Examples with ConnectorSession io.trino.spi.connector.ConnectorSession used on opensource projects

Example 31 with ConnectorSession

use of io.trino.spi.connector.ConnectorSession in project trino by trinodb.

the class HiveMetadata method finishOptimize.

private void finishOptimize(ConnectorSession session, ConnectorTableExecuteHandle tableExecuteHandle, Collection<Slice> fragments, List<Object> splitSourceInfo) {
    // TODO lots of that is copied from finishInsert; rafactoring opportunity
    HiveTableExecuteHandle handle = (HiveTableExecuteHandle) tableExecuteHandle;
    checkArgument(handle.getWriteDeclarationId().isPresent(), "no write declaration id present in tableExecuteHandle");
    List<PartitionUpdate> partitionUpdates = fragments.stream().map(Slice::getBytes).map(partitionUpdateCodec::fromJson).collect(toImmutableList());
    HiveStorageFormat tableStorageFormat = handle.getTableStorageFormat();
    partitionUpdates = PartitionUpdate.mergePartitionUpdates(partitionUpdates);
    Table table = metastore.getTable(handle.getSchemaName(), handle.getTableName()).orElseThrow(() -> new TableNotFoundException(handle.getSchemaTableName()));
    if (!table.getStorage().getStorageFormat().getInputFormat().equals(tableStorageFormat.getInputFormat()) && isRespectTableFormat(session)) {
        throw new TrinoException(HIVE_CONCURRENT_MODIFICATION_DETECTED, "Table format changed during optimize");
    }
    // Support for bucketed tables disabled mostly so we do not need to think about grouped execution in an initial version. Possibly no change apart from testing required.
    verify(handle.getBucketProperty().isEmpty(), "bucketed table not supported");
    for (PartitionUpdate partitionUpdate : partitionUpdates) {
        // sanity check
        verify(partitionUpdate.getUpdateMode() == APPEND, "Expected partionUpdate mode to be APPEND but got %s", partitionUpdate.getUpdateMode());
        if (partitionUpdate.getName().isEmpty()) {
            // operating on an unpartitioned table
            if (!table.getStorage().getStorageFormat().getInputFormat().equals(handle.getPartitionStorageFormat().getInputFormat()) && isRespectTableFormat(session)) {
                throw new TrinoException(HIVE_CONCURRENT_MODIFICATION_DETECTED, "Table format changed during optimize");
            }
            metastore.finishInsertIntoExistingTable(session, handle.getSchemaName(), handle.getTableName(), partitionUpdate.getWritePath(), partitionUpdate.getFileNames(), PartitionStatistics.empty(), handle.isRetriesEnabled());
        } else {
            // operating on a partition
            List<String> partitionValues = toPartitionValues(partitionUpdate.getName());
            metastore.finishInsertIntoExistingPartition(session, handle.getSchemaName(), handle.getTableName(), partitionValues, partitionUpdate.getWritePath(), partitionUpdate.getFileNames(), PartitionStatistics.empty(), handle.isRetriesEnabled());
        }
    }
    // get filesystem
    FileSystem fs;
    try {
        fs = hdfsEnvironment.getFileSystem(new HdfsContext(session), new Path(table.getStorage().getLocation()));
    } catch (IOException e) {
        throw new TrinoException(HIVE_FILESYSTEM_ERROR, e);
    }
    // path to be deleted
    Set<Path> scannedPaths = splitSourceInfo.stream().map(file -> new Path((String) file)).collect(toImmutableSet());
    // track remaining files to be delted for error reporting
    Set<Path> remainingFilesToDelete = new HashSet<>(scannedPaths);
    // delete loop
    boolean someDeleted = false;
    Optional<Path> firstScannedPath = Optional.empty();
    try {
        for (Path scannedPath : scannedPaths) {
            if (firstScannedPath.isEmpty()) {
                firstScannedPath = Optional.of(scannedPath);
            }
            retry().run("delete " + scannedPath, () -> fs.delete(scannedPath, false));
            someDeleted = true;
            remainingFilesToDelete.remove(scannedPath);
        }
    } catch (Exception e) {
        if (!someDeleted && (firstScannedPath.isEmpty() || exists(fs, firstScannedPath.get()))) {
            // fs.delete above could throw exception but file was actually deleted.
            throw new TrinoException(HIVE_FILESYSTEM_ERROR, "Error while deleting original files", e);
        }
        // If we already deleted some original files we disable rollback routine so written files are not deleted.
        // The reported exception message and log entry lists files which need to be cleaned up by user manually.
        // Until table is cleaned up there will duplicate rows present.
        metastore.dropDeclaredIntentionToWrite(handle.getWriteDeclarationId().get());
        String errorMessage = "Error while deleting data files in FINISH phase of OPTIMIZE for table " + table.getTableName() + "; remaining files need to be deleted manually:  " + remainingFilesToDelete;
        log.error(e, "%s", errorMessage);
        throw new TrinoException(HIVE_FILESYSTEM_ERROR, errorMessage, e);
    }
}

Example 32 with ConnectorSession

use of io.trino.spi.connector.ConnectorSession in project trino by trinodb.

the class HiveMetadata method getTableProperties.

@Override
public ConnectorTableProperties getTableProperties(ConnectorSession session, ConnectorTableHandle table) {
    HiveTableHandle hiveTable = (HiveTableHandle) table;
    List<ColumnHandle> partitionColumns = ImmutableList.copyOf(hiveTable.getPartitionColumns());
    TupleDomain<ColumnHandle> predicate = TupleDomain.all();
    Optional<DiscretePredicates> discretePredicates = Optional.empty();
    // So computation of  predicate and discretePredicates are not valid.
    if (hiveTable.getPartitionNames().isEmpty()) {
        Optional<List<HivePartition>> partitions = hiveTable.getPartitions().or(() -> {
            // We load the partitions to compute the predicates enforced by the table.
            // Note that the computation is not persisted in the table handle, so can be redone many times
            // TODO: https://github.com/trinodb/trino/issues/10980.
            HivePartitionResult partitionResult = partitionManager.getPartitions(metastore, table, new Constraint(hiveTable.getEnforcedConstraint()));
            if (partitionManager.canPartitionsBeLoaded(partitionResult)) {
                return Optional.of(partitionManager.getPartitionsAsList(partitionResult));
            }
            return Optional.empty();
        });
        if (partitions.isPresent()) {
            List<HivePartition> hivePartitions = partitions.orElseThrow();
            // Since the partitions are fully loaded now, we need to compute
            predicate = createPredicate(partitionColumns, hivePartitions);
            // this check allows us to ensure that table is partitioned
            if (!partitionColumns.isEmpty()) {
                // Do not create tuple domains for every partition at the same time!
                // There can be a huge number of partitions so use an iterable so
                // all domains do not need to be in memory at the same time.
                Iterable<TupleDomain<ColumnHandle>> partitionDomains = Iterables.transform(hivePartitions, hivePartition -> TupleDomain.fromFixedValues(hivePartition.getKeys()));
                discretePredicates = Optional.of(new DiscretePredicates(partitionColumns, partitionDomains));
            }
        }
    }
    Optional<ConnectorTablePartitioning> tablePartitioning = Optional.empty();
    List<LocalProperty<ColumnHandle>> sortingProperties = ImmutableList.of();
    if (hiveTable.getBucketHandle().isPresent()) {
        if (isPropagateTableScanSortingProperties(session) && !hiveTable.getBucketHandle().get().getSortedBy().isEmpty()) {
            // Populating SortingProperty guarantees to the engine that it is reading pre-sorted input.
            // We detect compatibility between table and partition level sorted_by properties
            // and fail the query if there is a mismatch in HiveSplitManager#getPartitionMetadata.
            // This can lead to incorrect results if a sorted_by property is defined over unsorted files.
            Map<String, ColumnHandle> columnHandles = getColumnHandles(session, table);
            sortingProperties = hiveTable.getBucketHandle().get().getSortedBy().stream().map(sortingColumn -> new SortingProperty<>(columnHandles.get(sortingColumn.getColumnName()), sortingColumn.getOrder().getSortOrder())).collect(toImmutableList());
        }
        if (isBucketExecutionEnabled(session)) {
            tablePartitioning = hiveTable.getBucketHandle().map(bucketing -> new ConnectorTablePartitioning(new HivePartitioningHandle(bucketing.getBucketingVersion(), bucketing.getReadBucketCount(), bucketing.getColumns().stream().map(HiveColumnHandle::getHiveType).collect(toImmutableList()), OptionalInt.empty(), false), bucketing.getColumns().stream().map(ColumnHandle.class::cast).collect(toImmutableList())));
        }
    }
    return new ConnectorTableProperties(predicate, tablePartitioning, Optional.empty(), discretePredicates, sortingProperties);
}

Example 33 with ConnectorSession

use of io.trino.spi.connector.ConnectorSession in project trino by trinodb.

the class HiveMetadata method beginInsert.

@Override
public HiveInsertTableHandle beginInsert(ConnectorSession session, ConnectorTableHandle tableHandle, List<ColumnHandle> columns, RetryMode retryMode) {
    SchemaTableName tableName = ((HiveTableHandle) tableHandle).getSchemaTableName();
    Table table = metastore.getTable(tableName.getSchemaName(), tableName.getTableName()).orElseThrow(() -> new TableNotFoundException(tableName));
    checkTableIsWritable(table, writesToNonManagedTablesEnabled);
    for (Column column : table.getDataColumns()) {
        if (!isWritableType(column.getType())) {
            throw new TrinoException(NOT_SUPPORTED, format("Inserting into Hive table %s with column type %s not supported", tableName, column.getType()));
        }
    }
    boolean isTransactional = isTransactionalTable(table.getParameters());
    if (isTransactional && retryMode != NO_RETRIES) {
        throw new TrinoException(NOT_SUPPORTED, "Inserting into Hive transactional tables is not supported with query retries enabled");
    }
    if (isTransactional && !autoCommit) {
        throw new TrinoException(NOT_SUPPORTED, "Inserting into Hive transactional tables is not supported in explicit transactions (use autocommit mode)");
    }
    if (isSparkBucketedTable(table)) {
        throw new TrinoException(NOT_SUPPORTED, "Inserting into Spark bucketed tables is not supported");
    }
    List<HiveColumnHandle> handles = hiveColumnHandles(table, typeManager, getTimestampPrecision(session)).stream().filter(columnHandle -> !columnHandle.isHidden()).collect(toImmutableList());
    HiveStorageFormat tableStorageFormat = extractHiveStorageFormat(table);
    Optional.ofNullable(table.getParameters().get(SKIP_HEADER_COUNT_KEY)).map(Integer::parseInt).ifPresent(headerSkipCount -> {
        if (headerSkipCount > 1) {
            throw new TrinoException(NOT_SUPPORTED, format("Inserting into Hive table with value of %s property greater than 1 is not supported", SKIP_HEADER_COUNT_KEY));
        }
    });
    if (table.getParameters().containsKey(SKIP_FOOTER_COUNT_KEY)) {
        throw new TrinoException(NOT_SUPPORTED, format("Inserting into Hive table with %s property not supported", SKIP_FOOTER_COUNT_KEY));
    }
    LocationHandle locationHandle = locationService.forExistingTable(metastore, session, table);
    AcidTransaction transaction = isTransactional ? metastore.beginInsert(session, table) : NO_ACID_TRANSACTION;
    HiveInsertTableHandle result = new HiveInsertTableHandle(tableName.getSchemaName(), tableName.getTableName(), handles, metastore.generatePageSinkMetadata(tableName), locationHandle, table.getStorage().getBucketProperty(), tableStorageFormat, isRespectTableFormat(session) ? tableStorageFormat : getHiveStorageFormat(session), transaction, retryMode != NO_RETRIES);
    WriteInfo writeInfo = locationService.getQueryWriteInfo(locationHandle);
    if (getInsertExistingPartitionsBehavior(session) == InsertExistingPartitionsBehavior.OVERWRITE && writeInfo.getWriteMode() == DIRECT_TO_TARGET_EXISTING_DIRECTORY) {
        if (isTransactional) {
            throw new TrinoException(NOT_SUPPORTED, "Overwriting existing partition in transactional tables doesn't support DIRECT_TO_TARGET_EXISTING_DIRECTORY write mode");
        }
        // Partition overwrite operation is nonatomic thus can't and shouldn't be used in non autocommit context.
        if (!autoCommit) {
            throw new TrinoException(NOT_SUPPORTED, "Overwriting existing partition in non auto commit context doesn't support DIRECT_TO_TARGET_EXISTING_DIRECTORY write mode");
        }
    }
    metastore.declareIntentionToWrite(session, writeInfo.getWriteMode(), writeInfo.getWritePath(), tableName);
    return result;
}

Example 34 with ConnectorSession

use of io.trino.spi.connector.ConnectorSession in project trino by trinodb.

the class HiveMetadata method getNewTableLayout.

@Override
public Optional<ConnectorTableLayout> getNewTableLayout(ConnectorSession session, ConnectorTableMetadata tableMetadata) {
    validateTimestampColumns(tableMetadata.getColumns(), getTimestampPrecision(session));
    validatePartitionColumns(tableMetadata);
    validateBucketColumns(tableMetadata);
    validateColumns(tableMetadata);
    Optional<HiveBucketProperty> bucketProperty = getBucketProperty(tableMetadata.getProperties());
    List<String> partitionedBy = getPartitionedBy(tableMetadata.getProperties());
    if (bucketProperty.isEmpty()) {
        // return preferred layout which is partitioned by partition columns
        if (partitionedBy.isEmpty()) {
            return Optional.empty();
        }
        return Optional.of(new ConnectorTableLayout(partitionedBy));
    }
    if (!bucketProperty.get().getSortedBy().isEmpty() && !isSortedWritingEnabled(session)) {
        throw new TrinoException(NOT_SUPPORTED, "Writing to bucketed sorted Hive tables is disabled");
    }
    List<String> bucketedBy = bucketProperty.get().getBucketedBy();
    Map<String, HiveType> hiveTypeMap = tableMetadata.getColumns().stream().collect(toMap(ColumnMetadata::getName, column -> toHiveType(column.getType())));
    return Optional.of(new ConnectorTableLayout(new HivePartitioningHandle(bucketProperty.get().getBucketingVersion(), bucketProperty.get().getBucketCount(), bucketedBy.stream().map(hiveTypeMap::get).collect(toImmutableList()), OptionalInt.of(bucketProperty.get().getBucketCount()), !partitionedBy.isEmpty() && isParallelPartitionedBucketedWrites(session)), ImmutableList.<String>builder().addAll(bucketedBy).addAll(partitionedBy).build()));
}

Example 35 with ConnectorSession

use of io.trino.spi.connector.ConnectorSession in project trino by trinodb.

the class HiveMetadata method finishInsert.

@Override
public Optional<ConnectorOutputMetadata> finishInsert(ConnectorSession session, ConnectorInsertTableHandle insertHandle, Collection<Slice> fragments, Collection<ComputedStatistics> computedStatistics) {
    HiveInsertTableHandle handle = (HiveInsertTableHandle) insertHandle;
    List<PartitionUpdate> partitionUpdates = fragments.stream().map(Slice::getBytes).map(partitionUpdateCodec::fromJson).collect(toImmutableList());
    HiveStorageFormat tableStorageFormat = handle.getTableStorageFormat();
    partitionUpdates = PartitionUpdate.mergePartitionUpdates(partitionUpdates);
    Table table = metastore.getTable(handle.getSchemaName(), handle.getTableName()).orElseThrow(() -> new TableNotFoundException(handle.getSchemaTableName()));
    if (!table.getStorage().getStorageFormat().getInputFormat().equals(tableStorageFormat.getInputFormat()) && isRespectTableFormat(session)) {
        throw new TrinoException(HIVE_CONCURRENT_MODIFICATION_DETECTED, "Table format changed during insert");
    }
    if (handle.getBucketProperty().isPresent() && isCreateEmptyBucketFiles(session)) {
        List<PartitionUpdate> partitionUpdatesForMissingBuckets = computePartitionUpdatesForMissingBuckets(session, handle, table, false, partitionUpdates);
        // replace partitionUpdates before creating the empty files so that those files will be cleaned up if we end up rollback
        partitionUpdates = PartitionUpdate.mergePartitionUpdates(concat(partitionUpdates, partitionUpdatesForMissingBuckets));
        for (PartitionUpdate partitionUpdate : partitionUpdatesForMissingBuckets) {
            Optional<Partition> partition = table.getPartitionColumns().isEmpty() ? Optional.empty() : Optional.of(buildPartitionObject(session, table, partitionUpdate));
            if (handle.isTransactional() && partition.isPresent()) {
                PartitionStatistics statistics = PartitionStatistics.builder().setBasicStatistics(partitionUpdate.getStatistics()).build();
                metastore.addPartition(session, handle.getSchemaName(), handle.getTableName(), partition.get(), partitionUpdate.getWritePath(), Optional.of(partitionUpdate.getFileNames()), statistics, handle.isRetriesEnabled());
            }
            createEmptyFiles(session, partitionUpdate.getWritePath(), table, partition, partitionUpdate.getFileNames());
        }
    }
    List<String> partitionedBy = table.getPartitionColumns().stream().map(Column::getName).collect(toImmutableList());
    Map<String, Type> columnTypes = handle.getInputColumns().stream().collect(toImmutableMap(HiveColumnHandle::getName, column -> column.getHiveType().getType(typeManager)));
    Map<List<String>, ComputedStatistics> partitionComputedStatistics = createComputedStatisticsToPartitionMap(computedStatistics, partitionedBy, columnTypes);
    for (PartitionUpdate partitionUpdate : partitionUpdates) {
        if (partitionUpdate.getName().isEmpty()) {
            // insert into unpartitioned table
            if (!table.getStorage().getStorageFormat().getInputFormat().equals(handle.getPartitionStorageFormat().getInputFormat()) && isRespectTableFormat(session)) {
                throw new TrinoException(HIVE_CONCURRENT_MODIFICATION_DETECTED, "Table format changed during insert");
            }
            PartitionStatistics partitionStatistics = createPartitionStatistics(partitionUpdate.getStatistics(), columnTypes, getColumnStatistics(partitionComputedStatistics, ImmutableList.of()));
            if (partitionUpdate.getUpdateMode() == OVERWRITE) {
                // get privileges from existing table
                PrincipalPrivileges principalPrivileges = fromHivePrivilegeInfos(metastore.listTablePrivileges(handle.getSchemaName(), handle.getTableName(), Optional.empty()));
                // first drop it
                metastore.dropTable(session, handle.getSchemaName(), handle.getTableName());
                // create the table with the new location
                metastore.createTable(session, table, principalPrivileges, Optional.of(partitionUpdate.getWritePath()), Optional.of(partitionUpdate.getFileNames()), false, partitionStatistics, handle.isRetriesEnabled());
            } else if (partitionUpdate.getUpdateMode() == NEW || partitionUpdate.getUpdateMode() == APPEND) {
                // insert into unpartitioned table
                metastore.finishInsertIntoExistingTable(session, handle.getSchemaName(), handle.getTableName(), partitionUpdate.getWritePath(), partitionUpdate.getFileNames(), partitionStatistics, handle.isRetriesEnabled());
            } else {
                throw new IllegalArgumentException("Unsupported update mode: " + partitionUpdate.getUpdateMode());
            }
        } else if (partitionUpdate.getUpdateMode() == APPEND) {
            // insert into existing partition
            List<String> partitionValues = toPartitionValues(partitionUpdate.getName());
            PartitionStatistics partitionStatistics = createPartitionStatistics(partitionUpdate.getStatistics(), columnTypes, getColumnStatistics(partitionComputedStatistics, partitionValues));
            metastore.finishInsertIntoExistingPartition(session, handle.getSchemaName(), handle.getTableName(), partitionValues, partitionUpdate.getWritePath(), partitionUpdate.getFileNames(), partitionStatistics, handle.isRetriesEnabled());
        } else if (partitionUpdate.getUpdateMode() == NEW || partitionUpdate.getUpdateMode() == OVERWRITE) {
            // insert into new partition or overwrite existing partition
            Partition partition = buildPartitionObject(session, table, partitionUpdate);
            if (!partition.getStorage().getStorageFormat().getInputFormat().equals(handle.getPartitionStorageFormat().getInputFormat()) && isRespectTableFormat(session)) {
                throw new TrinoException(HIVE_CONCURRENT_MODIFICATION_DETECTED, "Partition format changed during insert");
            }
            PartitionStatistics partitionStatistics = createPartitionStatistics(partitionUpdate.getStatistics(), columnTypes, getColumnStatistics(partitionComputedStatistics, partition.getValues()));
            if (partitionUpdate.getUpdateMode() == OVERWRITE) {
                if (handle.getLocationHandle().getWriteMode() == DIRECT_TO_TARGET_EXISTING_DIRECTORY) {
                    removeNonCurrentQueryFiles(session, partitionUpdate.getTargetPath());
                    if (handle.isRetriesEnabled()) {
                        HdfsContext hdfsContext = new HdfsContext(session);
                        cleanExtraOutputFiles(hdfsEnvironment, hdfsContext, session.getQueryId(), partitionUpdate.getTargetPath(), ImmutableSet.copyOf(partitionUpdate.getFileNames()));
                    }
                } else {
                    metastore.dropPartition(session, handle.getSchemaName(), handle.getTableName(), partition.getValues(), true);
                    metastore.addPartition(session, handle.getSchemaName(), handle.getTableName(), partition, partitionUpdate.getWritePath(), Optional.of(partitionUpdate.getFileNames()), partitionStatistics, handle.isRetriesEnabled());
                }
            } else {
                metastore.addPartition(session, handle.getSchemaName(), handle.getTableName(), partition, partitionUpdate.getWritePath(), Optional.of(partitionUpdate.getFileNames()), partitionStatistics, handle.isRetriesEnabled());
            }
        } else {
            throw new IllegalArgumentException(format("Unsupported update mode: %s", partitionUpdate.getUpdateMode()));
        }
    }
    if (isFullAcidTable(table.getParameters())) {
        HdfsContext context = new HdfsContext(session);
        for (PartitionUpdate update : partitionUpdates) {
            long writeId = handle.getTransaction().getWriteId();
            Path deltaDirectory = new Path(format("%s/%s/%s", table.getStorage().getLocation(), update.getName(), deltaSubdir(writeId, writeId, 0)));
            createOrcAcidVersionFile(context, deltaDirectory);
        }
    }
    return Optional.of(new HiveWrittenPartitions(partitionUpdates.stream().map(PartitionUpdate::getName).collect(toImmutableList())));
}