Examples with HoodieDataBlock org.apache.hudi.common.table.log.block.HoodieDataBlock used on opensource projects

Example 16 with HoodieDataBlock

use of org.apache.hudi.common.table.log.block.HoodieDataBlock in project hudi by apache.

the class TestHoodieLogFormat method testAvroLogRecordReaderWithMixedInsertsCorruptsAndRollback.

@ParameterizedTest
@MethodSource("testArguments")
public void testAvroLogRecordReaderWithMixedInsertsCorruptsAndRollback(ExternalSpillableMap.DiskMapType diskMapType, boolean isCompressionEnabled, boolean readBlocksLazily) throws IOException, URISyntaxException, InterruptedException {
    // Write a 3 Data blocs with same InstantTime (written in same batch)
    Schema schema = HoodieAvroUtils.addMetadataFields(getSimpleSchema());
    // Set a small threshold so that every block is a new version
    Writer writer = HoodieLogFormat.newWriterBuilder().onParentPath(partitionPath).withFileExtension(HoodieLogFile.DELTA_EXTENSION).withFileId("test-fileid1").overBaseCommit("100").withFs(fs).build();
    // Write 1
    List<IndexedRecord> records1 = SchemaTestUtil.generateHoodieTestRecords(0, 100);
    Map<HoodieLogBlock.HeaderMetadataType, String> header = new HashMap<>();
    header.put(HoodieLogBlock.HeaderMetadataType.INSTANT_TIME, "100");
    header.put(HoodieLogBlock.HeaderMetadataType.SCHEMA, schema.toString());
    HoodieDataBlock dataBlock = getDataBlock(DEFAULT_DATA_BLOCK_TYPE, records1, header);
    writer.appendBlock(dataBlock);
    writer.appendBlock(dataBlock);
    writer.appendBlock(dataBlock);
    writer.close();
    FileCreateUtils.createDeltaCommit(basePath, "100", fs);
    // Append some arbit byte[] to the end of the log (mimics a partially written commit)
    fs = FSUtils.getFs(fs.getUri().toString(), fs.getConf());
    FSDataOutputStream outputStream = fs.append(writer.getLogFile().getPath());
    // create a block with
    outputStream.write(HoodieLogFormat.MAGIC);
    outputStream.writeLong(1000);
    outputStream.writeInt(HoodieLogBlockType.AVRO_DATA_BLOCK.ordinal());
    outputStream.writeInt(HoodieLogFormat.CURRENT_VERSION);
    // Write out a length that does not confirm with the content
    outputStream.writeLong(100);
    outputStream.flush();
    outputStream.close();
    // Append some arbit byte[] to the end of the log (mimics a partially written commit)
    fs = FSUtils.getFs(fs.getUri().toString(), fs.getConf());
    outputStream = fs.append(writer.getLogFile().getPath());
    // create a block with
    outputStream.write(HoodieLogFormat.MAGIC);
    outputStream.writeLong(1000);
    outputStream.writeInt(HoodieLogBlockType.AVRO_DATA_BLOCK.ordinal());
    outputStream.writeInt(HoodieLogFormat.CURRENT_VERSION);
    // Write out a length that does not confirm with the content
    outputStream.writeLong(100);
    outputStream.flush();
    outputStream.close();
    writer = HoodieLogFormat.newWriterBuilder().onParentPath(partitionPath).withFileExtension(HoodieLogFile.DELTA_EXTENSION).withFileId("test-fileid1").overBaseCommit("100").withFs(fs).build();
    writer.appendBlock(dataBlock);
    writer.close();
    // Append some arbit byte[] to the end of the log (mimics a partially written commit)
    fs = FSUtils.getFs(fs.getUri().toString(), fs.getConf());
    outputStream = fs.append(writer.getLogFile().getPath());
    // create a block with
    outputStream.write(HoodieLogFormat.MAGIC);
    outputStream.writeLong(1000);
    outputStream.writeInt(HoodieLogBlockType.AVRO_DATA_BLOCK.ordinal());
    outputStream.writeInt(HoodieLogFormat.CURRENT_VERSION);
    // Write out a length that does not confirm with the content
    outputStream.writeLong(100);
    outputStream.flush();
    outputStream.close();
    writer = HoodieLogFormat.newWriterBuilder().onParentPath(partitionPath).withFileExtension(HoodieLogFile.DELTA_EXTENSION).withFileId("test-fileid1").overBaseCommit("100").withFs(fs).build();
    // Write 1 rollback block for the last commit instant
    header.put(HoodieLogBlock.HeaderMetadataType.INSTANT_TIME, "101");
    header.put(HeaderMetadataType.TARGET_INSTANT_TIME, "100");
    header.put(HoodieLogBlock.HeaderMetadataType.COMMAND_BLOCK_TYPE, String.valueOf(HoodieCommandBlock.HoodieCommandBlockTypeEnum.ROLLBACK_PREVIOUS_BLOCK.ordinal()));
    HoodieCommandBlock commandBlock = new HoodieCommandBlock(header);
    writer.appendBlock(commandBlock);
    writer.close();
    List<String> allLogFiles = FSUtils.getAllLogFiles(fs, partitionPath, "test-fileid1", HoodieLogFile.DELTA_EXTENSION, "100").map(s -> s.getPath().toString()).collect(Collectors.toList());
    HoodieMergedLogRecordScanner scanner = HoodieMergedLogRecordScanner.newBuilder().withFileSystem(fs).withBasePath(basePath).withLogFilePaths(allLogFiles).withReaderSchema(schema).withLatestInstantTime("101").withMaxMemorySizeInBytes(10240L).withReadBlocksLazily(readBlocksLazily).withReverseReader(false).withBufferSize(bufferSize).withSpillableMapBasePath(BASE_OUTPUT_PATH).withDiskMapType(diskMapType).withBitCaskDiskMapCompressionEnabled(isCompressionEnabled).build();
    assertEquals(0, scanner.getTotalLogRecords(), "We would read 0 records");
    FileCreateUtils.deleteDeltaCommit(basePath, "100", fs);
}

Example 17 with HoodieDataBlock

use of org.apache.hudi.common.table.log.block.HoodieDataBlock in project hudi by apache.

the class TestHoodieLogFormat method testHugeLogFileWrite.

@Test
public void testHugeLogFileWrite() throws IOException, URISyntaxException, InterruptedException {
    Writer writer = HoodieLogFormat.newWriterBuilder().onParentPath(partitionPath).withFileExtension(HoodieLogFile.DELTA_EXTENSION).withFileId("test-fileid1").overBaseCommit("100").withFs(fs).withSizeThreshold(3L * 1024 * 1024 * 1024).build();
    Schema schema = getSimpleSchema();
    List<IndexedRecord> records = SchemaTestUtil.generateTestRecords(0, 1000);
    List<IndexedRecord> copyOfRecords = records.stream().map(record -> HoodieAvroUtils.rewriteRecord((GenericRecord) record, schema)).collect(Collectors.toList());
    Map<HoodieLogBlock.HeaderMetadataType, String> header = new HashMap<>();
    header.put(HoodieLogBlock.HeaderMetadataType.INSTANT_TIME, "100");
    header.put(HoodieLogBlock.HeaderMetadataType.SCHEMA, getSimpleSchema().toString());
    byte[] dataBlockContentBytes = getDataBlock(DEFAULT_DATA_BLOCK_TYPE, records, header).getContentBytes();
    HoodieLogBlock.HoodieLogBlockContentLocation logBlockContentLoc = new HoodieLogBlock.HoodieLogBlockContentLocation(new Configuration(), null, 0, dataBlockContentBytes.length, 0);
    HoodieDataBlock reusableDataBlock = new HoodieAvroDataBlock(null, Option.ofNullable(dataBlockContentBytes), false, logBlockContentLoc, Option.ofNullable(getSimpleSchema()), header, new HashMap<>(), HoodieRecord.RECORD_KEY_METADATA_FIELD);
    long writtenSize = 0;
    int logBlockWrittenNum = 0;
    while (writtenSize < Integer.MAX_VALUE) {
        AppendResult appendResult = writer.appendBlock(reusableDataBlock);
        assertTrue(appendResult.size() > 0);
        writtenSize += appendResult.size();
        logBlockWrittenNum++;
    }
    writer.close();
    Reader reader = HoodieLogFormat.newReader(fs, writer.getLogFile(), SchemaTestUtil.getSimpleSchema(), true, true);
    assertTrue(reader.hasNext(), "We wrote a block, we should be able to read it");
    HoodieLogBlock nextBlock = reader.next();
    assertEquals(DEFAULT_DATA_BLOCK_TYPE, nextBlock.getBlockType(), "The next block should be a data block");
    HoodieDataBlock dataBlockRead = (HoodieDataBlock) nextBlock;
    List<IndexedRecord> recordsRead = getRecords(dataBlockRead);
    assertEquals(copyOfRecords.size(), recordsRead.size(), "Read records size should be equal to the written records size");
    assertEquals(copyOfRecords, recordsRead, "Both records lists should be the same. (ordering guaranteed)");
    int logBlockReadNum = 1;
    while (reader.hasNext()) {
        reader.next();
        logBlockReadNum++;
    }
    assertEquals(logBlockWrittenNum, logBlockReadNum, "All written log should be correctly found");
    reader.close();
    // test writing oversize data block which should be rejected
    Writer oversizeWriter = HoodieLogFormat.newWriterBuilder().onParentPath(partitionPath).withFileExtension(HoodieLogFile.DELTA_EXTENSION).withFileId("test-fileid1").overBaseCommit("100").withSizeThreshold(3L * 1024 * 1024 * 1024).withFs(fs).build();
    List<HoodieLogBlock> dataBlocks = new ArrayList<>(logBlockWrittenNum + 1);
    for (int i = 0; i < logBlockWrittenNum + 1; i++) {
        dataBlocks.add(reusableDataBlock);
    }
    assertThrows(HoodieIOException.class, () -> {
        oversizeWriter.appendBlocks(dataBlocks);
    }, "Blocks appended may overflow. Please decrease log block size or log block amount");
    oversizeWriter.close();
}

Example 18 with HoodieDataBlock

use of org.apache.hudi.common.table.log.block.HoodieDataBlock in project hudi by apache.

the class TestHoodieLogFormat method testMultipleAppend.

@ParameterizedTest
@EnumSource(names = { "AVRO_DATA_BLOCK", "HFILE_DATA_BLOCK", "PARQUET_DATA_BLOCK" })
public void testMultipleAppend(HoodieLogBlockType dataBlockType) throws IOException, URISyntaxException, InterruptedException {
    Writer writer = HoodieLogFormat.newWriterBuilder().onParentPath(partitionPath).withFileExtension(HoodieLogFile.DELTA_EXTENSION).withFileId("test-fileid1").overBaseCommit("100").withFs(fs).build();
    List<IndexedRecord> records = SchemaTestUtil.generateTestRecords(0, 100);
    Map<HoodieLogBlock.HeaderMetadataType, String> header = new HashMap<>();
    header.put(HoodieLogBlock.HeaderMetadataType.INSTANT_TIME, "100");
    header.put(HoodieLogBlock.HeaderMetadataType.SCHEMA, getSimpleSchema().toString());
    HoodieDataBlock dataBlock = getDataBlock(dataBlockType, records, header);
    writer.appendBlock(dataBlock);
    long size1 = writer.getCurrentSize();
    writer.close();
    writer = HoodieLogFormat.newWriterBuilder().onParentPath(partitionPath).withFileExtension(HoodieLogFile.DELTA_EXTENSION).withFileId("test-fileid1").overBaseCommit("100").withFs(fs).build();
    records = SchemaTestUtil.generateTestRecords(0, 100);
    header.put(HoodieLogBlock.HeaderMetadataType.SCHEMA, getSimpleSchema().toString());
    dataBlock = getDataBlock(dataBlockType, records, header);
    writer.appendBlock(dataBlock);
    long size2 = writer.getCurrentSize();
    assertTrue(size2 > size1, "We just wrote a new block - size2 should be > size1");
    assertEquals(size2, fs.getFileStatus(writer.getLogFile().getPath()).getLen(), "Write should be auto-flushed. The size reported by FileStatus and the writer should match");
    writer.close();
    // Close and Open again and append 100 more records
    writer = HoodieLogFormat.newWriterBuilder().onParentPath(partitionPath).withFileExtension(HoodieLogFile.DELTA_EXTENSION).withFileId("test-fileid1").overBaseCommit("100").withFs(fs).build();
    records = SchemaTestUtil.generateTestRecords(0, 100);
    header.put(HoodieLogBlock.HeaderMetadataType.SCHEMA, getSimpleSchema().toString());
    dataBlock = getDataBlock(dataBlockType, records, header);
    writer.appendBlock(dataBlock);
    long size3 = writer.getCurrentSize();
    assertTrue(size3 > size2, "We just wrote a new block - size3 should be > size2");
    assertEquals(size3, fs.getFileStatus(writer.getLogFile().getPath()).getLen(), "Write should be auto-flushed. The size reported by FileStatus and the writer should match");
    writer.close();
    // Cannot get the current size after closing the log
    final Writer closedWriter = writer;
    assertThrows(IllegalStateException.class, () -> {
        closedWriter.getCurrentSize();
    }, "getCurrentSize should fail after the logAppender is closed");
}

Example 19 with HoodieDataBlock

use of org.apache.hudi.common.table.log.block.HoodieDataBlock in project hudi by apache.

the class TestHoodieLogFormat method testValidateCorruptBlockEndPosition.

@Test
public void testValidateCorruptBlockEndPosition() throws IOException, URISyntaxException, InterruptedException {
    Writer writer = HoodieLogFormat.newWriterBuilder().onParentPath(partitionPath).withFileExtension(HoodieLogFile.DELTA_EXTENSION).withFileId("test-fileid1").overBaseCommit("100").withFs(fs).build();
    List<IndexedRecord> records = SchemaTestUtil.generateTestRecords(0, 100);
    Map<HoodieLogBlock.HeaderMetadataType, String> header = new HashMap<>();
    header.put(HoodieLogBlock.HeaderMetadataType.INSTANT_TIME, "100");
    header.put(HoodieLogBlock.HeaderMetadataType.SCHEMA, getSimpleSchema().toString());
    HoodieDataBlock dataBlock = getDataBlock(DEFAULT_DATA_BLOCK_TYPE, records, header);
    writer.appendBlock(dataBlock);
    writer.close();
    // Append some arbit byte[] to the end of the log (mimics a partially written commit)
    fs = FSUtils.getFs(fs.getUri().toString(), fs.getConf());
    FSDataOutputStream outputStream = fs.append(writer.getLogFile().getPath());
    // create a block with
    outputStream.write(HoodieLogFormat.MAGIC);
    // Write out a length that does not confirm with the content
    outputStream.writeLong(474);
    outputStream.writeInt(HoodieLogBlockType.AVRO_DATA_BLOCK.ordinal());
    outputStream.writeInt(HoodieLogFormat.CURRENT_VERSION);
    // Write out a length that does not confirm with the content
    outputStream.writeLong(400);
    // Write out incomplete content
    outputStream.write("something-random".getBytes());
    // get corrupt block end position
    long corruptBlockEndPos = outputStream.getPos();
    outputStream.flush();
    outputStream.close();
    // Append a proper block again
    writer = HoodieLogFormat.newWriterBuilder().onParentPath(partitionPath).withFileExtension(HoodieLogFile.DELTA_EXTENSION).withFileId("test-fileid1").overBaseCommit("100").withFs(fs).build();
    records = SchemaTestUtil.generateTestRecords(0, 10);
    header.put(HoodieLogBlock.HeaderMetadataType.SCHEMA, getSimpleSchema().toString());
    dataBlock = getDataBlock(DEFAULT_DATA_BLOCK_TYPE, records, header);
    writer.appendBlock(dataBlock);
    writer.close();
    // Read data and corrupt block
    Reader reader = HoodieLogFormat.newReader(fs, writer.getLogFile(), SchemaTestUtil.getSimpleSchema());
    assertTrue(reader.hasNext(), "First block should be available");
    reader.next();
    assertTrue(reader.hasNext(), "We should have corrupted block next");
    HoodieLogBlock block = reader.next();
    assertEquals(HoodieLogBlockType.CORRUPT_BLOCK, block.getBlockType(), "The read block should be a corrupt block");
    // validate the corrupt block end position correctly.
    assertEquals(corruptBlockEndPos, block.getBlockContentLocation().get().getBlockEndPos());
    assertTrue(reader.hasNext(), "Third block should be available");
    reader.next();
    assertFalse(reader.hasNext(), "There should be no more block left");
    reader.close();
}

Example 20 with HoodieDataBlock

use of org.apache.hudi.common.table.log.block.HoodieDataBlock in project hudi by apache.

the class TestHoodieLogFormat method testRollover.

@Test
public void testRollover() throws IOException, InterruptedException, URISyntaxException {
    Writer writer = HoodieLogFormat.newWriterBuilder().onParentPath(partitionPath).withFileExtension(HoodieLogFile.DELTA_EXTENSION).withFileId("test-fileid1").overBaseCommit("100").withFs(fs).build();
    List<IndexedRecord> records = SchemaTestUtil.generateTestRecords(0, 100);
    Map<HoodieLogBlock.HeaderMetadataType, String> header = new HashMap<>();
    header.put(HoodieLogBlock.HeaderMetadataType.INSTANT_TIME, "100");
    header.put(HoodieLogBlock.HeaderMetadataType.SCHEMA, getSimpleSchema().toString());
    HoodieDataBlock dataBlock = getDataBlock(DEFAULT_DATA_BLOCK_TYPE, records, header);
    // Write out a block
    AppendResult firstAppend = writer.appendBlock(dataBlock);
    // Get the size of the block
    long size = writer.getCurrentSize();
    writer.close();
    assertEquals(0, firstAppend.offset());
    assertEquals(size, firstAppend.size());
    // Create a writer with the size threshold as the size we just wrote - so this has to roll
    writer = HoodieLogFormat.newWriterBuilder().onParentPath(partitionPath).withFileExtension(HoodieLogFile.DELTA_EXTENSION).withFileId("test-fileid1").overBaseCommit("100").withFs(fs).withSizeThreshold(size - 1).build();
    records = SchemaTestUtil.generateTestRecords(0, 100);
    dataBlock = getDataBlock(DEFAULT_DATA_BLOCK_TYPE, records, header);
    AppendResult secondAppend = writer.appendBlock(dataBlock);
    assertEquals(firstAppend.logFile(), secondAppend.logFile());
    assertNotEquals(0, secondAppend.offset());
    assertEquals(0, writer.getCurrentSize(), "This should be a new log file and hence size should be 0");
    assertEquals(2, writer.getLogFile().getLogVersion(), "Version should be rolled to 2");
    Path logFilePath = writer.getLogFile().getPath();
    assertFalse(fs.exists(logFilePath), "Path (" + logFilePath + ") must not exist");
    // Write one more block, which should not go to the new log file.
    records = SchemaTestUtil.generateTestRecords(0, 100);
    dataBlock = getDataBlock(DEFAULT_DATA_BLOCK_TYPE, records, header);
    AppendResult rolloverAppend = writer.appendBlock(dataBlock);
    assertNotEquals(secondAppend.logFile(), rolloverAppend.logFile());
    assertEquals(0, rolloverAppend.offset());
    writer.close();
}