shangxinli commented on code in PR #14435: URL: https://github.com/apache/iceberg/pull/14435#discussion_r2563022736
########## parquet/src/main/java/org/apache/iceberg/parquet/ParquetFileMerger.java: ########## @@ -0,0 +1,421 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, + * software distributed under the License is distributed on an + * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY + * KIND, either express or implied. See the License for the + * specific language governing permissions and limitations + * under the License. + */ +package org.apache.iceberg.parquet; + +import static java.util.Collections.emptyMap; + +import java.io.IOException; +import java.util.List; +import java.util.Map; +import org.apache.iceberg.MetadataColumns; +import org.apache.iceberg.encryption.EncryptedOutputFile; +import org.apache.iceberg.io.InputFile; +import org.apache.iceberg.io.OutputFile; +import org.apache.iceberg.io.SeekableInputStream; +import org.apache.iceberg.relocated.com.google.common.collect.Lists; +import org.apache.parquet.bytes.BytesInput; +import org.apache.parquet.bytes.HeapByteBufferAllocator; +import org.apache.parquet.column.ColumnDescriptor; +import org.apache.parquet.column.Encoding; +import org.apache.parquet.column.ParquetProperties; +import org.apache.parquet.column.statistics.LongStatistics; +import org.apache.parquet.column.values.ValuesWriter; +import org.apache.parquet.column.values.delta.DeltaBinaryPackingValuesWriterForLong; +import org.apache.parquet.hadoop.ParquetFileReader; +import org.apache.parquet.hadoop.ParquetFileWriter; +import org.apache.parquet.hadoop.metadata.BlockMetaData; +import org.apache.parquet.hadoop.metadata.ColumnChunkMetaData; +import org.apache.parquet.hadoop.metadata.CompressionCodecName; +import org.apache.parquet.io.DelegatingSeekableInputStream; +import org.apache.parquet.schema.MessageType; +import org.apache.parquet.schema.PrimitiveType; +import org.apache.parquet.schema.Type; +import org.apache.parquet.schema.Types; + +/** + * Utility class for performing strict schema validation and merging of Parquet files at the + * row-group level. + * + * <p>This class ensures that all input files have identical Parquet schemas before merging. The + * merge operation is performed by copying row groups directly without + * serialization/deserialization, providing significant performance benefits over traditional + * read-rewrite approaches. + * + * <p>This class works with any Iceberg FileIO implementation (HadoopFileIO, S3FileIO, GCSFileIO, + * etc.), making it cloud-agnostic. + * + * <p>TODO: Encrypted tables are not supported + * + * <p>Key features: + * + * <ul> + * <li>Row group merging without deserialization using {@link ParquetFileWriter#appendFile} + * <li>Strict schema validation - all files must have identical {@link MessageType} + * <li>Metadata merging for Iceberg-specific footer data + * <li>Works with any FileIO implementation (local, S3, GCS, Azure, etc.) + * </ul> + * + * <p>Restrictions: + * + * <ul> + * <li>All files must have compatible schemas (identical {@link MessageType}) + * <li>Files must not be encrypted + * <li>Files must not have associated delete files or delete vectors + * <li>Table must not have a sort order (including z-ordered tables) + * </ul> + * + * <p>Typical usage: + * + * <pre> + * FileIO fileIO = table.io(); + * List<InputFile> inputFiles = Arrays.asList( + * fileIO.newInputFile("s3://bucket/file1.parquet"), + * fileIO.newInputFile("s3://bucket/file2.parquet") + * ); + * OutputFile outputFile = fileIO.newOutputFile("s3://bucket/merged.parquet"); + * long rowGroupSize = 128 * 1024 * 1024; // 128 MB + * int columnIndexTruncateLength = 64; // Default truncation length + * ParquetFileMerger.mergeFiles(inputFiles, outputFile, rowGroupSize, columnIndexTruncateLength, null); + * </pre> + */ +public class ParquetFileMerger { + // Default buffer sizes for DeltaBinaryPackingValuesWriter + private static final int DEFAULT_INITIAL_BUFFER_SIZE = 64 * 1024; // 64KB + private static final int DEFAULT_PAGE_SIZE_FOR_ENCODING = 64 * 1024; // 64KB + + private ParquetFileMerger() { + // Utility class - prevent instantiation + } + + /** + * Reads and validates that all input files have identical Parquet schemas. + * + * <p>This method works with any Iceberg FileIO implementation (S3FileIO, GCSFileIO, etc.). + * + * @param inputFiles List of Iceberg input files to validate + * @return the common Parquet schema if all files have identical schemas, null otherwise + */ + public static MessageType readAndValidateSchema(List<InputFile> inputFiles) { + try { + if (inputFiles == null || inputFiles.isEmpty()) { + return null; + } + + // Read schema from the first file + MessageType firstSchema = readSchema(inputFiles.get(0)); + + // Validate all remaining files have the same schema + for (int i = 1; i < inputFiles.size(); i++) { + MessageType currentSchema = readSchema(inputFiles.get(i)); + + if (!firstSchema.equals(currentSchema)) { + return null; + } + } + + return firstSchema; + } catch (IllegalArgumentException | IOException e) { + return null; + } + } + + /** + * Reads the Parquet schema from an Iceberg InputFile. + * + * @param inputFile Iceberg input file to read schema from + * @return MessageType schema of the Parquet file + * @throws IOException if reading fails + */ + private static MessageType readSchema(InputFile inputFile) throws IOException { + return ParquetFileReader.open(ParquetIO.file(inputFile)) + .getFooter() + .getFileMetaData() + .getSchema(); + } + + /** Internal method to merge files when schema is already known. */ + private static void mergeFilesWithSchema( + List<InputFile> inputFiles, + OutputFile outputFile, + MessageType schema, + long rowGroupSize, + int columnIndexTruncateLength, + Map<String, String> extraMetadata) + throws IOException { + try (ParquetFileWriter writer = + new ParquetFileWriter( + ParquetIO.file(outputFile), + schema, + ParquetFileWriter.Mode.CREATE, + rowGroupSize, + 0, // maxPaddingSize - hardcoded to 0 (same as ParquetWriter) + columnIndexTruncateLength, + ParquetProperties.DEFAULT_STATISTICS_TRUNCATE_LENGTH, + ParquetProperties.DEFAULT_PAGE_WRITE_CHECKSUM_ENABLED)) { + + writer.start(); + for (InputFile inputFile : inputFiles) { + writer.appendFile(ParquetIO.file(inputFile)); + } + + if (extraMetadata != null && !extraMetadata.isEmpty()) { + writer.end(extraMetadata); + } else { + writer.end(emptyMap()); + } + } + } + + /** Internal method to merge files with row IDs when base schema is already known. */ + private static void mergeFilesWithRowIdsAndSchema( + List<InputFile> inputFiles, + OutputFile outputFile, + List<Long> firstRowIds, + MessageType baseSchema, + long rowGroupSize, + int columnIndexTruncateLength, + Map<String, String> extraMetadata) + throws IOException { + // Extend schema to include _row_id column + MessageType extendedSchema = addRowIdColumn(baseSchema); + + // Create output writer with extended schema + try (ParquetFileWriter writer = + new ParquetFileWriter( + ParquetIO.file(outputFile), + extendedSchema, + ParquetFileWriter.Mode.CREATE, + rowGroupSize, + 0, // maxPaddingSize - hardcoded to 0 (same as ParquetWriter) + columnIndexTruncateLength, + ParquetProperties.DEFAULT_STATISTICS_TRUNCATE_LENGTH, + ParquetProperties.DEFAULT_PAGE_WRITE_CHECKSUM_ENABLED)) { + + writer.start(); + + // Get _row_id column descriptor from extended schema + ColumnDescriptor rowIdDescriptor = + extendedSchema.getColumnDescription(new String[] {MetadataColumns.ROW_ID.name()}); + + // Process each input file + for (int fileIdx = 0; fileIdx < inputFiles.size(); fileIdx++) { + InputFile inputFile = inputFiles.get(fileIdx); + long currentRowId = firstRowIds.get(fileIdx); + + try (ParquetFileReader reader = ParquetFileReader.open(ParquetIO.file(inputFile))) { + List<BlockMetaData> rowGroups = reader.getFooter().getBlocks(); + + for (BlockMetaData rowGroup : rowGroups) { + long rowCount = rowGroup.getRowCount(); + writer.startBlock(rowCount); + + // Copy all existing column chunks (binary copy) + copyColumnChunks(writer, baseSchema, inputFile, rowGroup); + + // Write new _row_id column chunk + writeRowIdColumnChunk(writer, rowIdDescriptor, currentRowId, rowCount); + currentRowId += rowCount; + writer.endBlock(); + } + } + } + + if (extraMetadata != null && !extraMetadata.isEmpty()) { + writer.end(extraMetadata); + } else { + writer.end(emptyMap()); + } + } + } + + /** + * Merges multiple Parquet files with optional row lineage preservation. + * + * <p>This method intelligently handles row lineage based on the input files and firstRowIds: + * + * <ul> + * <li>If firstRowIds is null/empty: performs simple binary copy merge + * <li>If files already have physical _row_id column: performs simple binary copy merge + * <li>Otherwise: synthesizes physical _row_id column from virtual metadata + * </ul> + * + * @param inputFiles List of Iceberg input files to merge + * @param encryptedOutputFile Encrypted output file for the merged result (encryption handled + * based on table configuration) + * @param schema Parquet schema from the input files (assumed already validated) + * @param firstRowIds Optional list of starting row IDs for each input file (null if no lineage + * needed) + * @param rowGroupSize Target row group size in bytes + * @param columnIndexTruncateLength Maximum length for min/max values in column index + * @param extraMetadata Additional metadata to include in the output file footer (can be null) + * @return true if output file has physical _row_id column, false otherwise + * @throws IOException if I/O error occurs during merge operation + */ + public static boolean mergeFilesWithOptionalRowIds( + List<InputFile> inputFiles, + EncryptedOutputFile encryptedOutputFile, + MessageType schema, + List<Long> firstRowIds, + long rowGroupSize, + int columnIndexTruncateLength, + Map<String, String> extraMetadata) + throws IOException { + // Get the encrypting output file (encryption applied based on table configuration) + OutputFile outputFile = encryptedOutputFile.encryptingOutputFile(); + + // Check if row lineage processing is requested + boolean needsRowLineageProcessing = firstRowIds != null && !firstRowIds.isEmpty(); + + if (needsRowLineageProcessing) { + if (schema.containsField(MetadataColumns.ROW_ID.name())) { + // Files already have physical _row_id - use simple binary copy (fastest!) + mergeFilesWithSchema( + inputFiles, outputFile, schema, rowGroupSize, columnIndexTruncateLength, extraMetadata); + return true; // Output has physical _row_id from input + } else { + // Files have virtual _row_id - synthesize physical column + mergeFilesWithRowIdsAndSchema( + inputFiles, + outputFile, + firstRowIds, + schema, + rowGroupSize, + columnIndexTruncateLength, + extraMetadata); + return true; // We just wrote physical _row_id + } + } else { + // No row lineage processing needed - simple binary merge + mergeFilesWithSchema( + inputFiles, outputFile, schema, rowGroupSize, columnIndexTruncateLength, extraMetadata); + return false; // No physical _row_id + } + } + + /** + * Extends a Parquet schema by adding the _row_id metadata column. + * + * @param baseSchema Original Parquet schema + * @return Extended schema with _row_id column added + */ + private static MessageType addRowIdColumn(MessageType baseSchema) { + // Create _row_id column: required int64 with Iceberg field ID for proper column mapping + PrimitiveType rowIdType = + Types.required(PrimitiveType.PrimitiveTypeName.INT64) + .id(MetadataColumns.ROW_ID.fieldId()) + .named(MetadataColumns.ROW_ID.name()); + + // Add to existing fields + List<Type> fields = Lists.newArrayList(baseSchema.getFields()); + fields.add(rowIdType); + + return new MessageType(baseSchema.getName(), fields); + } + + /** + * Writes a _row_id column chunk with sequential row IDs. + * + * <p>Uses DELTA_BINARY_PACKED encoding with ZSTD compression. For sequential data like row IDs, + * delta encoding is optimal as it encodes the deltas (which are all 1) rather than the absolute + * values, achieving excellent compression ratios. + * + * @param writer ParquetFileWriter to write to + * @param rowIdDescriptor Column descriptor for _row_id + * @param startRowId Starting row ID for this row group + * @param rowCount Number of rows in this row group + * @throws IOException if writing fails + */ + private static void writeRowIdColumnChunk( + ParquetFileWriter writer, ColumnDescriptor rowIdDescriptor, long startRowId, long rowCount) + throws IOException { + + // Start the column chunk with ZSTD compression for maximum compression of sequential data + writer.startColumn(rowIdDescriptor, rowCount, CompressionCodecName.ZSTD); Review Comment: Parquet can have column specific codec. It should be OK. But still better to use the same way. Modified the code to use the same compression codec as the existing columns instead of hardcoding ZSTD -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
