan and read for queries with supported filter expressions. Pruning data reduces the I/O, CPU, and network overhead to optimize query performance.

Tip

Enable INFO logging level for the DataSourceV2Strategy logger to be informed what filters were pushed down.

References

Videos


Last update: 2021-04-20