AHeise commented on a change in pull request #17501: URL: https://github.com/apache/flink/pull/17501#discussion_r768387505
########## File path: flink-formats/flink-parquet/src/main/java/org/apache/flink/formats/parquet/avro/AvroParquetRecordFormat.java ########## @@ -0,0 +1,236 @@ +/* + * 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.flink.formats.parquet.avro; + +import org.apache.flink.annotation.VisibleForTesting; +import org.apache.flink.api.common.typeinfo.TypeInformation; +import org.apache.flink.configuration.Configuration; +import org.apache.flink.connector.file.src.reader.StreamFormat; +import org.apache.flink.connector.file.src.util.CheckpointedPosition; +import org.apache.flink.core.fs.FSDataInputStream; + +import org.apache.avro.generic.GenericData; +import org.apache.avro.generic.GenericRecord; +import org.apache.avro.reflect.ReflectData; +import org.apache.avro.specific.SpecificData; +import org.apache.parquet.avro.AvroParquetReader; +import org.apache.parquet.hadoop.ParquetReader; +import org.apache.parquet.io.DelegatingSeekableInputStream; +import org.apache.parquet.io.InputFile; +import org.apache.parquet.io.SeekableInputStream; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +import javax.annotation.Nullable; + +import java.io.IOException; + +import static org.apache.flink.util.Preconditions.checkArgument; + +/** */ +public class AvroParquetRecordFormat<E> implements StreamFormat<E> { + + private static final long serialVersionUID = 1L; + + static final Logger LOG = LoggerFactory.getLogger(AvroParquetRecordFormat.class); + + private final TypeInformation<E> type; + + AvroParquetRecordFormat(TypeInformation<E> type) { + this.type = type; + } + + /** + * Creates a new reader to read avro {@link GenericRecord} from Parquet input stream. + * + * <p>Several wrapper classes haven be created to Flink abstraction become compatible with the + * parquet abstraction. Please refer to the inner classes {@link AvroParquetRecordReader}, + * {@link ParquetInputFile}, {@link FSDataInputStreamAdapter} for details. + */ + @Override + public Reader<E> createReader( + Configuration config, FSDataInputStream stream, long fileLen, long splitEnd) + throws IOException { + + // current version does not support splitting. + checkNotSplit(fileLen, splitEnd); + + return new AvroParquetRecordReader<E>( + AvroParquetReader.<E>builder(new ParquetInputFile(stream, fileLen)) + .withDataModel(getDataModel()) + .build()); + } + + /** + * Restores the reader from a checkpointed position. It is in fact identical since only {@link + * CheckpointedPosition#NO_OFFSET} as the {@code restoredOffset} is support. + */ + @Override + public Reader<E> restoreReader( + Configuration config, + FSDataInputStream stream, + long restoredOffset, + long fileLen, + long splitEnd) + throws IOException { + + // current version does not support splitting. + checkNotSplit(fileLen, splitEnd); + + checkArgument( + restoredOffset == CheckpointedPosition.NO_OFFSET, + "The restoredOffset should always be NO_OFFSET"); + + return createReader(config, stream, fileLen, splitEnd); + } + + @VisibleForTesting + GenericData getDataModel() { + Class<E> typeClass = getProducedType().getTypeClass(); + if (org.apache.avro.specific.SpecificRecordBase.class.isAssignableFrom(typeClass)) { + return SpecificData.get(); + } else if (org.apache.avro.generic.GenericRecord.class.isAssignableFrom(typeClass)) { + return GenericData.get(); + } else { + return ReflectData.get(); + } + } + + /** Current version does not support splitting. */ + @Override + public boolean isSplittable() { + return false; + } + + /** + * Gets the type produced by this format. This type will be the type produced by the file source + * as a whole. + */ + @Override + public TypeInformation<E> getProducedType() { + return type; + } + + private static void checkNotSplit(long fileLen, long splitEnd) { + if (splitEnd != fileLen) { + throw new IllegalArgumentException( + String.format( + "Current version of AvroParquetRecordFormat is not splittable, " + + "but found split end (%d) different from file length (%d)", + splitEnd, fileLen)); + } + } + + /** + * {@link StreamFormat.Reader} implementation. Using {@link ParquetReader} internally to read + * avro {@link GenericRecord} from parquet {@link InputFile}. + */ + private static class AvroParquetRecordReader<E> implements StreamFormat.Reader<E> { + + private final ParquetReader<E> parquetReader; + + private final long offset; + private long skipCount; + private final boolean checkpointed; + + private AvroParquetRecordReader(ParquetReader<E> parquetReader) { + this(parquetReader, CheckpointedPosition.NO_OFFSET, 0, false); + } + + private AvroParquetRecordReader( + ParquetReader<E> parquetReader, long offset, long skipCount, boolean checkpointed) { + this.parquetReader = parquetReader; + this.offset = offset; + this.skipCount = skipCount; + this.checkpointed = checkpointed; + } + + @Nullable + @Override + public E read() throws IOException { + E record = parquetReader.read(); + incrementPosition(); + return record; + } + + @Override + public void close() throws IOException { + parquetReader.close(); + } + + @Nullable + @Override + public CheckpointedPosition getCheckpointedPosition() { + return checkpointed ? new CheckpointedPosition(offset, skipCount) : null; Review comment: Yes, I think ultimately both `AvroParquetRecordFormat` and `ParquetVectorizedInputFormat` should share a common ancestor and be bulk-based. However, it's strictly speaking a performance improvement and probably improves maintainability in the long run (one base format where all the logic resides). For now, your solution already closes some gaps in the feature matrix and has a good value on its own. I'd pursue the other option later in a separate ticket/PR. -- 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: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
