Search in sources :

Example 26 with Stopwatch

use of org.apache.drill.shaded.guava.com.google.common.base.Stopwatch in project drill by apache.

the class PageReader method readCompressedPageV2.

/**
 * Reads a compressed v2 data page which excluded the repetition and definition level
 * sections from compression.
 * @return decompressed Parquet page data
 * @throws IOException
 */
protected DrillBuf readCompressedPageV2() throws IOException {
    Stopwatch timer = Stopwatch.createUnstarted();
    int inputSize = pageHeader.getCompressed_page_size();
    int repLevelSize = pageHeader.data_page_header_v2.getRepetition_levels_byte_length();
    int defLevelSize = pageHeader.data_page_header_v2.getDefinition_levels_byte_length();
    int compDataOffset = repLevelSize + defLevelSize;
    int outputSize = pageHeader.uncompressed_page_size;
    long start = dataReader.getPos();
    long timeToRead;
    DrillBuf inputPageData = null;
    DrillBuf outputPageData = this.allocator.buffer(outputSize);
    try {
        timer.start();
        // Read in both the uncompressed and compressed sections
        inputPageData = dataReader.getNext(inputSize);
        timeToRead = timer.elapsed(TimeUnit.NANOSECONDS);
        this.updateStats(pageHeader, "Page Read", start, timeToRead, inputSize, inputSize);
        timer.reset();
        timer.start();
        start = dataReader.getPos();
        // Write out the uncompressed section
        // Note that the following setBytes call to read the repetition and definition level sections
        // advances readerIndex in inputPageData but not writerIndex in outputPageData.
        outputPageData.setBytes(0, inputPageData, compDataOffset);
        // decompress from the start of compressed data to the end of the input buffer
        CompressionCodecName codecName = columnChunkMetaData.getCodec();
        BytesInputDecompressor decomp = codecFactory.getDecompressor(codecName);
        ByteBuffer input = inputPageData.nioBuffer(compDataOffset, inputSize - compDataOffset);
        ByteBuffer output = outputPageData.nioBuffer(compDataOffset, outputSize - compDataOffset);
        decomp.decompress(input, inputSize - compDataOffset, output, outputSize - compDataOffset);
        outputPageData.writerIndex(outputSize);
        timeToRead = timer.elapsed(TimeUnit.NANOSECONDS);
        if (logger.isTraceEnabled()) {
            logger.trace("Col: {}  readPos: {}  Uncompressed_size: {}  pageData: {}", columnChunkMetaData.toString(), dataReader.getPos(), outputSize, ByteBufUtil.hexDump(outputPageData));
        }
        this.updateStats(pageHeader, "Decompress", start, timeToRead, inputSize, outputSize);
    } finally {
        if (inputPageData != null) {
            inputPageData.release();
        }
    }
    return outputPageData;
}
Also used : CompressionCodecName(org.apache.parquet.hadoop.metadata.CompressionCodecName) Stopwatch(org.apache.drill.shaded.guava.com.google.common.base.Stopwatch) BytesInputDecompressor(org.apache.parquet.compression.CompressionCodecFactory.BytesInputDecompressor) ByteBuffer(java.nio.ByteBuffer) DrillBuf(io.netty.buffer.DrillBuf)

Example 27 with Stopwatch

use of org.apache.drill.shaded.guava.com.google.common.base.Stopwatch in project drill by apache.

the class PageReader method readPageHeader.

/**
 * Reads the next page header available in the backing input stream.
 * @throws IOException
 */
protected void readPageHeader() throws IOException {
    long start = dataReader.getPos();
    Stopwatch timer = Stopwatch.createStarted();
    this.pageHeader = Util.readPageHeader(dataReader);
    long timeToRead = timer.elapsed(TimeUnit.NANOSECONDS);
    long pageHeaderBytes = dataReader.getPos() - start;
    this.updateStats(pageHeader, "Page Header", start, timeToRead, pageHeaderBytes, pageHeaderBytes);
    if (logger.isTraceEnabled()) {
        logger.trace("ParquetTrace,{},{},{},{},{},{},{},{}", "Page Header Read", "", this.parentColumnReader.parentReader.getHadoopPath(), this.columnDescriptor.toString(), start, 0, 0, timeToRead);
    }
}
Also used : Stopwatch(org.apache.drill.shaded.guava.com.google.common.base.Stopwatch)

Example 28 with Stopwatch

use of org.apache.drill.shaded.guava.com.google.common.base.Stopwatch in project drill by apache.

the class Metadata method readBlockMeta.

/**
 * Read the parquet metadata from a file
 *
 * @param path to metadata file
 * @param dirsOnly true for {@link Metadata#METADATA_DIRECTORIES_FILENAME}
 *                 or false for {@link Metadata#OLD_METADATA_FILENAME} files reading
 * @param metaContext current metadata context
 */
private void readBlockMeta(Path path, boolean dirsOnly, MetadataContext metaContext, FileSystem fs) {
    Stopwatch timer = logger.isDebugEnabled() ? Stopwatch.createStarted() : null;
    Path metadataParentDir = Path.getPathWithoutSchemeAndAuthority(path.getParent());
    String metadataParentDirPath = metadataParentDir.toUri().getPath();
    ObjectMapper mapper = new ObjectMapper();
    final SimpleModule serialModule = new SimpleModule();
    serialModule.addDeserializer(SchemaPath.class, new SchemaPath.De());
    serialModule.addKeyDeserializer(Metadata_V2.ColumnTypeMetadata_v2.Key.class, new Metadata_V2.ColumnTypeMetadata_v2.Key.DeSerializer());
    serialModule.addKeyDeserializer(Metadata_V3.ColumnTypeMetadata_v3.Key.class, new Metadata_V3.ColumnTypeMetadata_v3.Key.DeSerializer());
    serialModule.addKeyDeserializer(ColumnTypeMetadata_v4.Key.class, new ColumnTypeMetadata_v4.Key.DeSerializer());
    AfterburnerModule module = new AfterburnerModule();
    module.setUseOptimizedBeanDeserializer(true);
    boolean isFileMetadata = path.toString().endsWith(METADATA_FILENAME);
    boolean isSummaryFile = path.toString().endsWith(METADATA_SUMMARY_FILENAME);
    mapper.registerModule(serialModule);
    mapper.registerModule(module);
    mapper.configure(DeserializationFeature.FAIL_ON_UNKNOWN_PROPERTIES, false);
    try (InputStream is = fs.open(path)) {
        boolean alreadyCheckedModification;
        boolean newMetadata = false;
        alreadyCheckedModification = metaContext.getStatus(metadataParentDirPath);
        if (dirsOnly) {
            parquetTableMetadataDirs = mapper.readValue(is, ParquetTableMetadataDirs.class);
            if (timer != null) {
                logger.debug("Took {} ms to read directories from directory cache file", timer.elapsed(TimeUnit.MILLISECONDS));
                timer.stop();
            }
            parquetTableMetadataDirs.updateRelativePaths(metadataParentDirPath);
            if (!alreadyCheckedModification && tableModified(parquetTableMetadataDirs.getDirectories(), path, metadataParentDir, metaContext, fs)) {
                parquetTableMetadataDirs = (createMetaFilesRecursivelyAsProcessUser(Path.getPathWithoutSchemeAndAuthority(path.getParent()), fs, true, null, true)).getRight();
                newMetadata = true;
            }
        } else {
            if (isFileMetadata) {
                parquetTableMetadata.assignFiles((mapper.readValue(is, FileMetadata.class)).getFiles());
                if (new MetadataVersion(parquetTableMetadata.getMetadataVersion()).isAtLeast(4, 0)) {
                    ((ParquetTableMetadata_v4) parquetTableMetadata).updateRelativePaths(metadataParentDirPath);
                }
                if (!alreadyCheckedModification && tableModified(parquetTableMetadata.getDirectories(), path, metadataParentDir, metaContext, fs)) {
                    parquetTableMetadata = (createMetaFilesRecursivelyAsProcessUser(Path.getPathWithoutSchemeAndAuthority(path.getParent()), fs, true, null, true)).getLeft();
                    newMetadata = true;
                }
            } else if (isSummaryFile) {
                MetadataSummary metadataSummary = mapper.readValue(is, Metadata_V4.MetadataSummary.class);
                parquetTableMetadata = new ParquetTableMetadata_v4(metadataSummary);
            } else {
                parquetTableMetadata = mapper.readValue(is, ParquetTableMetadataBase.class);
                if (new MetadataVersion(parquetTableMetadata.getMetadataVersion()).isAtLeast(3, 0)) {
                    ((Metadata_V3.ParquetTableMetadata_v3) parquetTableMetadata).updateRelativePaths(metadataParentDirPath);
                }
                if (!alreadyCheckedModification && tableModified((parquetTableMetadata.getDirectories()), path, metadataParentDir, metaContext, fs)) {
                    parquetTableMetadata = (createMetaFilesRecursivelyAsProcessUser(Path.getPathWithoutSchemeAndAuthority(path.getParent()), fs, true, null, true)).getLeft();
                    newMetadata = true;
                }
            }
            if (timer != null) {
                logger.debug("Took {} ms to read metadata from cache file", timer.elapsed(TimeUnit.MILLISECONDS));
                timer.stop();
            }
            if (!isSummaryFile) {
                List<? extends ParquetFileMetadata> files = parquetTableMetadata.getFiles();
                if (files != null) {
                    for (ParquetFileMetadata file : files) {
                        // DRILL-5009: Remove empty row groups unless it is the only row group
                        List<? extends RowGroupMetadata> rowGroups = file.getRowGroups();
                        if (rowGroups.size() == 1) {
                            continue;
                        }
                        rowGroups.removeIf(r -> r.getRowCount() == 0);
                    }
                }
            }
            if (newMetadata) {
                // if new metadata files were created, invalidate the existing metadata context
                metaContext.clear();
            }
        }
    } catch (IOException e) {
        logger.error("Failed to read '{}' metadata file", path, e);
        metaContext.setMetadataCacheCorrupted(true);
    }
}
Also used : ColumnTypeMetadata_v4(org.apache.drill.exec.store.parquet.metadata.Metadata_V4.ColumnTypeMetadata_v4) Stopwatch(org.apache.drill.shaded.guava.com.google.common.base.Stopwatch) FileMetadata(org.apache.drill.exec.store.parquet.metadata.Metadata_V4.FileMetadata) ParquetFileMetadata(org.apache.drill.exec.store.parquet.metadata.MetadataBase.ParquetFileMetadata) SchemaPath(org.apache.drill.common.expression.SchemaPath) ObjectMapper(com.fasterxml.jackson.databind.ObjectMapper) Path(org.apache.hadoop.fs.Path) SchemaPath(org.apache.drill.common.expression.SchemaPath) InputStream(java.io.InputStream) ParquetFileMetadata(org.apache.drill.exec.store.parquet.metadata.MetadataBase.ParquetFileMetadata) ParquetTableMetadata_v4(org.apache.drill.exec.store.parquet.metadata.Metadata_V4.ParquetTableMetadata_v4) IOException(java.io.IOException) AfterburnerModule(com.fasterxml.jackson.module.afterburner.AfterburnerModule) MetadataSummary(org.apache.drill.exec.store.parquet.metadata.Metadata_V4.MetadataSummary) SimpleModule(com.fasterxml.jackson.databind.module.SimpleModule)

Example 29 with Stopwatch

use of org.apache.drill.shaded.guava.com.google.common.base.Stopwatch in project drill by apache.

the class Metadata method createMetaFilesRecursively.

/**
 * Create the parquet metadata files for the directory at the given path and for any subdirectories.
 * Metadata cache files written to the disk contain relative paths. Returned Pair of metadata contains absolute paths.
 *
 * @param path to the directory of the parquet table
 * @param fs file system
 * @param allColumnsInteresting if set, store column metadata for all the columns
 * @param columnSet Set of columns for which column metadata has to be stored
 * @return Pair of parquet metadata. The left one is a parquet metadata for the table. The right one of the Pair is
 *         a metadata for all subdirectories (if they are present and there are no any parquet files in the
 *         {@code path} directory).
 * @throws IOException if parquet metadata can't be serialized and written to the json file
 */
private Pair<ParquetTableMetadata_v4, ParquetTableMetadataDirs> createMetaFilesRecursively(Path path, FileSystem fs, boolean allColumnsInteresting, Set<SchemaPath> columnSet) throws IOException {
    Stopwatch timer = logger.isDebugEnabled() ? Stopwatch.createStarted() : null;
    List<ParquetFileMetadata_v4> metaDataList = Lists.newArrayList();
    List<Path> directoryList = Lists.newArrayList();
    ConcurrentHashMap<ColumnTypeMetadata_v4.Key, ColumnTypeMetadata_v4> columnTypeInfoSet = new ConcurrentHashMap<>();
    FileStatus fileStatus = fs.getFileStatus(path);
    long dirTotalRowCount = 0;
    assert fileStatus.isDirectory() : "Expected directory";
    final Map<FileStatus, FileSystem> childFiles = new LinkedHashMap<>();
    for (final FileStatus file : DrillFileSystemUtil.listAll(fs, path, false)) {
        if (file.isDirectory()) {
            ParquetTableMetadata_v4 subTableMetadata = (createMetaFilesRecursively(file.getPath(), fs, allColumnsInteresting, columnSet)).getLeft();
            ConcurrentHashMap<ColumnTypeMetadata_v4.Key, ColumnTypeMetadata_v4> subTableColumnTypeInfo = subTableMetadata.getColumnTypeInfoMap();
            metaDataList.addAll((List<ParquetFileMetadata_v4>) subTableMetadata.getFiles());
            directoryList.addAll(subTableMetadata.getDirectories());
            directoryList.add(file.getPath());
            // TODO: We need a merge method that merges two columns with the same name but different types
            if (columnTypeInfoSet.isEmpty()) {
                columnTypeInfoSet.putAll(subTableColumnTypeInfo);
            } else {
                for (ColumnTypeMetadata_v4.Key key : subTableColumnTypeInfo.keySet()) {
                    ColumnTypeMetadata_v4 columnTypeMetadata_v4 = columnTypeInfoSet.get(key);
                    if (columnTypeMetadata_v4 == null) {
                        columnTypeMetadata_v4 = subTableColumnTypeInfo.get(key);
                    } else {
                        // as unknown
                        if (subTableColumnTypeInfo.get(key).totalNullCount < 0 || columnTypeMetadata_v4.totalNullCount < 0) {
                            columnTypeMetadata_v4.totalNullCount = NULL_COUNT_NOT_EXISTS;
                        } else {
                            columnTypeMetadata_v4.totalNullCount = columnTypeMetadata_v4.totalNullCount + subTableColumnTypeInfo.get(key).totalNullCount;
                        }
                    }
                    columnTypeInfoSet.put(key, columnTypeMetadata_v4);
                }
            }
            dirTotalRowCount = dirTotalRowCount + subTableMetadata.getTotalRowCount();
        } else {
            childFiles.put(file, fs);
        }
    }
    Metadata_V4.MetadataSummary metadataSummary = new Metadata_V4.MetadataSummary(SUPPORTED_VERSIONS.last().toString(), DrillVersionInfo.getVersion(), allColumnsInteresting || columnSet == null);
    ParquetTableMetadata_v4 parquetTableMetadata = new ParquetTableMetadata_v4(metadataSummary);
    if (childFiles.size() > 0) {
        List<ParquetFileAndRowCountMetadata> childFileAndRowCountMetadata = getParquetFileMetadata_v4(parquetTableMetadata, childFiles, allColumnsInteresting, columnSet);
        // If the columnTypeInfoSet is empty, add the columnTypeInfo from the parquetTableMetadata
        if (columnTypeInfoSet.isEmpty()) {
            columnTypeInfoSet.putAll(parquetTableMetadata.getColumnTypeInfoMap());
        }
        for (ParquetFileAndRowCountMetadata parquetFileAndRowCountMetadata : childFileAndRowCountMetadata) {
            metaDataList.add(parquetFileAndRowCountMetadata.getFileMetadata());
            dirTotalRowCount = dirTotalRowCount + parquetFileAndRowCountMetadata.getFileRowCount();
            Map<ColumnTypeMetadata_v4.Key, Long> totalNullCountMap = parquetFileAndRowCountMetadata.getTotalNullCountMap();
            for (ColumnTypeMetadata_v4.Key column : totalNullCountMap.keySet()) {
                ColumnTypeMetadata_v4 columnTypeMetadata_v4 = columnTypeInfoSet.get(column);
                // If the column is not present in columnTypeInfoSet, get it from parquetTableMetadata
                if (columnTypeMetadata_v4 == null) {
                    columnTypeMetadata_v4 = parquetTableMetadata.getColumnTypeInfoMap().get(column);
                }
                // as unknown
                if (columnTypeMetadata_v4.totalNullCount < 0 || totalNullCountMap.get(column) < 0) {
                    columnTypeMetadata_v4.totalNullCount = NULL_COUNT_NOT_EXISTS;
                } else {
                    columnTypeMetadata_v4.totalNullCount += totalNullCountMap.get(column);
                }
                columnTypeInfoSet.put(column, columnTypeMetadata_v4);
            }
        }
    }
    metadataSummary.directories = directoryList;
    parquetTableMetadata.assignFiles(metaDataList);
    // TODO: We need a merge method that merges two columns with the same name but different types
    if (metadataSummary.columnTypeInfo == null) {
        metadataSummary.columnTypeInfo = new ConcurrentHashMap<>();
    }
    metadataSummary.columnTypeInfo.putAll(columnTypeInfoSet);
    metadataSummary.allColumnsInteresting = allColumnsInteresting;
    metadataSummary.totalRowCount = dirTotalRowCount;
    parquetTableMetadata.metadataSummary = metadataSummary;
    for (String oldName : OLD_METADATA_FILENAMES) {
        fs.delete(new Path(path, oldName), false);
    }
    // relative paths in the metadata are only necessary for meta cache files.
    ParquetTableMetadata_v4 metadataTableWithRelativePaths = MetadataPathUtils.createMetadataWithRelativePaths(parquetTableMetadata, path);
    writeFile(metadataTableWithRelativePaths.fileMetadata, new Path(path, METADATA_FILENAME), fs);
    writeFile(metadataTableWithRelativePaths.getSummary(), new Path(path, METADATA_SUMMARY_FILENAME), fs);
    Metadata_V4.MetadataSummary metadataSummaryWithRelativePaths = metadataTableWithRelativePaths.getSummary();
    // Directories list will be empty at the leaf level directories. For sub-directories with both files and directories,
    // only the directories will be included in the list.
    writeFile(new ParquetTableMetadataDirs(metadataSummaryWithRelativePaths.directories), new Path(path, METADATA_DIRECTORIES_FILENAME), fs);
    if (timer != null) {
        logger.debug("Creating metadata files recursively took {} ms", timer.elapsed(TimeUnit.MILLISECONDS));
        timer.stop();
    }
    return Pair.of(parquetTableMetadata, new ParquetTableMetadataDirs(directoryList));
}
Also used : ColumnTypeMetadata_v4(org.apache.drill.exec.store.parquet.metadata.Metadata_V4.ColumnTypeMetadata_v4) FileStatus(org.apache.hadoop.fs.FileStatus) Stopwatch(org.apache.drill.shaded.guava.com.google.common.base.Stopwatch) LinkedHashMap(java.util.LinkedHashMap) MetadataSummary(org.apache.drill.exec.store.parquet.metadata.Metadata_V4.MetadataSummary) FileSystem(org.apache.hadoop.fs.FileSystem) ConcurrentHashMap(java.util.concurrent.ConcurrentHashMap) Path(org.apache.hadoop.fs.Path) SchemaPath(org.apache.drill.common.expression.SchemaPath) ParquetFileAndRowCountMetadata(org.apache.drill.exec.store.parquet.metadata.Metadata_V4.ParquetFileAndRowCountMetadata) ParquetTableMetadata_v4(org.apache.drill.exec.store.parquet.metadata.Metadata_V4.ParquetTableMetadata_v4) MetadataSummary(org.apache.drill.exec.store.parquet.metadata.Metadata_V4.MetadataSummary) ParquetFileMetadata_v4(org.apache.drill.exec.store.parquet.metadata.Metadata_V4.ParquetFileMetadata_v4)

Example 30 with Stopwatch

use of org.apache.drill.shaded.guava.com.google.common.base.Stopwatch in project drill by apache.

the class AssignmentCreator method getMappings.

/**
 * Does the work of creating the mappings for this AssignmentCreator
 * @return the minor fragment id to work units mapping
 */
private ListMultimap<Integer, T> getMappings() {
    Stopwatch watch = Stopwatch.createStarted();
    maxWork = (int) Math.ceil(units.size() / ((float) incomingEndpoints.size()));
    LinkedList<WorkEndpointListPair<T>> workList = getWorkList();
    LinkedList<WorkEndpointListPair<T>> unassignedWorkList;
    Map<DrillbitEndpoint, FragIteratorWrapper> endpointIterators = getEndpointIterators();
    // Assign up to maxCount per node based on locality.
    unassignedWorkList = assign(workList, endpointIterators, false);
    // Assign up to minCount per node in a round robin fashion.
    assignLeftovers(unassignedWorkList, endpointIterators, true);
    // Assign up to maxCount + leftovers per node based on locality.
    unassignedWorkList = assign(unassignedWorkList, endpointIterators, true);
    // Assign up to maxCount + leftovers per node in a round robin fashion.
    assignLeftovers(unassignedWorkList, endpointIterators, false);
    if (unassignedWorkList.size() != 0) {
        throw new DrillRuntimeException("There are still unassigned work units");
    }
    logger.debug("Took {} ms to assign {} work units to {} fragments", watch.elapsed(TimeUnit.MILLISECONDS), units.size(), incomingEndpoints.size());
    return mappings;
}
Also used : DrillbitEndpoint(org.apache.drill.exec.proto.CoordinationProtos.DrillbitEndpoint) Stopwatch(org.apache.drill.shaded.guava.com.google.common.base.Stopwatch) DrillRuntimeException(org.apache.drill.common.exceptions.DrillRuntimeException)

Aggregations

Stopwatch (org.apache.drill.shaded.guava.com.google.common.base.Stopwatch)68 IOException (java.io.IOException)13 Path (org.apache.hadoop.fs.Path)12 ArrayList (java.util.ArrayList)8 DrillRuntimeException (org.apache.drill.common.exceptions.DrillRuntimeException)8 FileStatus (org.apache.hadoop.fs.FileStatus)8 DrillBuf (io.netty.buffer.DrillBuf)7 ByteBuffer (java.nio.ByteBuffer)7 SchemaPath (org.apache.drill.common.expression.SchemaPath)7 HashMap (java.util.HashMap)5 RelNode (org.apache.calcite.rel.RelNode)5 SchemaChangeException (org.apache.drill.exec.exception.SchemaChangeException)4 DrillbitEndpoint (org.apache.drill.exec.proto.CoordinationProtos.DrillbitEndpoint)4 VectorContainer (org.apache.drill.exec.record.VectorContainer)4 SelectionVector4 (org.apache.drill.exec.record.selection.SelectionVector4)4 ValueVector (org.apache.drill.exec.vector.ValueVector)4 CompressionCodecName (org.apache.parquet.hadoop.metadata.CompressionCodecName)4 File (java.io.File)3 ResultSet (java.sql.ResultSet)3 ResultSetMetaData (java.sql.ResultSetMetaData)3