Examples with OrcRecordReader io.prestosql.orc.OrcRecordReader used on opensource projects

Example 1 with OrcRecordReader

use of io.prestosql.orc.OrcRecordReader in project hetu-core by openlookeng.

the class OrcPageSourceFactory method createOrcPageSource.

public static OrcPageSource createOrcPageSource(HdfsEnvironment hdfsEnvironment, String sessionUser, Configuration configuration, Path path, long start, long length, long fileSize, List<HiveColumnHandle> columns, boolean useOrcColumnNames, boolean isFullAcid, TupleDomain<HiveColumnHandle> effectivePredicate, DateTimeZone legacyFileTimeZone, TypeManager typeManager, DataSize maxMergeDistance, DataSize maxBufferSize, DataSize streamBufferSize, DataSize tinyStripeThreshold, DataSize maxReadBlockSize, boolean lazyReadSmallRanges, boolean orcBloomFiltersEnabled, FileFormatDataSourceStats stats, Optional<DynamicFilterSupplier> dynamicFilters, Optional<DeleteDeltaLocations> deleteDeltaLocations, Optional<Long> startRowOffsetOfFile, Optional<List<IndexMetadata>> indexes, SplitMetadata splitMetadata, OrcCacheStore orcCacheStore, OrcCacheProperties orcCacheProperties, int domainCompactionThreshold, boolean pageMetadataEnabled, long dataSourceLastModifiedTime) {
    for (HiveColumnHandle column : columns) {
        checkArgument(column.getColumnType() == HiveColumnHandle.ColumnType.REGULAR || column.getHiveColumnIndex() == HiveColumnHandle.ROW_ID__COLUMN_INDEX, "column type must be regular: %s", column);
    }
    checkArgument(!effectivePredicate.isNone());
    OrcDataSource orcDataSource;
    try {
        // Always create a lazy Stream. HDFS stream opened only when required.
        FSDataInputStream inputStream = new FSDataInputStream(new LazyFSInputStream(() -> {
            FileSystem fileSystem = hdfsEnvironment.getFileSystem(sessionUser, path, configuration);
            return hdfsEnvironment.doAs(sessionUser, () -> fileSystem.open(path));
        }));
        orcDataSource = new HdfsOrcDataSource(new OrcDataSourceId(path.toString()), fileSize, maxMergeDistance, maxBufferSize, streamBufferSize, lazyReadSmallRanges, inputStream, stats, dataSourceLastModifiedTime);
    } 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);
    }
    AggregatedMemoryContext systemMemoryUsage = newSimpleAggregatedMemoryContext();
    try {
        OrcDataSource readerLocalDataSource = OrcReader.wrapWithCacheIfTiny(orcDataSource, tinyStripeThreshold);
        OrcFileTail fileTail;
        if (orcCacheProperties.isFileTailCacheEnabled()) {
            try {
                OrcDataSourceIdWithTimeStamp orcDataSourceIdWithTimeStamp = new OrcDataSourceIdWithTimeStamp(readerLocalDataSource.getId(), readerLocalDataSource.getLastModifiedTime());
                fileTail = orcCacheStore.getFileTailCache().get(new OrcFileTailCacheKey(orcDataSourceIdWithTimeStamp), () -> OrcPageSourceFactory.createFileTail(orcDataSource));
            } catch (UncheckedExecutionException | ExecutionException executionException) {
                handleCacheLoadException(executionException);
                log.debug(executionException.getCause(), "Error while caching the Orc file tail. Falling back to default flow");
                fileTail = OrcPageSourceFactory.createFileTail(orcDataSource);
            }
        } else {
            fileTail = OrcPageSourceFactory.createFileTail(orcDataSource);
        }
        OrcReader reader = new OrcReader(readerLocalDataSource, fileTail, maxMergeDistance, tinyStripeThreshold, maxReadBlockSize);
        List<OrcColumn> fileColumns = reader.getRootColumn().getNestedColumns();
        List<OrcColumn> fileReadColumns = isFullAcid ? new ArrayList<>(columns.size() + 5) : new ArrayList<>(columns.size());
        List<Type> fileReadTypes = isFullAcid ? new ArrayList<>(columns.size() + 5) : new ArrayList<>(columns.size());
        ImmutableList<String> acidColumnNames = null;
        List<ColumnAdaptation> columnAdaptations = new ArrayList<>(columns.size());
        // Only Hive ACID files will begin with bucket_
        boolean fileNameContainsBucket = path.getName().contains("bucket");
        if (isFullAcid && fileNameContainsBucket) {
            // Skip the acid schema check in case of non-ACID files
            acidColumnNames = ImmutableList.<String>builder().add(ACID_COLUMN_ORIGINAL_TRANSACTION, ACID_COLUMN_BUCKET, ACID_COLUMN_ROW_ID, ACID_COLUMN_CURRENT_TRANSACTION, ACID_COLUMN_OPERATION).build();
            verifyAcidSchema(reader, path);
            Map<String, OrcColumn> acidColumnsByName = uniqueIndex(fileColumns, orcColumn -> orcColumn.getColumnName().toLowerCase(ENGLISH));
            if (AcidUtils.isDeleteDelta(path.getParent())) {
                // Avoid reading column data from delete_delta files.
                // Call will come here in case of Minor VACUUM where all delete_delta files are merge together.
                fileColumns = ImmutableList.of();
            } else {
                fileColumns = ensureColumnNameConsistency(acidColumnsByName.get(ACID_COLUMN_ROW_STRUCT).getNestedColumns(), columns);
            }
            fileReadColumns.add(acidColumnsByName.get(ACID_COLUMN_ORIGINAL_TRANSACTION.toLowerCase(ENGLISH)));
            fileReadTypes.add(BIGINT);
            fileReadColumns.add(acidColumnsByName.get(ACID_COLUMN_BUCKET.toLowerCase(ENGLISH)));
            fileReadTypes.add(INTEGER);
            fileReadColumns.add(acidColumnsByName.get(ACID_COLUMN_ROW_ID.toLowerCase(ENGLISH)));
            fileReadTypes.add(BIGINT);
            fileReadColumns.add(acidColumnsByName.get(ACID_COLUMN_CURRENT_TRANSACTION.toLowerCase(ENGLISH)));
            fileReadTypes.add(BIGINT);
            fileReadColumns.add(acidColumnsByName.get(ACID_COLUMN_OPERATION.toLowerCase(ENGLISH)));
            fileReadTypes.add(INTEGER);
        }
        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));
        }
        TupleDomainOrcPredicateBuilder predicateBuilder = TupleDomainOrcPredicate.builder().setBloomFiltersEnabled(orcBloomFiltersEnabled);
        Map<HiveColumnHandle, Domain> effectivePredicateDomains = effectivePredicate.getDomains().orElseThrow(() -> new IllegalArgumentException("Effective predicate is none"));
        for (HiveColumnHandle column : columns) {
            OrcColumn orcColumn = null;
            if (useOrcColumnNames || isFullAcid) {
                orcColumn = fileColumnsByName.get(column.getName());
            } else if (column.getHiveColumnIndex() >= 0 && column.getHiveColumnIndex() < fileColumns.size()) {
                orcColumn = fileColumns.get(column.getHiveColumnIndex());
            }
            Type readType = typeManager.getType(column.getTypeSignature());
            if (orcColumn != null) {
                int sourceIndex = fileReadColumns.size();
                columnAdaptations.add(ColumnAdaptation.sourceColumn(sourceIndex));
                fileReadColumns.add(orcColumn);
                fileReadTypes.add(readType);
                Domain domain = effectivePredicateDomains.get(column);
                if (domain != null) {
                    predicateBuilder.addColumn(orcColumn.getColumnId(), domain);
                }
            } else if (isFullAcid && readType instanceof RowType && column.getName().equalsIgnoreCase(HiveColumnHandle.UPDATE_ROW_ID_COLUMN_NAME)) {
                HiveType hiveType = column.getHiveType();
                StructTypeInfo structTypeInfo = (StructTypeInfo) hiveType.getTypeInfo();
                ImmutableList.Builder<ColumnAdaptation> builder = new ImmutableList.Builder<>();
                ArrayList<String> fieldNames = structTypeInfo.getAllStructFieldNames();
                List<ColumnAdaptation> adaptations = fieldNames.stream().map(acidColumnNames::indexOf).map(c -> ColumnAdaptation.sourceColumn(c, false)).collect(Collectors.toList());
                columnAdaptations.add(ColumnAdaptation.structColumn(structTypeInfo, adaptations));
            } else {
                columnAdaptations.add(ColumnAdaptation.nullColumn(readType));
            }
        }
        Map<String, Domain> domains = effectivePredicate.getDomains().get().entrySet().stream().collect(toMap(e -> e.getKey().getName(), Map.Entry::getValue));
        OrcRecordReader recordReader = reader.createRecordReader(fileReadColumns, fileReadTypes, predicateBuilder.build(), start, length, legacyFileTimeZone, systemMemoryUsage, INITIAL_BATCH_SIZE, exception -> handleException(orcDataSource.getId(), exception), indexes, splitMetadata, domains, orcCacheStore, orcCacheProperties, pageMetadataEnabled);
        OrcDeletedRows deletedRows = new OrcDeletedRows(path.getName(), deleteDeltaLocations, new OrcDeleteDeltaPageSourceFactory(sessionUser, configuration, hdfsEnvironment, maxMergeDistance, maxBufferSize, streamBufferSize, maxReadBlockSize, tinyStripeThreshold, lazyReadSmallRanges, orcBloomFiltersEnabled, stats), sessionUser, configuration, hdfsEnvironment, startRowOffsetOfFile);
        boolean eagerload = false;
        if (indexes.isPresent()) {
            eagerload = indexes.get().stream().anyMatch(indexMetadata -> EAGER_LOAD_INDEX_ID.contains(indexMetadata.getIndex().getId()));
        }
        return new OrcPageSource(recordReader, columnAdaptations, orcDataSource, deletedRows, eagerload, systemMemoryUsage, stats);
    } 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 instanceof BlockMissingException) {
            throw new PrestoException(HIVE_MISSING_DATA, message, e);
        }
        throw new PrestoException(HIVE_CANNOT_OPEN_SPLIT, message, e);
    }
}

Example 2 with OrcRecordReader

use of io.prestosql.orc.OrcRecordReader in project boostkit-bigdata by kunpengcompute.

the class OrcPageSourceFactory method createOrcPageSource.

public static OrcPageSource createOrcPageSource(HdfsEnvironment hdfsEnvironment, String sessionUser, Configuration configuration, Path path, long start, long length, long fileSize, List<HiveColumnHandle> columns, boolean useOrcColumnNames, boolean isFullAcid, TupleDomain<HiveColumnHandle> effectivePredicate, DateTimeZone legacyFileTimeZone, TypeManager typeManager, DataSize maxMergeDistance, DataSize maxBufferSize, DataSize streamBufferSize, DataSize tinyStripeThreshold, DataSize maxReadBlockSize, boolean lazyReadSmallRanges, boolean orcBloomFiltersEnabled, FileFormatDataSourceStats stats, Optional<DynamicFilterSupplier> dynamicFilters, Optional<DeleteDeltaLocations> deleteDeltaLocations, Optional<Long> startRowOffsetOfFile, Optional<List<IndexMetadata>> indexes, SplitMetadata splitMetadata, OrcCacheStore orcCacheStore, OrcCacheProperties orcCacheProperties, int domainCompactionThreshold, boolean pageMetadataEnabled, long dataSourceLastModifiedTime) {
    for (HiveColumnHandle column : columns) {
        checkArgument(column.getColumnType() == HiveColumnHandle.ColumnType.REGULAR || column.getHiveColumnIndex() == HiveColumnHandle.ROW_ID__COLUMN_INDEX, "column type must be regular: %s", column);
    }
    checkArgument(!effectivePredicate.isNone());
    OrcDataSource orcDataSource;
    try {
        // Always create a lazy Stream. HDFS stream opened only when required.
        FSDataInputStream inputStream = new FSDataInputStream(new LazyFSInputStream(() -> {
            FileSystem fileSystem = hdfsEnvironment.getFileSystem(sessionUser, path, configuration);
            return hdfsEnvironment.doAs(sessionUser, () -> fileSystem.open(path));
        }));
        orcDataSource = new HdfsOrcDataSource(new OrcDataSourceId(path.toString()), fileSize, maxMergeDistance, maxBufferSize, streamBufferSize, lazyReadSmallRanges, inputStream, stats, dataSourceLastModifiedTime);
    } 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);
    }
    AggregatedMemoryContext systemMemoryUsage = newSimpleAggregatedMemoryContext();
    try {
        OrcDataSource readerLocalDataSource = OrcReader.wrapWithCacheIfTiny(orcDataSource, tinyStripeThreshold);
        OrcFileTail fileTail;
        if (orcCacheProperties.isFileTailCacheEnabled()) {
            try {
                OrcDataSourceIdWithTimeStamp orcDataSourceIdWithTimeStamp = new OrcDataSourceIdWithTimeStamp(readerLocalDataSource.getId(), readerLocalDataSource.getLastModifiedTime());
                fileTail = orcCacheStore.getFileTailCache().get(new OrcFileTailCacheKey(orcDataSourceIdWithTimeStamp), () -> OrcPageSourceFactory.createFileTail(orcDataSource));
            } catch (UncheckedExecutionException | ExecutionException executionException) {
                handleCacheLoadException(executionException);
                log.debug(executionException.getCause(), "Error while caching the Orc file tail. Falling back to default flow");
                fileTail = OrcPageSourceFactory.createFileTail(orcDataSource);
            }
        } else {
            fileTail = OrcPageSourceFactory.createFileTail(orcDataSource);
        }
        OrcReader reader = new OrcReader(readerLocalDataSource, fileTail, maxMergeDistance, tinyStripeThreshold, maxReadBlockSize);
        List<OrcColumn> fileColumns = reader.getRootColumn().getNestedColumns();
        List<OrcColumn> fileReadColumns = isFullAcid ? new ArrayList<>(columns.size() + 5) : new ArrayList<>(columns.size());
        List<Type> fileReadTypes = isFullAcid ? new ArrayList<>(columns.size() + 5) : new ArrayList<>(columns.size());
        ImmutableList<String> acidColumnNames = null;
        List<ColumnAdaptation> columnAdaptations = new ArrayList<>(columns.size());
        // Only Hive ACID files will begin with bucket_
        boolean fileNameContainsBucket = path.getName().contains("bucket");
        if (isFullAcid && fileNameContainsBucket) {
            // Skip the acid schema check in case of non-ACID files
            acidColumnNames = ImmutableList.<String>builder().add(ACID_COLUMN_ORIGINAL_TRANSACTION, ACID_COLUMN_BUCKET, ACID_COLUMN_ROW_ID, ACID_COLUMN_CURRENT_TRANSACTION, ACID_COLUMN_OPERATION).build();
            verifyAcidSchema(reader, path);
            Map<String, OrcColumn> acidColumnsByName = uniqueIndex(fileColumns, orcColumn -> orcColumn.getColumnName().toLowerCase(ENGLISH));
            if (AcidUtils.isDeleteDelta(path.getParent())) {
                // Avoid reading column data from delete_delta files.
                // Call will come here in case of Minor VACUUM where all delete_delta files are merge together.
                fileColumns = ImmutableList.of();
            } else {
                fileColumns = ensureColumnNameConsistency(acidColumnsByName.get(ACID_COLUMN_ROW_STRUCT).getNestedColumns(), columns);
            }
            fileReadColumns.add(acidColumnsByName.get(ACID_COLUMN_ORIGINAL_TRANSACTION.toLowerCase(ENGLISH)));
            fileReadTypes.add(BIGINT);
            fileReadColumns.add(acidColumnsByName.get(ACID_COLUMN_BUCKET.toLowerCase(ENGLISH)));
            fileReadTypes.add(INTEGER);
            fileReadColumns.add(acidColumnsByName.get(ACID_COLUMN_ROW_ID.toLowerCase(ENGLISH)));
            fileReadTypes.add(BIGINT);
            fileReadColumns.add(acidColumnsByName.get(ACID_COLUMN_CURRENT_TRANSACTION.toLowerCase(ENGLISH)));
            fileReadTypes.add(BIGINT);
            fileReadColumns.add(acidColumnsByName.get(ACID_COLUMN_OPERATION.toLowerCase(ENGLISH)));
            fileReadTypes.add(INTEGER);
        }
        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));
        }
        TupleDomainOrcPredicateBuilder predicateBuilder = TupleDomainOrcPredicate.builder().setBloomFiltersEnabled(orcBloomFiltersEnabled);
        Map<HiveColumnHandle, Domain> effectivePredicateDomains = effectivePredicate.getDomains().orElseThrow(() -> new IllegalArgumentException("Effective predicate is none"));
        for (HiveColumnHandle column : columns) {
            OrcColumn orcColumn = null;
            if (useOrcColumnNames || isFullAcid) {
                orcColumn = fileColumnsByName.get(column.getName());
            } else if (column.getHiveColumnIndex() >= 0 && column.getHiveColumnIndex() < fileColumns.size()) {
                orcColumn = fileColumns.get(column.getHiveColumnIndex());
            }
            Type readType = typeManager.getType(column.getTypeSignature());
            if (orcColumn != null) {
                int sourceIndex = fileReadColumns.size();
                columnAdaptations.add(ColumnAdaptation.sourceColumn(sourceIndex));
                fileReadColumns.add(orcColumn);
                fileReadTypes.add(readType);
                Domain domain = effectivePredicateDomains.get(column);
                if (domain != null) {
                    predicateBuilder.addColumn(orcColumn.getColumnId(), domain);
                }
            } else if (isFullAcid && readType instanceof RowType && column.getName().equalsIgnoreCase(HiveColumnHandle.UPDATE_ROW_ID_COLUMN_NAME)) {
                HiveType hiveType = column.getHiveType();
                StructTypeInfo structTypeInfo = (StructTypeInfo) hiveType.getTypeInfo();
                ImmutableList.Builder<ColumnAdaptation> builder = new ImmutableList.Builder<>();
                ArrayList<String> fieldNames = structTypeInfo.getAllStructFieldNames();
                List<ColumnAdaptation> adaptations = fieldNames.stream().map(acidColumnNames::indexOf).map(c -> ColumnAdaptation.sourceColumn(c, false)).collect(Collectors.toList());
                columnAdaptations.add(ColumnAdaptation.structColumn(structTypeInfo, adaptations));
            } else {
                columnAdaptations.add(ColumnAdaptation.nullColumn(readType));
            }
        }
        Map<String, Domain> domains = effectivePredicate.getDomains().get().entrySet().stream().collect(toMap(e -> e.getKey().getName(), Map.Entry::getValue));
        OrcRecordReader recordReader = reader.createRecordReader(fileReadColumns, fileReadTypes, predicateBuilder.build(), start, length, legacyFileTimeZone, systemMemoryUsage, INITIAL_BATCH_SIZE, exception -> handleException(orcDataSource.getId(), exception), indexes, splitMetadata, domains, orcCacheStore, orcCacheProperties, pageMetadataEnabled);
        OrcDeletedRows deletedRows = new OrcDeletedRows(path.getName(), deleteDeltaLocations, new OrcDeleteDeltaPageSourceFactory(sessionUser, configuration, hdfsEnvironment, maxMergeDistance, maxBufferSize, streamBufferSize, maxReadBlockSize, tinyStripeThreshold, lazyReadSmallRanges, orcBloomFiltersEnabled, stats), sessionUser, configuration, hdfsEnvironment, startRowOffsetOfFile);
        boolean eagerload = false;
        if (indexes.isPresent()) {
            eagerload = indexes.get().stream().anyMatch(indexMetadata -> EAGER_LOAD_INDEX_ID.contains(indexMetadata.getIndex().getId()));
        }
        return new OrcPageSource(recordReader, columnAdaptations, orcDataSource, deletedRows, eagerload, systemMemoryUsage, stats);
    } 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 instanceof BlockMissingException) {
            throw new PrestoException(HIVE_MISSING_DATA, message, e);
        }
        throw new PrestoException(HIVE_CANNOT_OPEN_SPLIT, message, e);
    }
}