Package org.apache.sysds.runtime.io

Class WriterDelta

java.lang.Object

org.apache.sysds.runtime.io.MatrixWriter

org.apache.sysds.runtime.io.WriterDelta

public class WriterDelta
extends MatrixWriter

Single-threaded native Delta Lake writer for matrices, built on the Spark-free Delta Kernel library. It creates a Delta table at the target directory with an all-double schema c0..c(n-1) (replacing any existing table at that path), streams the MatrixBlock rows as columnar batches into parquet data files via the kernel's default engine, and commits the corresponding add-file actions to the transaction log.

Constructor Summary

Constructors

Constructor	Description
WriterDelta()

Method Summary

Modifier and Type	Method	Description
void	writeEmptyMatrixToHDFS(String fname, long rlen, long clen, int blen)	Writes a minimal entry to represent an empty matrix on hdfs.
long	writeMatrixFromStream(String fname, OOCStream<IndexedMatrixValue> stream, long rlen, long clen, int blen)	Consumes an out-of-core stream of matrix blocks and writes them to the target output path.
void	writeMatrixToHDFS(MatrixBlock src, String fname, long rlen, long clen, int blen, long nnz, boolean diag)

Methods inherited from class org.apache.sysds.runtime.io.MatrixWriter

createMatrixBlocksForReuse, getMatrixBlockForReuse, setForcedParallel, writeMatrixToHDFS

Methods inherited from class java.lang.Object

equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Details

WriterDelta

public WriterDelta()

Method Details

writeMatrixToHDFS

public void writeMatrixToHDFS(MatrixBlock src,
 String fname,
 long rlen,
 long clen,
 int blen,
 long nnz,
 boolean diag)
                       throws IOException

Specified by:

writeMatrixToHDFS in class MatrixWriter

Throws:

IOException

writeEmptyMatrixToHDFS

public void writeEmptyMatrixToHDFS(String fname,
 long rlen,
 long clen,
 int blen)
                            throws IOException

Description copied from class: MatrixWriter

Writes a minimal entry to represent an empty matrix on hdfs.

Specified by:

writeEmptyMatrixToHDFS in class MatrixWriter

Parameters:

fname - file name

rlen - number of rows

clen - number of columns

blen - number of rows/cols in block

Throws:

IOException - if IOException occurs

writeMatrixFromStream

public long writeMatrixFromStream(String fname,
 OOCStream<IndexedMatrixValue> stream,
 long rlen,
 long clen,
 int blen)
                           throws IOException

Description copied from class: MatrixWriter

Consumes an out-of-core stream of matrix blocks and writes them to the target output path. Implementations may choose single-file or multipart output depending on format and parallelism.

Specified by:

writeMatrixFromStream in class MatrixWriter

Parameters:

fname - The target output filename

stream - The OOC stream of matrix blocks to consume

rlen - The total number of rows in the matrix

clen - The total number of columns in the matrix

blen - The block size

Throws:

IOException - if an I/O error occurs