Examples with OrcFileWriterFactory io.trino.plugin.hive.orc.OrcFileWriterFactory used on opensource projects

Example 1 with OrcFileWriterFactory

use of io.trino.plugin.hive.orc.OrcFileWriterFactory in project trino by trinodb.

the class TestHiveFileFormats method testOrcOptimizedWriter.

@Test(dataProvider = "validRowAndFileSizePadding")
public void testOrcOptimizedWriter(int rowCount, long fileSizePadding) throws Exception {
    HiveSessionProperties hiveSessionProperties = new HiveSessionProperties(new HiveConfig(), new OrcReaderConfig(), new OrcWriterConfig().setValidationPercentage(100.0), new ParquetReaderConfig(), new ParquetWriterConfig());
    ConnectorSession session = TestingConnectorSession.builder().setPropertyMetadata(hiveSessionProperties.getSessionProperties()).build();
    // A Trino page cannot contain a map with null keys, so a page based writer cannot write null keys
    List<TestColumn> testColumns = TEST_COLUMNS.stream().filter(TestHiveFileFormats::withoutNullMapKeyTests).collect(toList());
    assertThatFileFormat(ORC).withColumns(testColumns).withRowsCount(rowCount).withSession(session).withFileSizePadding(fileSizePadding).withFileWriterFactory(new OrcFileWriterFactory(HDFS_ENVIRONMENT, TESTING_TYPE_MANAGER, new NodeVersion("test"), STATS, new OrcWriterOptions())).isReadableByRecordCursor(createGenericHiveRecordCursorProvider(HDFS_ENVIRONMENT)).isReadableByPageSource(new OrcPageSourceFactory(new OrcReaderOptions(), HDFS_ENVIRONMENT, STATS, UTC));
}

Example 2 with OrcFileWriterFactory

use of io.trino.plugin.hive.orc.OrcFileWriterFactory in project trino by trinodb.

the class HiveWriterFactory method createWriter.

public HiveWriter createWriter(Page partitionColumns, int position, OptionalInt bucketNumber) {
    if (bucketCount.isPresent()) {
        checkArgument(bucketNumber.isPresent(), "Bucket not provided for bucketed table");
        checkArgument(bucketNumber.getAsInt() < bucketCount.getAsInt(), "Bucket number %s must be less than bucket count %s", bucketNumber, bucketCount);
    } else {
        checkArgument(bucketNumber.isEmpty(), "Bucket number provided by for table that is not bucketed");
    }
    List<String> partitionValues = createPartitionValues(partitionColumnTypes, partitionColumns, position);
    Optional<String> partitionName;
    if (!partitionColumnNames.isEmpty()) {
        partitionName = Optional.of(FileUtils.makePartName(partitionColumnNames, partitionValues));
    } else {
        partitionName = Optional.empty();
    }
    // attempt to get the existing partition (if this is an existing partitioned table)
    Optional<Partition> partition = Optional.empty();
    if (!partitionValues.isEmpty() && table != null) {
        partition = pageSinkMetadataProvider.getPartition(partitionValues);
    }
    UpdateMode updateMode;
    Properties schema;
    WriteInfo writeInfo;
    StorageFormat outputStorageFormat;
    if (partition.isEmpty()) {
        if (table == null) {
            // Write to: a new partition in a new partitioned table,
            // or a new unpartitioned table.
            updateMode = UpdateMode.NEW;
            schema = new Properties();
            schema.setProperty(IOConstants.COLUMNS, dataColumns.stream().map(DataColumn::getName).collect(joining(",")));
            schema.setProperty(IOConstants.COLUMNS_TYPES, dataColumns.stream().map(DataColumn::getHiveType).map(HiveType::getHiveTypeName).map(HiveTypeName::toString).collect(joining(":")));
            if (partitionName.isEmpty()) {
                // new unpartitioned table
                writeInfo = locationService.getTableWriteInfo(locationHandle, false);
            } else {
                // a new partition in a new partitioned table
                writeInfo = locationService.getPartitionWriteInfo(locationHandle, partition, partitionName.get());
                if (!writeInfo.getWriteMode().isWritePathSameAsTargetPath()) {
                    // verify that the target directory for the partition does not already exist
                    if (HiveWriteUtils.pathExists(new HdfsContext(session), hdfsEnvironment, writeInfo.getTargetPath())) {
                        throw new TrinoException(HIVE_PATH_ALREADY_EXISTS, format("Target directory for new partition '%s' of table '%s.%s' already exists: %s", partitionName, schemaName, tableName, writeInfo.getTargetPath()));
                    }
                }
            }
        } else {
            // or an existing unpartitioned table
            if (partitionName.isPresent()) {
                // a new partition in an existing partitioned table
                updateMode = UpdateMode.NEW;
                writeInfo = locationService.getPartitionWriteInfo(locationHandle, partition, partitionName.get());
            } else {
                switch(insertExistingPartitionsBehavior) {
                    case APPEND:
                        updateMode = UpdateMode.APPEND;
                        writeInfo = locationService.getTableWriteInfo(locationHandle, false);
                        break;
                    case OVERWRITE:
                        updateMode = UpdateMode.OVERWRITE;
                        writeInfo = locationService.getTableWriteInfo(locationHandle, true);
                        break;
                    case ERROR:
                        throw new TrinoException(HIVE_TABLE_READ_ONLY, "Unpartitioned Hive tables are immutable");
                    default:
                        throw new IllegalArgumentException("Unsupported insert existing table behavior: " + insertExistingPartitionsBehavior);
                }
            }
            schema = getHiveSchema(table);
        }
        if (partitionName.isPresent()) {
            // Write to a new partition
            outputStorageFormat = fromHiveStorageFormat(partitionStorageFormat);
        } else {
            // Write to a new/existing unpartitioned table
            outputStorageFormat = fromHiveStorageFormat(tableStorageFormat);
        }
    } else {
        switch(insertExistingPartitionsBehavior) {
            // Write to: an existing partition in an existing partitioned table
            case APPEND:
                // Append to an existing partition
                updateMode = UpdateMode.APPEND;
                // Check the column types in partition schema match the column types in table schema
                List<Column> tableColumns = table.getDataColumns();
                List<Column> existingPartitionColumns = partition.get().getColumns();
                for (int i = 0; i < min(existingPartitionColumns.size(), tableColumns.size()); i++) {
                    HiveType tableType = tableColumns.get(i).getType();
                    HiveType partitionType = existingPartitionColumns.get(i).getType();
                    if (!tableType.equals(partitionType)) {
                        throw new TrinoException(HIVE_PARTITION_SCHEMA_MISMATCH, format("" + "You are trying to write into an existing partition in a table. " + "The table schema has changed since the creation of the partition. " + "Inserting rows into such partition is not supported. " + "The column '%s' in table '%s' is declared as type '%s', " + "but partition '%s' declared column '%s' as type '%s'.", tableColumns.get(i).getName(), tableName, tableType, partitionName, existingPartitionColumns.get(i).getName(), partitionType));
                    }
                }
                HiveWriteUtils.checkPartitionIsWritable(partitionName.get(), partition.get());
                outputStorageFormat = partition.get().getStorage().getStorageFormat();
                schema = getHiveSchema(partition.get(), table);
                writeInfo = locationService.getPartitionWriteInfo(locationHandle, partition, partitionName.get());
                break;
            case OVERWRITE:
                // Overwrite an existing partition
                // 
                // The behavior of overwrite considered as if first dropping the partition and inserting a new partition, thus:
                // * No partition writable check is required.
                // * Table schema and storage format is used for the new partition (instead of existing partition schema and storage format).
                updateMode = UpdateMode.OVERWRITE;
                outputStorageFormat = fromHiveStorageFormat(partitionStorageFormat);
                schema = getHiveSchema(table);
                writeInfo = locationService.getPartitionWriteInfo(locationHandle, Optional.empty(), partitionName.get());
                break;
            case ERROR:
                throw new TrinoException(HIVE_PARTITION_READ_ONLY, "Cannot insert into an existing partition of Hive table: " + partitionName.get());
            default:
                throw new IllegalArgumentException(format("Unsupported insert existing partitions behavior: %s", insertExistingPartitionsBehavior));
        }
    }
    additionalTableParameters.forEach(schema::setProperty);
    validateSchema(partitionName, schema);
    int bucketToUse = bucketNumber.isEmpty() ? 0 : bucketNumber.getAsInt();
    Path path;
    String fileNameWithExtension;
    if (transaction.isAcidTransactionRunning()) {
        String subdir = computeAcidSubdir(transaction);
        Path subdirPath = new Path(writeInfo.getWritePath(), subdir);
        path = createHiveBucketPath(subdirPath, bucketToUse, table.getParameters());
        fileNameWithExtension = path.getName();
    } else {
        String fileName = computeFileName(bucketNumber);
        fileNameWithExtension = fileName + getFileExtension(conf, outputStorageFormat);
        path = new Path(writeInfo.getWritePath(), fileNameWithExtension);
    }
    boolean useAcidSchema = isCreateTransactionalTable || (table != null && isFullAcidTable(table.getParameters()));
    FileWriter hiveFileWriter = null;
    for (HiveFileWriterFactory fileWriterFactory : fileWriterFactories) {
        Optional<FileWriter> fileWriter = fileWriterFactory.createFileWriter(path, dataColumns.stream().map(DataColumn::getName).collect(toList()), outputStorageFormat, schema, conf, session, bucketNumber, transaction, useAcidSchema, WriterKind.INSERT);
        if (fileWriter.isPresent()) {
            hiveFileWriter = fileWriter.get();
            break;
        }
    }
    if (hiveFileWriter == null) {
        hiveFileWriter = new RecordFileWriter(path, dataColumns.stream().map(DataColumn::getName).collect(toList()), outputStorageFormat, schema, partitionStorageFormat.getEstimatedWriterMemoryUsage(), conf, typeManager, parquetTimeZone, session);
    }
    String writerImplementation = hiveFileWriter.getClass().getName();
    Consumer<HiveWriter> onCommit = hiveWriter -> {
        Optional<Long> size;
        try {
            size = Optional.of(hiveWriter.getWrittenBytes());
        } catch (RuntimeException e) {
            // Do not fail the query if file system is not available
            size = Optional.empty();
        }
        eventClient.post(new WriteCompletedEvent(session.getQueryId(), path.toString(), schemaName, tableName, partitionName.orElse(null), outputStorageFormat.getOutputFormat(), writerImplementation, nodeManager.getCurrentNode().getVersion(), nodeManager.getCurrentNode().getHost(), session.getIdentity().getPrincipal().map(Principal::getName).orElse(null), nodeManager.getEnvironment(), sessionProperties, size.orElse(null), hiveWriter.getRowCount()));
    };
    if (!sortedBy.isEmpty()) {
        FileSystem fileSystem;
        Path tempFilePath;
        if (sortedWritingTempStagingPathEnabled) {
            String tempPrefix = sortedWritingTempStagingPath.replace("${USER}", new HdfsContext(session).getIdentity().getUser());
            tempFilePath = new Path(tempPrefix, ".tmp-sort." + path.getParent().getName() + "." + path.getName());
        } else {
            tempFilePath = new Path(path.getParent(), ".tmp-sort." + path.getName());
        }
        try {
            Configuration configuration = new Configuration(conf);
            // Explicitly set the default FS to local file system to avoid getting HDFS when sortedWritingTempStagingPath specifies no scheme
            configuration.set(FS_DEFAULT_NAME_KEY, "file:///");
            fileSystem = hdfsEnvironment.getFileSystem(session.getIdentity(), tempFilePath, configuration);
        } catch (IOException e) {
            throw new TrinoException(HIVE_WRITER_OPEN_ERROR, e);
        }
        List<Type> types = dataColumns.stream().map(column -> column.getHiveType().getType(typeManager, getTimestampPrecision(session))).collect(toImmutableList());
        Map<String, Integer> columnIndexes = new HashMap<>();
        for (int i = 0; i < dataColumns.size(); i++) {
            columnIndexes.put(dataColumns.get(i).getName(), i);
        }
        List<Integer> sortFields = new ArrayList<>();
        List<SortOrder> sortOrders = new ArrayList<>();
        for (SortingColumn column : sortedBy) {
            Integer index = columnIndexes.get(column.getColumnName());
            if (index == null) {
                throw new TrinoException(HIVE_INVALID_METADATA, format("Sorting column '%s' does exist in table '%s.%s'", column.getColumnName(), schemaName, tableName));
            }
            sortFields.add(index);
            sortOrders.add(column.getOrder().getSortOrder());
        }
        hiveFileWriter = new SortingFileWriter(fileSystem, tempFilePath, hiveFileWriter, sortBufferSize, maxOpenSortFiles, types, sortFields, sortOrders, pageSorter, typeManager.getTypeOperators(), OrcFileWriterFactory::createOrcDataSink);
    }
    return new HiveWriter(hiveFileWriter, partitionName, updateMode, fileNameWithExtension, writeInfo.getWritePath().toString(), writeInfo.getTargetPath().toString(), onCommit, hiveWriterStats);
}