Class FrameReaderDelta

java.lang.Object
org.apache.sysds.runtime.io.FrameReader
org.apache.sysds.runtime.io.FrameReaderDelta
Direct Known Subclasses:
FrameReaderDeltaParallel

public class FrameReaderDelta extends FrameReader
Single-threaded native Delta Lake reader for frames, built on the Spark-free Delta Kernel library. It opens the latest snapshot of a Delta table through the kernel (log replay, schema, data-file listing) and decodes the parquet data files directly into pre-allocated columns via parquet-mr's column API; tables with deletion vectors, partition columns or missing row statistics are read through the kernel's default engine instead (which applies deletion vectors and splices partition values). Schema and column names are derived from the Delta table schema.

Data is extracted column-at-a-time into primitive arrays (no per-cell boxing or FrameBlock.set dispatch) and the frame is constructed directly from typed column Arrays. Supported column types map to SystemDS value types: double, float, long, int, short, byte, boolean, and string. Neither the schema nor the dimensions need to be supplied; they are discovered from the table.