Implementation:Apache Flink EnrichedRowData

Knowledge Sources	Apache_Flink
Domains	Connectors, Table
Last Updated	2026-02-09 00:00 GMT

Overview

EnrichedRowData is an internal implementation of the RowData interface that merges two separate row data instances (a fixed row and a mutable row) into a single composite row, using an index mapping to route field accesses to the correct underlying row.

Description

EnrichedRowData enables the file connector to augment physical rows read from files with additional columns such as partition values or file metadata without copying the original row data. It is backed by two RowData instances:

• A fixedRow that holds static, immutable data (e.g., partition column values).
• A mutableRow that can be swapped at runtime for performant iteration in hot code paths.

The class uses an indexMapping array to determine which underlying row provides each field. Positive indexes refer to positions in the mutable row, while negative indexes (with a -1 offset) refer to positions in the fixed row. For example, an index mapping of [0, 1, -1, -2, 2] means:

• Index 0 maps to mutable row position 0
• Index 1 maps to mutable row position 1
• Index -1 maps to fixed row position 0
• Index -2 maps to fixed row position 1
• Index 2 maps to mutable row position 2

The RowKind is always inherited from the mutable row. All RowData accessor methods (getBoolean, getInt, getString, getDecimal, etc.) delegate to either the fixed or mutable row based on the index mapping.

A static factory method from() computes the index mapping automatically from lists of field names for the produced row, the mutable row, and the fixed row. The companion static method computeIndexMapping() performs the actual index mapping computation.

Usage

Use EnrichedRowData when reading from file-based sources where the physical row data needs to be enriched with additional columns such as partition values, file metadata, or other computed fields. This is an internal class annotated with @Internal and is not intended for direct use by end users.

The typical workflow is:

1. Create an EnrichedRowData instance via the from() factory method, passing the fixed row and the field name lists.
2. For each record read from a file, call replaceMutableRow() to swap in the new physical row data.
3. Access the merged row through the standard RowData interface.

Code Reference

Source Location

• Repository: Apache_Flink
• File: flink-connectors/flink-connector-files/src/main/java/org/apache/flink/connector/file/table/EnrichedRowData.java
• Lines: 1-334

Signature

@Internal
public class EnrichedRowData implements RowData {

    public EnrichedRowData(RowData fixedRow, int[] indexMapping)

    public EnrichedRowData replaceMutableRow(RowData mutableRow)

    public static EnrichedRowData from(
            RowData fixedRow,
            List<String> producedRowFields,
            List<String> mutableRowFields,
            List<String> fixedRowFields)

    public static int[] computeIndexMapping(
            List<String> producedRowFields,
            List<String> mutableRowFields,
            List<String> fixedRowFields)
}

Import

import org.apache.flink.connector.file.table.EnrichedRowData;

I/O Contract

Inputs

Name	Type	Required	Description
fixedRow	RowData	Yes	The immutable row containing static field values (e.g., partition columns)
indexMapping	int[]	Yes	Array mapping composite row positions to underlying row positions; positive values index the mutable row, negative values (with -1 offset) index the fixed row
mutableRow	RowData	Yes (via replaceMutableRow)	The swappable row containing the physical record data read from a file
producedRowFields	List<String>	Yes (for factory method)	Ordered field names of the full composite row
mutableRowFields	List<String>	Yes (for factory method)	Ordered field names of the mutable row
fixedRowFields	List<String>	Yes (for factory method)	Ordered field names of the fixed row

Outputs

Name	Type	Description
EnrichedRowData	RowData	A composite RowData instance that transparently merges the fixed and mutable rows according to the index mapping
indexMapping	int[]	The computed index mapping array (from computeIndexMapping)

Usage Examples

Basic Usage

// Define field names for the composite, mutable, and fixed rows
List<String> producedFields = Arrays.asList("id", "name", "partition_date");
List<String> mutableFields = Arrays.asList("id", "name");
List<String> fixedFields = Arrays.asList("partition_date");

// Create the fixed row with partition value
GenericRowData fixedRow = new GenericRowData(1);
fixedRow.setField(0, StringData.fromString("2024-01-15"));

// Create the EnrichedRowData using the factory method
EnrichedRowData enrichedRow = EnrichedRowData.from(
        fixedRow, producedFields, mutableFields, fixedFields);

// For each record read from a file, swap in the mutable row
RowData physicalRow = readNextFromFile();
enrichedRow.replaceMutableRow(physicalRow);

// Access fields through the composite RowData interface
int id = enrichedRow.getInt(0);             // from mutable row
StringData name = enrichedRow.getString(1);  // from mutable row
StringData date = enrichedRow.getString(2);  // from fixed row

Related Pages

• Principle:Apache_Flink_Enriched_Row_Composition