Reading fails when using DirectCodecFactory

Start with the supplied reproducer and the DirectCodecFactory.createDirectCodecFactory call, then compare its behavior with the Hadoop codec factory used by ParquetReader. Use the attached test.parquet file and verify that reading completes with 51,000 records and preserves the expected key/value content.

Describe the bug, including details regarding any error messages, version, and platform.

Hello folks,

I'm currently working on removing the hadoop-common dependency from the runtime in one of my projects.

As part of this process, I need to replace the "Hadoop" codec factory, which relies on certain classes from hadoop-common, with a codec factory that exclusively uses classes from parquet-hadoop. One potential option is DirectCodecFactory, but it comes with a problem.

For demonstration purposes, I’m using a simple key-value Snappy-compressed parquet file containing 51,000 records. Here’s a sample of the data:

When attempting to read this file using DirectCodecFactory, I encounter two issues:

1. The file fails to read completely. Near the end, it throws an error: Can't read value in column [key] optional binary key (STRING) = 0 at value 49,534 out of 51,000, 9,534 out of 11,000 in currentPage. Repetition level: 0, definition level: 1.
2. At record number 40,001, the key and value columns get mixed up, with the key column unexpectedly containing a value.

Parquet version: 1.15.0
Hadoop version: 3.4.1

Observations:

1. With the Hadoop codec factory the file can be read without any issues.

Here are tests demonstrating the issues:

package sandbox.parquet.reader;

import org.apache.hadoop.conf.Configuration;
import org.apache.parquet.bytes.DirectByteBufferAllocator;
import org.apache.parquet.column.ParquetProperties;
import org.apache.parquet.conf.ParquetConfiguration;
import org.apache.parquet.conf.PlainParquetConfiguration;
import org.apache.parquet.hadoop.CodecFactory;
import org.apache.parquet.hadoop.ParquetReader;
import org.apache.parquet.hadoop.api.InitContext;
import org.apache.parquet.hadoop.api.ReadSupport;
import org.apache.parquet.io.InputFile;
import org.apache.parquet.io.LocalInputFile;
import org.apache.parquet.io.api.Binary;
import org.apache.parquet.io.api.Converter;
import org.apache.parquet.io.api.GroupConverter;
import org.apache.parquet.io.api.PrimitiveConverter;
import org.apache.parquet.io.api.RecordMaterializer;
import org.apache.parquet.schema.GroupType;
import org.apache.parquet.schema.MessageType;
import org.apache.parquet.schema.Type;
import org.junit.jupiter.api.Assertions;
import org.junit.jupiter.api.Test;

import java.nio.file.Path;
import java.nio.file.Paths;
import java.util.Map;

public class ParquetReaderTest {

    @Test
    void readParquetFileAndVerifyRecordCount_usingDirectCodeFactory() throws Exception {
        Path filePath = Paths.get(ClassLoader.getSystemResource("test.parquet").toURI());
        int recordCount = 0;
        try(ParquetReader<String[]> parquetReader = createReaderWithDirectCodeFactory(filePath)) {
            while (parquetReader.read() != null) {
                recordCount++;
            }
        }

        Assertions.assertEquals(51_000, recordCount);
    }

    @Test
    void readParquetFileAndVerifyContent_usingDirectCodeFactory() throws Exception {
        Path filePath = Paths.get(ClassLoader.getSystemResource("test.parquet").toURI());
        int recordCount = 0;
        try(ParquetReader<String[]> parquetReader = createReaderWithHadoopCodeFactory(filePath)) {
            String[] record;
            while ((record = parquetReader.read()) != null) {
                Assertions.assertEquals("key_" + (recordCount + 1), record[0]);
                Assertions.assertEquals("value_" + (recordCount + 1), record[1]);
                recordCount++;
            }
        }

        Assertions.assertEquals(51_000, recordCount);
    }

    @Test
    void readParquetFileAndVerifyRecordCount_usingHadoopCodeFactory() throws Exception {
        Path filePath = Paths.get(ClassLoader.getSystemResource("test.parquet").toURI());
        int recordCount = 0;
        try(ParquetReader<String[]> parquetReader = createReaderWithHadoopCodeFactory(filePath)) {
            while (parquetReader.read() != null) {
                recordCount++;
            }
        }

        Assertions.assertEquals(51_000, recordCount);
    }

    @Test
    void readParquetFileAndVerifyContent_usingHadoopCodeFactory() throws Exception {
        Path filePath = Paths.get(ClassLoader.getSystemResource("test.parquet").toURI());
        int recordCount = 0;
        try(ParquetReader<String[]> parquetReader = createReaderWithDirectCodeFactory(filePath)) {
            String[] record;
            while ((record = parquetReader.read()) != null) {
                Assertions.assertEquals("key_" + (recordCount + 1), record[0]);
                Assertions.assertEquals("value_" + (recordCount + 1), record[1]);
                recordCount++;
            }
        }

        Assertions.assertEquals(51_000, recordCount);
    }

    private ParquetReader<String[]> createReaderWithDirectCodeFactory(Path file) throws Exception {
        return new ParquetReaderBuilder(new LocalInputFile(file), new PlainParquetConfiguration())
                .withCodecFactory(CodecFactory.createDirectCodecFactory(null,
                        DirectByteBufferAllocator.getInstance(),
                        ParquetProperties.DEFAULT_PAGE_SIZE))
                .build();
    }

    private ParquetReader<String[]> createReaderWithHadoopCodeFactory(Path file) throws Exception {
        // the Hadoop codec factory is created in ParquetReadOptions.Builder#build
        return new ParquetReaderBuilder(new LocalInputFile(file), new PlainParquetConfiguration())
                .build();
    }


    static class ParquetReaderBuilder extends ParquetReader.Builder<String[]> {

        ParquetReaderBuilder(InputFile file, ParquetConfiguration conf) {
            super(file, conf);
        }

        @Override
        protected ReadSupport<String[]> getReadSupport() {
            return new TestReadSupport();
        }
    }

    static class TestReadSupport extends ReadSupport<String[]> {


        @Override
        public ReadContext init(InitContext context) {
            return new ReadContext(context.getFileSchema());
        }

        @Override
        public RecordMaterializer<String[]> prepareForRead(Configuration configuration,
                                                           Map<String, String> keyValueMetaData,
                                                           MessageType fileSchema,
                                                           ReadContext readContext) {
            return new TestRecordMaterializer(fileSchema);
        }

        @Override
        public RecordMaterializer<String[]> prepareForRead(ParquetConfiguration configuration,
                                                           Map<String, String> keyValueMetaData,
                                                           MessageType fileSchema,
                                                           ReadContext readContext) {
            return new TestRecordMaterializer(fileSchema);
        }
    }

    static class TestRecordMaterializer extends RecordMaterializer<String[]> {

        private final TestRootGroupConverter root;

        TestRecordMaterializer(MessageType schema) {
            this.root = new TestRootGroupConverter(schema);
        }

        @Override
        public String[] getCurrentRecord() {
            return root.getCurrentRecord();
        }

        @Override
        public GroupConverter getRootConverter() {
            return root;
        }
    }

    static class TestRootGroupConverter extends GroupConverter {
        private String[] currentRecord;
        private final Converter[] converters;

        TestRootGroupConverter(GroupType schema) {
            converters = new Converter[schema.getFieldCount()];

            for (int i = 0; i < converters.length; i++) {
                final Type type = schema.getType(i);
                if (type.isPrimitive()) {
                    converters[i] = new TestPrimitiveConverter(this, i);
                } else {
                    throw new RuntimeException("Nested records not supported!");
                }
            }
        }

        @Override
        public Converter getConverter(int fieldIndex) {
            return converters[fieldIndex];
        }

        @Override
        public void start() {
            currentRecord = new String[converters.length];
        }

        @Override
        public void end() {
        }

        String[] getCurrentRecord() {
            return currentRecord;
        }
    }

    static class TestPrimitiveConverter extends PrimitiveConverter {

        private final TestRootGroupConverter parent;
        private final int index;

        TestPrimitiveConverter(TestRootGroupConverter parent, int index) {
            this.parent = parent;
            this.index = index;
        }

        @Override
        public void addBinary(Binary value) {
            parent.getCurrentRecord()[index] = value.toStringUsingUTF8();
        }

        @Override
        public void addBoolean(boolean value) {
            throw new UnsupportedOperationException();
        }

        @Override
        public void addDouble(double value) {
            throw new UnsupportedOperationException();
        }

        @Override
        public void addFloat(float value) {
            throw new UnsupportedOperationException();
        }

        @Override
        public void addInt(int value) {
            throw new UnsupportedOperationException();
        }

        @Override
        public void addLong(long value) {
            throw new UnsupportedOperationException();
        }
    }
}

Attaching the parquet file and a demo application:

• test.parquet.zip
• parquet-reader.zip

Component(s)

Core