Examples with HdfsContext io.trino.plugin.hive.HdfsEnvironment.HdfsContext used on opensource projects

Example 1 with HdfsContext

use of io.trino.plugin.hive.HdfsEnvironment.HdfsContext 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 2 with HdfsContext

use of io.trino.plugin.hive.HdfsEnvironment.HdfsContext in project trino by trinodb.

the class HiveMetadata method finishDelete.

@Override
public void finishDelete(ConnectorSession session, ConnectorTableHandle tableHandle, Collection<Slice> fragments) {
    HiveTableHandle handle = (HiveTableHandle) tableHandle;
    checkArgument(handle.isAcidDelete(), "handle should be a delete handle, but is %s", handle);
    requireNonNull(fragments, "fragments is null");
    SchemaTableName tableName = handle.getSchemaTableName();
    Table table = metastore.getTable(tableName.getSchemaName(), tableName.getTableName()).orElseThrow(() -> new TableNotFoundException(tableName));
    ensureTableSupportsDelete(table);
    List<PartitionAndStatementId> partitionAndStatementIds = fragments.stream().map(Slice::getBytes).map(PartitionAndStatementId.CODEC::fromJson).collect(toImmutableList());
    HdfsContext context = new HdfsContext(session);
    for (PartitionAndStatementId ps : partitionAndStatementIds) {
        createOrcAcidVersionFile(context, new Path(ps.getDeleteDeltaDirectory()));
    }
    LocationHandle locationHandle = locationService.forExistingTable(metastore, session, table);
    WriteInfo writeInfo = locationService.getQueryWriteInfo(locationHandle);
    metastore.finishRowLevelDelete(session, table.getDatabaseName(), table.getTableName(), writeInfo.getWritePath(), partitionAndStatementIds);
}

Example 3 with HdfsContext

use of io.trino.plugin.hive.HdfsEnvironment.HdfsContext in project trino by trinodb.

the class HiveMetadata method finishUpdate.

@Override
public void finishUpdate(ConnectorSession session, ConnectorTableHandle tableHandle, Collection<Slice> fragments) {
    HiveTableHandle handle = (HiveTableHandle) tableHandle;
    checkArgument(handle.isAcidUpdate(), "handle should be a update handle, but is %s", handle);
    requireNonNull(fragments, "fragments is null");
    SchemaTableName tableName = handle.getSchemaTableName();
    Table table = metastore.getTable(tableName.getSchemaName(), tableName.getTableName()).orElseThrow(() -> new TableNotFoundException(tableName));
    List<PartitionAndStatementId> partitionAndStatementIds = fragments.stream().map(Slice::getBytes).map(PartitionAndStatementId.CODEC::fromJson).collect(toImmutableList());
    HdfsContext context = new HdfsContext(session);
    for (PartitionAndStatementId ps : partitionAndStatementIds) {
        createOrcAcidVersionFile(context, new Path(ps.getDeleteDeltaDirectory()));
    }
    LocationHandle locationHandle = locationService.forExistingTable(metastore, session, table);
    WriteInfo writeInfo = locationService.getQueryWriteInfo(locationHandle);
    metastore.finishUpdate(session, table.getDatabaseName(), table.getTableName(), writeInfo.getWritePath(), partitionAndStatementIds);
}

Example 4 with HdfsContext

use of io.trino.plugin.hive.HdfsEnvironment.HdfsContext in project trino by trinodb.

the class HiveMetadata method createSchema.

@Override
public void createSchema(ConnectorSession session, String schemaName, Map<String, Object> properties, TrinoPrincipal owner) {
    Optional<String> location = HiveSchemaProperties.getLocation(properties).map(locationUri -> {
        try {
            hdfsEnvironment.getFileSystem(new HdfsContext(session), new Path(locationUri));
        } catch (IOException e) {
            throw new TrinoException(INVALID_SCHEMA_PROPERTY, "Invalid location URI: " + locationUri, e);
        }
        return locationUri;
    });
    Database database = Database.builder().setDatabaseName(schemaName).setLocation(location).setOwnerType(accessControlMetadata.isUsingSystemSecurity() ? Optional.empty() : Optional.of(owner.getType())).setOwnerName(accessControlMetadata.isUsingSystemSecurity() ? Optional.empty() : Optional.of(owner.getName())).build();
    metastore.createDatabase(database);
}

Example 5 with HdfsContext

use of io.trino.plugin.hive.HdfsEnvironment.HdfsContext 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())));
}