Examples with HdfsEnvironment com.facebook.presto.hive.HdfsEnvironment used on opensource projects

Example 6 with HdfsEnvironment

use of com.facebook.presto.hive.HdfsEnvironment in project presto by prestodb.

the class ParquetPageSourceFactory method createParquetPageSource.

public static ConnectorPageSource createParquetPageSource(HdfsEnvironment hdfsEnvironment, String user, Configuration configuration, Path path, long start, long length, long fileSize, List<HiveColumnHandle> columns, SchemaTableName tableName, boolean useParquetColumnNames, DataSize maxReadBlockSize, boolean batchReaderEnabled, boolean verificationEnabled, TypeManager typeManager, StandardFunctionResolution functionResolution, TupleDomain<HiveColumnHandle> effectivePredicate, FileFormatDataSourceStats stats, HiveFileContext hiveFileContext, ParquetMetadataSource parquetMetadataSource, boolean columnIndexFilterEnabled) {
    AggregatedMemoryContext systemMemoryContext = newSimpleAggregatedMemoryContext();
    ParquetDataSource dataSource = null;
    try {
        FSDataInputStream inputStream = hdfsEnvironment.getFileSystem(user, path, configuration).openFile(path, hiveFileContext);
        dataSource = buildHdfsParquetDataSource(inputStream, path, stats);
        ParquetMetadata parquetMetadata = parquetMetadataSource.getParquetMetadata(dataSource, fileSize, hiveFileContext.isCacheable()).getParquetMetadata();
        if (!columns.isEmpty() && columns.stream().allMatch(hiveColumnHandle -> hiveColumnHandle.getColumnType() == AGGREGATED)) {
            return new AggregatedParquetPageSource(columns, parquetMetadata, typeManager, functionResolution);
        }
        FileMetaData fileMetaData = parquetMetadata.getFileMetaData();
        MessageType fileSchema = fileMetaData.getSchema();
        Optional<MessageType> message = columns.stream().filter(column -> column.getColumnType() == REGULAR || isPushedDownSubfield(column)).map(column -> getColumnType(typeManager.getType(column.getTypeSignature()), fileSchema, useParquetColumnNames, column, tableName, path)).filter(Optional::isPresent).map(Optional::get).map(type -> new MessageType(fileSchema.getName(), type)).reduce(MessageType::union);
        MessageType requestedSchema = message.orElse(new MessageType(fileSchema.getName(), ImmutableList.of()));
        ImmutableList.Builder<BlockMetaData> footerBlocks = ImmutableList.builder();
        for (BlockMetaData block : parquetMetadata.getBlocks()) {
            long firstDataPage = block.getColumns().get(0).getFirstDataPageOffset();
            if (firstDataPage >= start && firstDataPage < start + length) {
                footerBlocks.add(block);
            }
        }
        Map<List<String>, RichColumnDescriptor> descriptorsByPath = getDescriptors(fileSchema, requestedSchema);
        TupleDomain<ColumnDescriptor> parquetTupleDomain = getParquetTupleDomain(descriptorsByPath, effectivePredicate);
        Predicate parquetPredicate = buildPredicate(requestedSchema, parquetTupleDomain, descriptorsByPath);
        final ParquetDataSource finalDataSource = dataSource;
        ImmutableList.Builder<BlockMetaData> blocks = ImmutableList.builder();
        List<ColumnIndexStore> blockIndexStores = new ArrayList<>();
        for (BlockMetaData block : footerBlocks.build()) {
            Optional<ColumnIndexStore> columnIndexStore = ColumnIndexFilterUtils.getColumnIndexStore(parquetPredicate, finalDataSource, block, descriptorsByPath, columnIndexFilterEnabled);
            if (predicateMatches(parquetPredicate, block, finalDataSource, descriptorsByPath, parquetTupleDomain, columnIndexStore, columnIndexFilterEnabled)) {
                blocks.add(block);
                blockIndexStores.add(columnIndexStore.orElse(null));
                hiveFileContext.incrementCounter("parquet.blocksRead", 1);
                hiveFileContext.incrementCounter("parquet.rowsRead", block.getRowCount());
                hiveFileContext.incrementCounter("parquet.totalBytesRead", block.getTotalByteSize());
            } else {
                hiveFileContext.incrementCounter("parquet.blocksSkipped", 1);
                hiveFileContext.incrementCounter("parquet.rowsSkipped", block.getRowCount());
                hiveFileContext.incrementCounter("parquet.totalBytesSkipped", block.getTotalByteSize());
            }
        }
        MessageColumnIO messageColumnIO = getColumnIO(fileSchema, requestedSchema);
        ParquetReader parquetReader = new ParquetReader(messageColumnIO, blocks.build(), dataSource, systemMemoryContext, maxReadBlockSize, batchReaderEnabled, verificationEnabled, parquetPredicate, blockIndexStores, columnIndexFilterEnabled);
        ImmutableList.Builder<String> namesBuilder = ImmutableList.builder();
        ImmutableList.Builder<Type> typesBuilder = ImmutableList.builder();
        ImmutableList.Builder<Optional<Field>> fieldsBuilder = ImmutableList.builder();
        for (HiveColumnHandle column : columns) {
            checkArgument(column.getColumnType() == REGULAR || column.getColumnType() == SYNTHESIZED, "column type must be regular or synthesized column");
            String name = column.getName();
            Type type = typeManager.getType(column.getTypeSignature());
            namesBuilder.add(name);
            typesBuilder.add(type);
            if (column.getColumnType() == SYNTHESIZED) {
                Subfield pushedDownSubfield = getPushedDownSubfield(column);
                List<String> nestedColumnPath = nestedColumnPath(pushedDownSubfield);
                Optional<ColumnIO> columnIO = findNestedColumnIO(lookupColumnByName(messageColumnIO, pushedDownSubfield.getRootName()), nestedColumnPath);
                if (columnIO.isPresent()) {
                    fieldsBuilder.add(constructField(type, columnIO.get()));
                } else {
                    fieldsBuilder.add(Optional.empty());
                }
            } else if (getParquetType(type, fileSchema, useParquetColumnNames, column, tableName, path).isPresent()) {
                String columnName = useParquetColumnNames ? name : fileSchema.getFields().get(column.getHiveColumnIndex()).getName();
                fieldsBuilder.add(constructField(type, lookupColumnByName(messageColumnIO, columnName)));
            } else {
                fieldsBuilder.add(Optional.empty());
            }
        }
        return new ParquetPageSource(parquetReader, typesBuilder.build(), fieldsBuilder.build(), namesBuilder.build(), hiveFileContext.getStats());
    } catch (Exception e) {
        try {
            if (dataSource != null) {
                dataSource.close();
            }
        } catch (IOException ignored) {
        }
        if (e instanceof PrestoException) {
            throw (PrestoException) e;
        }
        if (e instanceof ParquetCorruptionException) {
            throw new PrestoException(HIVE_BAD_DATA, e);
        }
        if (e instanceof AccessControlException) {
            throw new PrestoException(PERMISSION_DENIED, e.getMessage(), e);
        }
        if (nullToEmpty(e.getMessage()).trim().equals("Filesystem closed") || e instanceof FileNotFoundException) {
            throw new PrestoException(HIVE_CANNOT_OPEN_SPLIT, e);
        }
        String message = format("Error opening Hive split %s (offset=%s, length=%s): %s", path, start, length, e.getMessage());
        if (e.getClass().getSimpleName().equals("BlockMissingException")) {
            throw new PrestoException(HIVE_MISSING_DATA, message, e);
        }
        throw new PrestoException(HIVE_CANNOT_OPEN_SPLIT, message, e);
    }
}

Example 7 with HdfsEnvironment

use of com.facebook.presto.hive.HdfsEnvironment in project presto by prestodb.

the class OrcBatchPageSourceFactory method createOrcPageSource.

public static ConnectorPageSource createOrcPageSource(OrcEncoding orcEncoding, HdfsEnvironment hdfsEnvironment, String sessionUser, Configuration configuration, Path path, long start, long length, long fileSize, List<HiveColumnHandle> columns, boolean useOrcColumnNames, TupleDomain<HiveColumnHandle> effectivePredicate, DateTimeZone hiveStorageTimeZone, TypeManager typeManager, StandardFunctionResolution functionResolution, DataSize maxBufferSize, DataSize streamBufferSize, boolean lazyReadSmallRanges, boolean orcBloomFiltersEnabled, FileFormatDataSourceStats stats, int domainCompactionThreshold, OrcFileTailSource orcFileTailSource, StripeMetadataSourceFactory stripeMetadataSourceFactory, HiveFileContext hiveFileContext, OrcReaderOptions orcReaderOptions, Optional<EncryptionInformation> encryptionInformation, DwrfEncryptionProvider dwrfEncryptionProvider) {
    checkArgument(domainCompactionThreshold >= 1, "domainCompactionThreshold must be at least 1");
    OrcDataSource orcDataSource;
    try {
        FSDataInputStream inputStream = hdfsEnvironment.getFileSystem(sessionUser, path, configuration).openFile(path, hiveFileContext);
        orcDataSource = new HdfsOrcDataSource(new OrcDataSourceId(path.toString()), fileSize, orcReaderOptions.getMaxMergeDistance(), maxBufferSize, streamBufferSize, lazyReadSmallRanges, inputStream, stats);
    } catch (Exception e) {
        if (nullToEmpty(e.getMessage()).trim().equals("Filesystem closed") || e instanceof FileNotFoundException) {
            throw new PrestoException(HIVE_CANNOT_OPEN_SPLIT, e);
        }
        throw new PrestoException(HIVE_CANNOT_OPEN_SPLIT, splitError(e, path, start, length), e);
    }
    OrcAggregatedMemoryContext systemMemoryUsage = new HiveOrcAggregatedMemoryContext();
    try {
        DwrfKeyProvider dwrfKeyProvider = new ProjectionBasedDwrfKeyProvider(encryptionInformation, columns, useOrcColumnNames, path);
        OrcReader reader = new OrcReader(orcDataSource, orcEncoding, orcFileTailSource, stripeMetadataSourceFactory, new HiveOrcAggregatedMemoryContext(), orcReaderOptions, hiveFileContext.isCacheable(), dwrfEncryptionProvider, dwrfKeyProvider, hiveFileContext.getStats());
        List<HiveColumnHandle> physicalColumns = getPhysicalHiveColumnHandles(columns, useOrcColumnNames, reader.getTypes(), path);
        ImmutableMap.Builder<Integer, Type> includedColumns = ImmutableMap.builder();
        ImmutableList.Builder<ColumnReference<HiveColumnHandle>> columnReferences = ImmutableList.builder();
        for (HiveColumnHandle column : physicalColumns) {
            if (column.getColumnType() == REGULAR) {
                Type type = typeManager.getType(column.getTypeSignature());
                includedColumns.put(column.getHiveColumnIndex(), type);
                columnReferences.add(new ColumnReference<>(column, column.getHiveColumnIndex(), type));
            }
        }
        if (!physicalColumns.isEmpty() && physicalColumns.stream().allMatch(hiveColumnHandle -> hiveColumnHandle.getColumnType() == AGGREGATED)) {
            return new AggregatedOrcPageSource(physicalColumns, reader.getFooter(), typeManager, functionResolution);
        }
        OrcPredicate predicate = new TupleDomainOrcPredicate<>(effectivePredicate, columnReferences.build(), orcBloomFiltersEnabled, Optional.of(domainCompactionThreshold));
        OrcBatchRecordReader recordReader = reader.createBatchRecordReader(includedColumns.build(), predicate, start, length, hiveStorageTimeZone, systemMemoryUsage, INITIAL_BATCH_SIZE);
        return new OrcBatchPageSource(recordReader, reader.getOrcDataSource(), physicalColumns, typeManager, systemMemoryUsage, stats, hiveFileContext.getStats());
    } catch (Exception e) {
        try {
            orcDataSource.close();
        } catch (IOException ignored) {
        }
        if (e instanceof PrestoException) {
            throw (PrestoException) e;
        }
        String message = splitError(e, path, start, length);
        if (e.getClass().getSimpleName().equals("BlockMissingException")) {
            throw new PrestoException(HIVE_MISSING_DATA, message, e);
        }
        throw new PrestoException(HIVE_CANNOT_OPEN_SPLIT, message, e);
    }
}

Example 8 with HdfsEnvironment

use of com.facebook.presto.hive.HdfsEnvironment in project presto by prestodb.

the class IcebergPageSourceProvider method createBatchOrcPageSource.

private static ConnectorPageSource createBatchOrcPageSource(HdfsEnvironment hdfsEnvironment, String user, Configuration configuration, Path path, long start, long length, boolean isCacheable, List<IcebergColumnHandle> regularColumns, TypeManager typeManager, TupleDomain<IcebergColumnHandle> effectivePredicate, OrcReaderOptions options, OrcEncoding orcEncoding, DataSize maxBufferSize, DataSize streamBufferSize, boolean lazyReadSmallRanges, boolean orcBloomFiltersEnabled, int domainCompactionThreshold, OrcFileTailSource orcFileTailSource, StripeMetadataSourceFactory stripeMetadataSourceFactory, FileFormatDataSourceStats stats, Optional<EncryptionInformation> encryptionInformation, DwrfEncryptionProvider dwrfEncryptionProvider) {
    OrcDataSource orcDataSource = null;
    try {
        ExtendedFileSystem fileSystem = hdfsEnvironment.getFileSystem(user, path, configuration);
        FileStatus fileStatus = fileSystem.getFileStatus(path);
        long fileSize = fileStatus.getLen();
        long modificationTime = fileStatus.getModificationTime();
        HiveFileContext hiveFileContext = new HiveFileContext(true, NO_CACHE_CONSTRAINTS, Optional.empty(), Optional.of(fileSize), modificationTime, false);
        FSDataInputStream inputStream = hdfsEnvironment.doAs(user, () -> fileSystem.openFile(path, hiveFileContext));
        orcDataSource = new HdfsOrcDataSource(new OrcDataSourceId(path.toString()), fileSize, options.getMaxMergeDistance(), maxBufferSize, streamBufferSize, lazyReadSmallRanges, inputStream, stats);
        // Todo: pass real columns to ProjectionBasedDwrfKeyProvider instead of ImmutableList.of()
        DwrfKeyProvider dwrfKeyProvider = new ProjectionBasedDwrfKeyProvider(encryptionInformation, ImmutableList.of(), true, path);
        RuntimeStats runtimeStats = new RuntimeStats();
        OrcReader reader = new OrcReader(orcDataSource, orcEncoding, orcFileTailSource, stripeMetadataSourceFactory, new HiveOrcAggregatedMemoryContext(), options, isCacheable, dwrfEncryptionProvider, dwrfKeyProvider, runtimeStats);
        List<HiveColumnHandle> physicalColumnHandles = new ArrayList<>(regularColumns.size());
        ImmutableMap.Builder<Integer, Type> includedColumns = ImmutableMap.builder();
        ImmutableList.Builder<TupleDomainOrcPredicate.ColumnReference<HiveColumnHandle>> columnReferences = ImmutableList.builder();
        List<IcebergOrcColumn> fileOrcColumns = getFileOrcColumns(reader);
        Map<Integer, IcebergOrcColumn> fileOrcColumnByIcebergId = fileOrcColumns.stream().filter(orcColumn -> orcColumn.getAttributes().containsKey(ORC_ICEBERG_ID_KEY)).collect(toImmutableMap(orcColumn -> Integer.parseInt(orcColumn.getAttributes().get(ORC_ICEBERG_ID_KEY)), orcColumn -> IcebergOrcColumn.copy(orcColumn).setIcebergColumnId(Optional.of(Integer.parseInt(orcColumn.getAttributes().get(ORC_ICEBERG_ID_KEY))))));
        Map<String, IcebergOrcColumn> fileOrcColumnsByName = uniqueIndex(fileOrcColumns, orcColumn -> orcColumn.getColumnName().toLowerCase(ENGLISH));
        int nextMissingColumnIndex = fileOrcColumnsByName.size();
        for (IcebergColumnHandle column : regularColumns) {
            IcebergOrcColumn icebergOrcColumn;
            boolean isExcludeColumn = false;
            if (fileOrcColumnByIcebergId.isEmpty()) {
                icebergOrcColumn = fileOrcColumnsByName.get(column.getName());
            } else {
                icebergOrcColumn = fileOrcColumnByIcebergId.get(column.getId());
                if (icebergOrcColumn == null) {
                    // Cannot get orc column from 'fileOrcColumnByIcebergId', which means SchemaEvolution may have happened, so we get orc column by column name.
                    icebergOrcColumn = fileOrcColumnsByName.get(column.getName());
                    if (icebergOrcColumn != null) {
                        isExcludeColumn = true;
                    }
                }
            }
            if (icebergOrcColumn != null) {
                HiveColumnHandle columnHandle = new HiveColumnHandle(// Todo: using orc file column name
                column.getName(), toHiveType(column.getType()), column.getType().getTypeSignature(), icebergOrcColumn.getOrcColumnId(), icebergOrcColumn.getColumnType(), Optional.empty(), Optional.empty());
                physicalColumnHandles.add(columnHandle);
                // Skip SchemaEvolution column
                if (!isExcludeColumn) {
                    includedColumns.put(columnHandle.getHiveColumnIndex(), typeManager.getType(columnHandle.getTypeSignature()));
                    columnReferences.add(new TupleDomainOrcPredicate.ColumnReference<>(columnHandle, columnHandle.getHiveColumnIndex(), typeManager.getType(columnHandle.getTypeSignature())));
                }
            } else {
                physicalColumnHandles.add(new HiveColumnHandle(column.getName(), toHiveType(column.getType()), column.getType().getTypeSignature(), nextMissingColumnIndex++, REGULAR, Optional.empty(), Optional.empty()));
            }
        }
        TupleDomain<HiveColumnHandle> hiveColumnHandleTupleDomain = effectivePredicate.transform(column -> {
            IcebergOrcColumn icebergOrcColumn;
            if (fileOrcColumnByIcebergId.isEmpty()) {
                icebergOrcColumn = fileOrcColumnsByName.get(column.getName());
            } else {
                icebergOrcColumn = fileOrcColumnByIcebergId.get(column.getId());
                if (icebergOrcColumn == null) {
                    // Cannot get orc column from 'fileOrcColumnByIcebergId', which means SchemaEvolution may have happened, so we get orc column by column name.
                    icebergOrcColumn = fileOrcColumnsByName.get(column.getName());
                }
            }
            return new HiveColumnHandle(column.getName(), toHiveType(column.getType()), column.getType().getTypeSignature(), // Note: the HiveColumnHandle.hiveColumnIndex starts from '0' while the IcebergColumnHandle.id starts from '1'
            icebergOrcColumn != null ? icebergOrcColumn.getOrcColumnId() : column.getId() - 1, icebergOrcColumn != null ? icebergOrcColumn.getColumnType() : REGULAR, Optional.empty(), Optional.empty());
        });
        OrcPredicate predicate = new TupleDomainOrcPredicate<>(hiveColumnHandleTupleDomain, columnReferences.build(), orcBloomFiltersEnabled, Optional.of(domainCompactionThreshold));
        OrcAggregatedMemoryContext systemMemoryUsage = new HiveOrcAggregatedMemoryContext();
        OrcBatchRecordReader recordReader = reader.createBatchRecordReader(includedColumns.build(), predicate, start, length, UTC, systemMemoryUsage, INITIAL_BATCH_SIZE);
        return new OrcBatchPageSource(recordReader, orcDataSource, physicalColumnHandles, typeManager, systemMemoryUsage, stats, runtimeStats);
    } catch (Exception e) {
        if (orcDataSource != null) {
            try {
                orcDataSource.close();
            } catch (IOException ignored) {
            }
        }
        if (e instanceof PrestoException) {
            throw (PrestoException) e;
        }
        String message = format("Error opening Iceberg split %s (offset=%s, length=%s): %s", path, start, length, e.getMessage());
        if (e instanceof BlockMissingException) {
            throw new PrestoException(ICEBERG_MISSING_DATA, message, e);
        }
        throw new PrestoException(ICEBERG_CANNOT_OPEN_SPLIT, message, e);
    }
}

Example 9 with HdfsEnvironment

use of com.facebook.presto.hive.HdfsEnvironment in project presto by prestodb.

the class IcebergFileWriterFactory method createParquetWriter.

private IcebergFileWriter createParquetWriter(Path outputPath, Schema icebergSchema, JobConf jobConf, ConnectorSession session, HdfsContext hdfsContext) {
    List<String> fileColumnNames = icebergSchema.columns().stream().map(Types.NestedField::name).collect(toImmutableList());
    List<Type> fileColumnTypes = icebergSchema.columns().stream().map(column -> toPrestoType(column.type(), typeManager)).collect(toImmutableList());
    try {
        FileSystem fileSystem = hdfsEnvironment.getFileSystem(session.getUser(), outputPath, jobConf);
        Callable<Void> rollbackAction = () -> {
            fileSystem.delete(outputPath, false);
            return null;
        };
        ParquetWriterOptions parquetWriterOptions = ParquetWriterOptions.builder().setMaxPageSize(getParquetWriterPageSize(session)).setMaxPageSize(getParquetWriterBlockSize(session)).build();
        return new IcebergParquetFileWriter(hdfsEnvironment.doAs(session.getUser(), () -> fileSystem.create(outputPath)), rollbackAction, fileColumnNames, fileColumnTypes, convert(icebergSchema, "table"), makeTypeMap(fileColumnTypes, fileColumnNames), parquetWriterOptions, IntStream.range(0, fileColumnNames.size()).toArray(), getCompressionCodec(session).getParquetCompressionCodec().get(), outputPath, hdfsEnvironment, hdfsContext);
    } catch (IOException e) {
        throw new PrestoException(ICEBERG_WRITER_OPEN_ERROR, "Error creating Parquet file", e);
    }
}

Example 10 with HdfsEnvironment

use of com.facebook.presto.hive.HdfsEnvironment in project presto by prestodb.

the class DeltaPageSourceProvider method createPageSource.

@Override
public ConnectorPageSource createPageSource(ConnectorTransactionHandle transactionHandle, ConnectorSession session, ConnectorSplit split, ConnectorTableLayoutHandle layout, List<ColumnHandle> columns, SplitContext splitContext) {
    DeltaSplit deltaSplit = (DeltaSplit) split;
    DeltaTableLayoutHandle deltaTableLayoutHandle = (DeltaTableLayoutHandle) layout;
    DeltaTableHandle deltaTableHandle = deltaTableLayoutHandle.getTable();
    HdfsContext hdfsContext = new HdfsContext(session, deltaSplit.getSchema(), deltaSplit.getTable(), deltaSplit.getFilePath(), false);
    Path filePath = new Path(deltaSplit.getFilePath());
    List<DeltaColumnHandle> deltaColumnHandles = columns.stream().map(DeltaColumnHandle.class::cast).collect(Collectors.toList());
    List<DeltaColumnHandle> regularColumnHandles = deltaColumnHandles.stream().filter(columnHandle -> columnHandle.getColumnType() != PARTITION).collect(Collectors.toList());
    ConnectorPageSource dataPageSource = createParquetPageSource(hdfsEnvironment, session.getUser(), hdfsEnvironment.getConfiguration(hdfsContext, filePath), filePath, deltaSplit.getStart(), deltaSplit.getLength(), deltaSplit.getFileSize(), regularColumnHandles, deltaTableHandle.toSchemaTableName(), getParquetMaxReadBlockSize(session), isParquetBatchReadsEnabled(session), isParquetBatchReaderVerificationEnabled(session), typeManager, deltaTableLayoutHandle.getPredicate(), fileFormatDataSourceStats, false);
    return new DeltaPageSource(deltaColumnHandles, convertPartitionValues(deltaColumnHandles, deltaSplit.getPartitionValues()), dataPageSource);
}