Examples with FileFormatDataSourceStats io.trino.plugin.hive.FileFormatDataSourceStats used on opensource projects

Example 1 with FileFormatDataSourceStats

use of io.trino.plugin.hive.FileFormatDataSourceStats in project trino by trinodb.

the class OrcPageSourceFactory method createOrcPageSource.

private ConnectorPageSource createOrcPageSource(HdfsEnvironment hdfsEnvironment, ConnectorIdentity identity, Configuration configuration, Path path, long start, long length, long estimatedFileSize, List<HiveColumnHandle> columns, List<HiveColumnHandle> projections, boolean useOrcColumnNames, boolean isFullAcid, TupleDomain<HiveColumnHandle> effectivePredicate, DateTimeZone legacyFileTimeZone, OrcReaderOptions options, Optional<AcidInfo> acidInfo, OptionalInt bucketNumber, boolean originalFile, AcidTransaction transaction, FileFormatDataSourceStats stats) {
    for (HiveColumnHandle column : columns) {
        checkArgument(column.getColumnType() == REGULAR, "column type must be regular: %s", column);
    }
    checkArgument(!effectivePredicate.isNone());
    OrcDataSource orcDataSource;
    boolean originalFilesPresent = acidInfo.isPresent() && !acidInfo.get().getOriginalFiles().isEmpty();
    try {
        FileSystem fileSystem = hdfsEnvironment.getFileSystem(identity, path, configuration);
        FSDataInputStream inputStream = hdfsEnvironment.doAs(identity, () -> fileSystem.open(path));
        orcDataSource = new HdfsOrcDataSource(new OrcDataSourceId(path.toString()), estimatedFileSize, options, inputStream, stats);
    } catch (Exception e) {
        if (nullToEmpty(e.getMessage()).trim().equals("Filesystem closed") || e instanceof FileNotFoundException) {
            throw new TrinoException(HIVE_CANNOT_OPEN_SPLIT, e);
        }
        throw new TrinoException(HIVE_CANNOT_OPEN_SPLIT, splitError(e, path, start, length), e);
    }
    AggregatedMemoryContext memoryUsage = newSimpleAggregatedMemoryContext();
    try {
        Optional<OrcReader> optionalOrcReader = OrcReader.createOrcReader(orcDataSource, options);
        if (optionalOrcReader.isEmpty()) {
            return new EmptyPageSource();
        }
        OrcReader reader = optionalOrcReader.get();
        if (!originalFile && acidInfo.isPresent() && !acidInfo.get().isOrcAcidVersionValidated()) {
            validateOrcAcidVersion(path, reader);
        }
        List<OrcColumn> fileColumns = reader.getRootColumn().getNestedColumns();
        int actualColumnCount = columns.size() + (isFullAcid ? 3 : 0);
        List<OrcColumn> fileReadColumns = new ArrayList<>(actualColumnCount);
        List<Type> fileReadTypes = new ArrayList<>(actualColumnCount);
        List<OrcReader.ProjectedLayout> fileReadLayouts = new ArrayList<>(actualColumnCount);
        if (isFullAcid && !originalFilesPresent) {
            verifyAcidSchema(reader, path);
            Map<String, OrcColumn> acidColumnsByName = uniqueIndex(fileColumns, orcColumn -> orcColumn.getColumnName().toLowerCase(ENGLISH));
            fileColumns = ensureColumnNameConsistency(acidColumnsByName.get(AcidSchema.ACID_COLUMN_ROW_STRUCT.toLowerCase(ENGLISH)).getNestedColumns(), columns);
            fileReadColumns.add(acidColumnsByName.get(AcidSchema.ACID_COLUMN_ORIGINAL_TRANSACTION.toLowerCase(ENGLISH)));
            fileReadTypes.add(BIGINT);
            fileReadLayouts.add(fullyProjectedLayout());
            fileReadColumns.add(acidColumnsByName.get(AcidSchema.ACID_COLUMN_BUCKET.toLowerCase(ENGLISH)));
            fileReadTypes.add(INTEGER);
            fileReadLayouts.add(fullyProjectedLayout());
            fileReadColumns.add(acidColumnsByName.get(AcidSchema.ACID_COLUMN_ROW_ID.toLowerCase(ENGLISH)));
            fileReadTypes.add(BIGINT);
            fileReadLayouts.add(fullyProjectedLayout());
        }
        Map<String, OrcColumn> fileColumnsByName = ImmutableMap.of();
        if (useOrcColumnNames || isFullAcid) {
            verifyFileHasColumnNames(fileColumns, path);
            // Convert column names read from ORC files to lower case to be consistent with those stored in Hive Metastore
            fileColumnsByName = uniqueIndex(fileColumns, orcColumn -> orcColumn.getColumnName().toLowerCase(ENGLISH));
        }
        Map<String, List<List<String>>> projectionsByColumnName = ImmutableMap.of();
        Map<Integer, List<List<String>>> projectionsByColumnIndex = ImmutableMap.of();
        if (useOrcColumnNames || isFullAcid) {
            projectionsByColumnName = projections.stream().collect(Collectors.groupingBy(HiveColumnHandle::getBaseColumnName, mapping(OrcPageSourceFactory::getDereferencesAsList, toList())));
        } else {
            projectionsByColumnIndex = projections.stream().collect(Collectors.groupingBy(HiveColumnHandle::getBaseHiveColumnIndex, mapping(OrcPageSourceFactory::getDereferencesAsList, toList())));
        }
        TupleDomainOrcPredicateBuilder predicateBuilder = TupleDomainOrcPredicate.builder().setBloomFiltersEnabled(options.isBloomFiltersEnabled()).setDomainCompactionThreshold(domainCompactionThreshold);
        Map<HiveColumnHandle, Domain> effectivePredicateDomains = effectivePredicate.getDomains().orElseThrow(() -> new IllegalArgumentException("Effective predicate is none"));
        List<ColumnAdaptation> columnAdaptations = new ArrayList<>(columns.size());
        for (HiveColumnHandle column : columns) {
            OrcColumn orcColumn = null;
            OrcReader.ProjectedLayout projectedLayout = null;
            Map<Optional<HiveColumnProjectionInfo>, Domain> columnDomains = null;
            if (useOrcColumnNames || isFullAcid) {
                String columnName = column.getName().toLowerCase(ENGLISH);
                orcColumn = fileColumnsByName.get(columnName);
                if (orcColumn != null) {
                    projectedLayout = createProjectedLayout(orcColumn, projectionsByColumnName.get(columnName));
                    columnDomains = effectivePredicateDomains.entrySet().stream().filter(columnDomain -> columnDomain.getKey().getBaseColumnName().toLowerCase(ENGLISH).equals(columnName)).collect(toImmutableMap(columnDomain -> columnDomain.getKey().getHiveColumnProjectionInfo(), Map.Entry::getValue));
                }
            } else if (column.getBaseHiveColumnIndex() < fileColumns.size()) {
                orcColumn = fileColumns.get(column.getBaseHiveColumnIndex());
                if (orcColumn != null) {
                    projectedLayout = createProjectedLayout(orcColumn, projectionsByColumnIndex.get(column.getBaseHiveColumnIndex()));
                    columnDomains = effectivePredicateDomains.entrySet().stream().filter(columnDomain -> columnDomain.getKey().getBaseHiveColumnIndex() == column.getBaseHiveColumnIndex()).collect(toImmutableMap(columnDomain -> columnDomain.getKey().getHiveColumnProjectionInfo(), Map.Entry::getValue));
                }
            }
            Type readType = column.getType();
            if (orcColumn != null) {
                int sourceIndex = fileReadColumns.size();
                columnAdaptations.add(ColumnAdaptation.sourceColumn(sourceIndex));
                fileReadColumns.add(orcColumn);
                fileReadTypes.add(readType);
                fileReadLayouts.add(projectedLayout);
                // Add predicates on top-level and nested columns
                for (Map.Entry<Optional<HiveColumnProjectionInfo>, Domain> columnDomain : columnDomains.entrySet()) {
                    OrcColumn nestedColumn = getNestedColumn(orcColumn, columnDomain.getKey());
                    if (nestedColumn != null) {
                        predicateBuilder.addColumn(nestedColumn.getColumnId(), columnDomain.getValue());
                    }
                }
            } else {
                columnAdaptations.add(ColumnAdaptation.nullColumn(readType));
            }
        }
        OrcRecordReader recordReader = reader.createRecordReader(fileReadColumns, fileReadTypes, fileReadLayouts, predicateBuilder.build(), start, length, legacyFileTimeZone, memoryUsage, INITIAL_BATCH_SIZE, exception -> handleException(orcDataSource.getId(), exception), NameBasedFieldMapper::create);
        Optional<OrcDeletedRows> deletedRows = acidInfo.map(info -> new OrcDeletedRows(path.getName(), new OrcDeleteDeltaPageSourceFactory(options, identity, configuration, hdfsEnvironment, stats), identity, configuration, hdfsEnvironment, info, bucketNumber, memoryUsage));
        Optional<Long> originalFileRowId = acidInfo.filter(OrcPageSourceFactory::hasOriginalFiles).map(info -> OriginalFilesUtils.getPrecedingRowCount(acidInfo.get().getOriginalFiles(), path, hdfsEnvironment, identity, options, configuration, stats));
        if (transaction.isDelete()) {
            if (originalFile) {
                int bucket = bucketNumber.orElse(0);
                long startingRowId = originalFileRowId.orElse(0L);
                columnAdaptations.add(ColumnAdaptation.originalFileRowIdColumn(startingRowId, bucket));
            } else {
                columnAdaptations.add(ColumnAdaptation.rowIdColumn());
            }
        } else if (transaction.isUpdate()) {
            HiveUpdateProcessor updateProcessor = transaction.getUpdateProcessor().orElseThrow(() -> new IllegalArgumentException("updateProcessor not present"));
            List<HiveColumnHandle> dependencyColumns = projections.stream().filter(HiveColumnHandle::isBaseColumn).collect(toImmutableList());
            if (originalFile) {
                int bucket = bucketNumber.orElse(0);
                long startingRowId = originalFileRowId.orElse(0L);
                columnAdaptations.add(updatedRowColumnsWithOriginalFiles(startingRowId, bucket, updateProcessor, dependencyColumns));
            } else {
                columnAdaptations.add(updatedRowColumns(updateProcessor, dependencyColumns));
            }
        }
        return new OrcPageSource(recordReader, columnAdaptations, orcDataSource, deletedRows, originalFileRowId, memoryUsage, stats);
    } catch (Exception e) {
        try {
            orcDataSource.close();
        } catch (IOException ignored) {
        }
        if (e instanceof TrinoException) {
            throw (TrinoException) e;
        }
        String message = splitError(e, path, start, length);
        if (e instanceof BlockMissingException) {
            throw new TrinoException(HIVE_MISSING_DATA, message, e);
        }
        throw new TrinoException(HIVE_CANNOT_OPEN_SPLIT, message, e);
    }
}

Example 2 with FileFormatDataSourceStats

use of io.trino.plugin.hive.FileFormatDataSourceStats in project trino by trinodb.

the class ParquetPageSourceFactory method createPageSource.

/**
 * This method is available for other callers to use directly.
 */
public static ReaderPageSource createPageSource(Path path, long start, long length, long estimatedFileSize, List<HiveColumnHandle> columns, TupleDomain<HiveColumnHandle> effectivePredicate, boolean useColumnNames, HdfsEnvironment hdfsEnvironment, Configuration configuration, ConnectorIdentity identity, DateTimeZone timeZone, FileFormatDataSourceStats stats, ParquetReaderOptions options) {
    // Ignore predicates on partial columns for now.
    effectivePredicate = effectivePredicate.filter((column, domain) -> column.isBaseColumn());
    MessageType fileSchema;
    MessageType requestedSchema;
    MessageColumnIO messageColumn;
    ParquetReader parquetReader;
    ParquetDataSource dataSource = null;
    try {
        FileSystem fileSystem = hdfsEnvironment.getFileSystem(identity, path, configuration);
        FSDataInputStream inputStream = hdfsEnvironment.doAs(identity, () -> fileSystem.open(path));
        dataSource = new HdfsParquetDataSource(new ParquetDataSourceId(path.toString()), estimatedFileSize, inputStream, stats, options);
        ParquetMetadata parquetMetadata = MetadataReader.readFooter(dataSource);
        FileMetaData fileMetaData = parquetMetadata.getFileMetaData();
        fileSchema = fileMetaData.getSchema();
        Optional<MessageType> message = projectSufficientColumns(columns).map(projection -> projection.get().stream().map(HiveColumnHandle.class::cast).collect(toUnmodifiableList())).orElse(columns).stream().filter(column -> column.getColumnType() == REGULAR).map(column -> getColumnType(column, fileSchema, useColumnNames)).filter(Optional::isPresent).map(Optional::get).map(type -> new MessageType(fileSchema.getName(), type)).reduce(MessageType::union);
        requestedSchema = message.orElse(new MessageType(fileSchema.getName(), ImmutableList.of()));
        messageColumn = getColumnIO(fileSchema, requestedSchema);
        Map<List<String>, RichColumnDescriptor> descriptorsByPath = getDescriptors(fileSchema, requestedSchema);
        TupleDomain<ColumnDescriptor> parquetTupleDomain = options.isIgnoreStatistics() ? TupleDomain.all() : getParquetTupleDomain(descriptorsByPath, effectivePredicate, fileSchema, useColumnNames);
        Predicate parquetPredicate = buildPredicate(requestedSchema, parquetTupleDomain, descriptorsByPath, timeZone);
        long nextStart = 0;
        ImmutableList.Builder<BlockMetaData> blocks = ImmutableList.builder();
        ImmutableList.Builder<Long> blockStarts = ImmutableList.builder();
        ImmutableList.Builder<Optional<ColumnIndexStore>> columnIndexes = ImmutableList.builder();
        for (BlockMetaData block : parquetMetadata.getBlocks()) {
            long firstDataPage = block.getColumns().get(0).getFirstDataPageOffset();
            Optional<ColumnIndexStore> columnIndex = getColumnIndexStore(dataSource, block, descriptorsByPath, parquetTupleDomain, options);
            if (start <= firstDataPage && firstDataPage < start + length && predicateMatches(parquetPredicate, block, dataSource, descriptorsByPath, parquetTupleDomain, columnIndex)) {
                blocks.add(block);
                blockStarts.add(nextStart);
                columnIndexes.add(columnIndex);
            }
            nextStart += block.getRowCount();
        }
        parquetReader = new ParquetReader(Optional.ofNullable(fileMetaData.getCreatedBy()), messageColumn, blocks.build(), Optional.of(blockStarts.build()), dataSource, timeZone, newSimpleAggregatedMemoryContext(), options, parquetPredicate, columnIndexes.build());
    } catch (Exception e) {
        try {
            if (dataSource != null) {
                dataSource.close();
            }
        } catch (IOException ignored) {
        }
        if (e instanceof TrinoException) {
            throw (TrinoException) e;
        }
        if (e instanceof ParquetCorruptionException) {
            throw new TrinoException(HIVE_BAD_DATA, e);
        }
        if (nullToEmpty(e.getMessage()).trim().equals("Filesystem closed") || e instanceof FileNotFoundException) {
            throw new TrinoException(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 instanceof BlockMissingException) {
            throw new TrinoException(HIVE_MISSING_DATA, message, e);
        }
        throw new TrinoException(HIVE_CANNOT_OPEN_SPLIT, message, e);
    }
    Optional<ReaderColumns> readerProjections = projectBaseColumns(columns);
    List<HiveColumnHandle> baseColumns = readerProjections.map(projection -> projection.get().stream().map(HiveColumnHandle.class::cast).collect(toUnmodifiableList())).orElse(columns);
    for (HiveColumnHandle column : baseColumns) {
        checkArgument(column == PARQUET_ROW_INDEX_COLUMN || column.getColumnType() == REGULAR, "column type must be REGULAR: %s", column);
    }
    ImmutableList.Builder<Type> trinoTypes = ImmutableList.builder();
    ImmutableList.Builder<Optional<Field>> internalFields = ImmutableList.builder();
    ImmutableList.Builder<Boolean> rowIndexColumns = ImmutableList.builder();
    for (HiveColumnHandle column : baseColumns) {
        trinoTypes.add(column.getBaseType());
        rowIndexColumns.add(column == PARQUET_ROW_INDEX_COLUMN);
        if (column == PARQUET_ROW_INDEX_COLUMN) {
            internalFields.add(Optional.empty());
        } else {
            internalFields.add(Optional.ofNullable(getParquetType(column, fileSchema, useColumnNames)).flatMap(field -> {
                String columnName = useColumnNames ? column.getBaseColumnName() : fileSchema.getFields().get(column.getBaseHiveColumnIndex()).getName();
                return constructField(column.getBaseType(), lookupColumnByName(messageColumn, columnName));
            }));
        }
    }
    ConnectorPageSource parquetPageSource = new ParquetPageSource(parquetReader, trinoTypes.build(), rowIndexColumns.build(), internalFields.build());
    return new ReaderPageSource(parquetPageSource, readerProjections);
}

Example 3 with FileFormatDataSourceStats

use of io.trino.plugin.hive.FileFormatDataSourceStats in project trino by trinodb.

the class TestCheckpointEntryIterator method createCheckpointEntryIterator.

private CheckpointEntryIterator createCheckpointEntryIterator(URI checkpointUri, Set<CheckpointEntryIterator.EntryType> entryTypes, Optional<MetadataEntry> metadataEntry) {
    Path path = new Path(checkpointUri);
    long fileSize = new File(checkpointUri).length();
    return new CheckpointEntryIterator(path, SESSION, fileSize, checkpointSchemaManager, TESTING_TYPE_MANAGER, entryTypes, metadataEntry, hdfsEnvironment, new FileFormatDataSourceStats(), new ParquetReaderConfig().toParquetReaderOptions(), true);
}

Example 4 with FileFormatDataSourceStats

use of io.trino.plugin.hive.FileFormatDataSourceStats in project trino by trinodb.

the class TestDeltaLakeFileStatistics method testParseParquetStatistics.

@Test
public void testParseParquetStatistics() throws Exception {
    File statsFile = new File(getClass().getResource("/databricks/pruning/parquet_struct_statistics/_delta_log/00000000000000000010.checkpoint.parquet").getFile());
    Path checkpointPath = new Path(statsFile.toURI());
    TypeManager typeManager = TESTING_TYPE_MANAGER;
    CheckpointSchemaManager checkpointSchemaManager = new CheckpointSchemaManager(typeManager);
    HdfsConfig hdfsConfig = new HdfsConfig();
    HdfsConfiguration hdfsConfiguration = new HiveHdfsConfiguration(new HdfsConfigurationInitializer(hdfsConfig), ImmutableSet.of());
    HdfsEnvironment hdfsEnvironment = new HdfsEnvironment(hdfsConfiguration, hdfsConfig, new NoHdfsAuthentication());
    FileSystem fs = hdfsEnvironment.getFileSystem(new HdfsEnvironment.HdfsContext(SESSION), checkpointPath);
    CheckpointEntryIterator metadataEntryIterator = new CheckpointEntryIterator(checkpointPath, SESSION, fs.getFileStatus(checkpointPath).getLen(), checkpointSchemaManager, typeManager, ImmutableSet.of(METADATA), Optional.empty(), hdfsEnvironment, new FileFormatDataSourceStats(), new ParquetReaderConfig().toParquetReaderOptions(), true);
    MetadataEntry metadataEntry = getOnlyElement(metadataEntryIterator).getMetaData();
    CheckpointEntryIterator checkpointEntryIterator = new CheckpointEntryIterator(checkpointPath, SESSION, fs.getFileStatus(checkpointPath).getLen(), checkpointSchemaManager, typeManager, ImmutableSet.of(CheckpointEntryIterator.EntryType.ADD), Optional.of(metadataEntry), hdfsEnvironment, new FileFormatDataSourceStats(), new ParquetReaderConfig().toParquetReaderOptions(), true);
    DeltaLakeTransactionLogEntry matchingAddFileEntry = null;
    while (checkpointEntryIterator.hasNext()) {
        DeltaLakeTransactionLogEntry entry = checkpointEntryIterator.next();
        if (entry.getAdd() != null && entry.getAdd().getPath().contains("part-00000-17951bea-0d04-43c1-979c-ea1fac19b382-c000.snappy.parquet")) {
            assertNull(matchingAddFileEntry);
            matchingAddFileEntry = entry;
        }
    }
    assertNotNull(matchingAddFileEntry);
    assertThat(matchingAddFileEntry.getAdd().getStats()).isPresent();
    testStatisticsValues(matchingAddFileEntry.getAdd().getStats().get());
}

Example 5 with FileFormatDataSourceStats

use of io.trino.plugin.hive.FileFormatDataSourceStats in project trino by trinodb.

the class TestDeltaLakeMetastoreStatistics method setupMetastore.

@BeforeClass
public void setupMetastore() {
    TestingConnectorContext context = new TestingConnectorContext();
    TypeManager typeManager = context.getTypeManager();
    CheckpointSchemaManager checkpointSchemaManager = new CheckpointSchemaManager(typeManager);
    HdfsConfig hdfsConfig = new HdfsConfig();
    HdfsConfiguration hdfsConfiguration = new HiveHdfsConfiguration(new HdfsConfigurationInitializer(hdfsConfig), ImmutableSet.of());
    HdfsEnvironment hdfsEnvironment = new HdfsEnvironment(hdfsConfiguration, hdfsConfig, new NoHdfsAuthentication());
    FileFormatDataSourceStats fileFormatDataSourceStats = new FileFormatDataSourceStats();
    TransactionLogAccess transactionLogAccess = new TransactionLogAccess(typeManager, checkpointSchemaManager, new DeltaLakeConfig(), fileFormatDataSourceStats, hdfsEnvironment, new ParquetReaderConfig(), new DeltaLakeConfig());
    File tmpDir = Files.createTempDir();
    File metastoreDir = new File(tmpDir, "metastore");
    hiveMetastore = new FileHiveMetastore(new NodeVersion("test_version"), hdfsEnvironment, new MetastoreConfig(), new FileHiveMetastoreConfig().setCatalogDirectory(metastoreDir.toURI().toString()).setMetastoreUser("test"));
    hiveMetastore.createDatabase(new Database("db_name", Optional.empty(), Optional.of("test"), Optional.of(PrincipalType.USER), Optional.empty(), ImmutableMap.of()));
    CachingDeltaLakeStatisticsAccess statistics = new CachingDeltaLakeStatisticsAccess(new MetaDirStatisticsAccess(hdfsEnvironment, new JsonCodecFactory().jsonCodec(DeltaLakeStatistics.class)));
    deltaLakeMetastore = new HiveMetastoreBackedDeltaLakeMetastore(hiveMetastore, transactionLogAccess, typeManager, statistics);
}