Examples with ComputedStatistics com.facebook.presto.spi.statistics.ComputedStatistics used on opensource projects

Example 1 with ComputedStatistics

use of com.facebook.presto.spi.statistics.ComputedStatistics in project presto by prestodb.

the class HiveMetadata method finishInsertInternal.

private Optional<ConnectorOutputMetadata> finishInsertInternal(ConnectorSession session, ConnectorInsertTableHandle insertHandle, Collection<Slice> fragments, Collection<ComputedStatistics> computedStatistics) {
    HiveInsertTableHandle handle = (HiveInsertTableHandle) insertHandle;
    List<PartitionUpdate> partitionUpdates = getPartitionUpdates(session, fragments);
    HiveStorageFormat tableStorageFormat = handle.getTableStorageFormat();
    partitionUpdates = PartitionUpdate.mergePartitionUpdates(partitionUpdates);
    MetastoreContext metastoreContext = getMetastoreContext(session);
    Table table = metastore.getTable(metastoreContext, handle.getSchemaName(), handle.getTableName()).orElseThrow(() -> new TableNotFoundException(handle.getSchemaTableName()));
    if (!table.getStorage().getStorageFormat().getInputFormat().equals(tableStorageFormat.getInputFormat()) && isRespectTableFormat(session)) {
        throw new PrestoException(HIVE_CONCURRENT_MODIFICATION_DETECTED, "Table format changed during insert");
    }
    if (handle.getBucketProperty().isPresent() && isCreateEmptyBucketFiles(session)) {
        List<PartitionUpdate> partitionUpdatesForMissingBuckets = computePartitionUpdatesForMissingBuckets(session, handle, table, partitionUpdates);
        // replace partitionUpdates before creating the zero-row files so that those files will be cleaned up if we end up rollback
        partitionUpdates = PartitionUpdate.mergePartitionUpdates(concat(partitionUpdates, partitionUpdatesForMissingBuckets));
        HdfsContext hdfsContext = new HdfsContext(session, table.getDatabaseName(), table.getTableName(), table.getStorage().getLocation(), false);
        for (PartitionUpdate partitionUpdate : partitionUpdatesForMissingBuckets) {
            Optional<Partition> partition = table.getPartitionColumns().isEmpty() ? Optional.empty() : Optional.of(partitionObjectBuilder.buildPartitionObject(session, table, partitionUpdate, prestoVersion, handle.getEncryptionInformation().map(encryptionInfo -> encryptionInfo.getDwrfEncryptionMetadata().map(DwrfEncryptionMetadata::getExtraMetadata).orElseGet(ImmutableMap::of)).orElseGet(ImmutableMap::of)));
            zeroRowFileCreator.createFiles(session, hdfsContext, partitionUpdate.getWritePath(), getTargetFileNames(partitionUpdate.getFileWriteInfos()), getStorageFormat(partition, table), handle.getCompressionCodec(), getSchema(partition, table));
        }
    }
    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);
    Set<String> existingPartitions = getExistingPartitionNames(session.getIdentity(), metastoreContext, handle.getSchemaName(), handle.getTableName(), partitionUpdates);
    for (PartitionUpdate partitionUpdate : partitionUpdates) {
        if (partitionUpdate.getName().isEmpty()) {
            if (handle.getEncryptionInformation().isPresent()) {
                throw new PrestoException(HIVE_UNSUPPORTED_ENCRYPTION_OPERATION, "Inserting into an existing table with encryption enabled is not supported yet");
            }
            // insert into unpartitioned table
            PartitionStatistics partitionStatistics = createPartitionStatistics(session, partitionUpdate.getStatistics(), columnTypes, getColumnStatistics(partitionComputedStatistics, ImmutableList.of()));
            metastore.finishInsertIntoExistingTable(session, handle.getSchemaName(), handle.getTableName(), partitionUpdate.getWritePath(), getTargetFileNames(partitionUpdate.getFileWriteInfos()), partitionStatistics);
        } else if (partitionUpdate.getUpdateMode() == APPEND) {
            if (handle.getEncryptionInformation().isPresent()) {
                throw new PrestoException(HIVE_UNSUPPORTED_ENCRYPTION_OPERATION, "Inserting into an existing partition with encryption enabled is not supported yet");
            }
            // insert into existing partition
            List<String> partitionValues = toPartitionValues(partitionUpdate.getName());
            PartitionStatistics partitionStatistics = createPartitionStatistics(session, partitionUpdate.getStatistics(), columnTypes, getColumnStatistics(partitionComputedStatistics, partitionValues));
            metastore.finishInsertIntoExistingPartition(session, handle.getSchemaName(), handle.getTableName(), handle.getLocationHandle().getTargetPath().toString(), partitionValues, partitionUpdate.getWritePath(), getTargetFileNames(partitionUpdate.getFileWriteInfos()), partitionStatistics);
        } else if (partitionUpdate.getUpdateMode() == NEW || partitionUpdate.getUpdateMode() == OVERWRITE) {
            Map<String, String> extraPartitionMetadata = handle.getEncryptionInformation().map(encryptionInfo -> encryptionInfo.getDwrfEncryptionMetadata().map(DwrfEncryptionMetadata::getExtraMetadata).orElseGet(ImmutableMap::of)).orElseGet(ImmutableMap::of);
            if (isPreferManifestsToListFiles(session) && isFileRenamingEnabled(session)) {
                // Store list of file names and sizes in partition metadata when prefer_manifests_to_list_files and file_renaming_enabled are set to true
                extraPartitionMetadata = updatePartitionMetadataWithFileNamesAndSizes(partitionUpdate, extraPartitionMetadata);
            }
            // Track the manifest blob size
            getManifestSizeInBytes(session, partitionUpdate, extraPartitionMetadata).ifPresent(hivePartitionStats::addManifestSizeInBytes);
            // insert into new partition or overwrite existing partition
            Partition partition = partitionObjectBuilder.buildPartitionObject(session, table, partitionUpdate, prestoVersion, extraPartitionMetadata);
            if (!partition.getStorage().getStorageFormat().getInputFormat().equals(handle.getPartitionStorageFormat().getInputFormat()) && isRespectTableFormat(session)) {
                throw new PrestoException(HIVE_CONCURRENT_MODIFICATION_DETECTED, "Partition format changed during insert");
            }
            if (existingPartitions.contains(partitionUpdate.getName())) {
                if (partitionUpdate.getUpdateMode() == OVERWRITE) {
                    metastore.dropPartition(session, handle.getSchemaName(), handle.getTableName(), handle.getLocationHandle().getTargetPath().toString(), partition.getValues());
                } else {
                    throw new PrestoException(HIVE_PARTITION_READ_ONLY, "Cannot insert into an existing partition of Hive table: " + partitionUpdate.getName());
                }
            }
            PartitionStatistics partitionStatistics = createPartitionStatistics(session, partitionUpdate.getStatistics(), columnTypes, getColumnStatistics(partitionComputedStatistics, partition.getValues()));
            metastore.addPartition(session, handle.getSchemaName(), handle.getTableName(), table.getStorage().getLocation(), false, partition, partitionUpdate.getWritePath(), partitionStatistics);
        } else {
            throw new IllegalArgumentException(format("Unsupported update mode: %s", partitionUpdate.getUpdateMode()));
        }
    }
    return Optional.of(new HiveWrittenPartitions(partitionUpdates.stream().map(PartitionUpdate::getName).map(name -> name.isEmpty() ? UNPARTITIONED_ID : name).collect(toList())));
}

Example 2 with ComputedStatistics

use of com.facebook.presto.spi.statistics.ComputedStatistics in project presto by prestodb.

the class HiveMetadata method finishCreateTable.

@Override
public Optional<ConnectorOutputMetadata> finishCreateTable(ConnectorSession session, ConnectorOutputTableHandle tableHandle, Collection<Slice> fragments, Collection<ComputedStatistics> computedStatistics) {
    HiveOutputTableHandle handle = (HiveOutputTableHandle) tableHandle;
    List<PartitionUpdate> partitionUpdates = getPartitionUpdates(session, fragments);
    Map<String, String> tableEncryptionParameters = ImmutableMap.of();
    Map<String, String> partitionEncryptionParameters = ImmutableMap.of();
    if (handle.getEncryptionInformation().isPresent()) {
        EncryptionInformation encryptionInformation = handle.getEncryptionInformation().get();
        if (encryptionInformation.getDwrfEncryptionMetadata().isPresent()) {
            if (handle.getPartitionedBy().isEmpty()) {
                tableEncryptionParameters = ImmutableMap.copyOf(encryptionInformation.getDwrfEncryptionMetadata().get().getExtraMetadata());
            } else {
                partitionEncryptionParameters = ImmutableMap.copyOf(encryptionInformation.getDwrfEncryptionMetadata().get().getExtraMetadata());
            }
        }
    }
    WriteInfo writeInfo = locationService.getQueryWriteInfo(handle.getLocationHandle());
    MetastoreContext metastoreContext = getMetastoreContext(session);
    Table table = buildTableObject(session.getQueryId(), handle.getSchemaName(), handle.getTableName(), handle.getTableOwner(), handle.getInputColumns(), handle.getTableStorageFormat(), handle.getPartitionedBy(), handle.getBucketProperty(), handle.getPreferredOrderingColumns(), ImmutableMap.<String, String>builder().putAll(handle.getAdditionalTableParameters()).putAll(tableEncryptionParameters).build(), writeInfo.getTargetPath(), MANAGED_TABLE, prestoVersion, metastoreContext);
    PrincipalPrivileges principalPrivileges = buildInitialPrivilegeSet(handle.getTableOwner());
    partitionUpdates = PartitionUpdate.mergePartitionUpdates(partitionUpdates);
    if (handle.getBucketProperty().isPresent() && isCreateEmptyBucketFiles(session)) {
        List<PartitionUpdate> partitionUpdatesForMissingBuckets = computePartitionUpdatesForMissingBuckets(session, handle, table, partitionUpdates);
        // replace partitionUpdates before creating the zero-row files so that those files will be cleaned up if we end up rollback
        partitionUpdates = PartitionUpdate.mergePartitionUpdates(concat(partitionUpdates, partitionUpdatesForMissingBuckets));
        HdfsContext hdfsContext = new HdfsContext(session, table.getDatabaseName(), table.getTableName(), table.getStorage().getLocation(), true);
        for (PartitionUpdate partitionUpdate : partitionUpdatesForMissingBuckets) {
            Optional<Partition> partition = table.getPartitionColumns().isEmpty() ? Optional.empty() : Optional.of(partitionObjectBuilder.buildPartitionObject(session, table, partitionUpdate, prestoVersion, partitionEncryptionParameters));
            zeroRowFileCreator.createFiles(session, hdfsContext, partitionUpdate.getWritePath(), getTargetFileNames(partitionUpdate.getFileWriteInfos()), getStorageFormat(partition, table), handle.getCompressionCodec(), getSchema(partition, table));
        }
    }
    Map<String, Type> columnTypes = handle.getInputColumns().stream().collect(toImmutableMap(HiveColumnHandle::getName, column -> column.getHiveType().getType(typeManager)));
    Map<List<String>, ComputedStatistics> partitionComputedStatistics = createComputedStatisticsToPartitionMap(computedStatistics, handle.getPartitionedBy(), columnTypes);
    PartitionStatistics tableStatistics;
    if (table.getPartitionColumns().isEmpty()) {
        HiveBasicStatistics basicStatistics = partitionUpdates.stream().map(PartitionUpdate::getStatistics).reduce((first, second) -> reduce(first, second, ADD)).orElse(createZeroStatistics());
        tableStatistics = createPartitionStatistics(session, basicStatistics, columnTypes, getColumnStatistics(partitionComputedStatistics, ImmutableList.of()));
    } else {
        tableStatistics = new PartitionStatistics(createEmptyStatistics(), ImmutableMap.of());
    }
    metastore.createTable(session, table, principalPrivileges, Optional.of(writeInfo.getWritePath()), false, tableStatistics);
    if (handle.getPartitionedBy().isEmpty()) {
        return Optional.of(new HiveWrittenPartitions(ImmutableList.of(UNPARTITIONED_ID)));
    }
    if (isRespectTableFormat(session)) {
        verify(handle.getPartitionStorageFormat() == handle.getTableStorageFormat());
    }
    for (PartitionUpdate update : partitionUpdates) {
        Map<String, String> partitionParameters = partitionEncryptionParameters;
        if (isPreferManifestsToListFiles(session) && isFileRenamingEnabled(session)) {
            // Store list of file names and sizes in partition metadata when prefer_manifests_to_list_files and file_renaming_enabled are set to true
            partitionParameters = updatePartitionMetadataWithFileNamesAndSizes(update, partitionParameters);
        }
        Partition partition = partitionObjectBuilder.buildPartitionObject(session, table, update, prestoVersion, partitionParameters);
        PartitionStatistics partitionStatistics = createPartitionStatistics(session, update.getStatistics(), columnTypes, getColumnStatistics(partitionComputedStatistics, partition.getValues()));
        metastore.addPartition(session, handle.getSchemaName(), handle.getTableName(), table.getStorage().getLocation(), true, partitionObjectBuilder.buildPartitionObject(session, table, update, prestoVersion, partitionParameters), update.getWritePath(), partitionStatistics);
    }
    return Optional.of(new HiveWrittenPartitions(partitionUpdates.stream().map(PartitionUpdate::getName).collect(toList())));
}

Example 3 with ComputedStatistics

use of com.facebook.presto.spi.statistics.ComputedStatistics in project presto by prestodb.

the class IcebergAbstractMetadata method finishInsert.

@Override
public Optional<ConnectorOutputMetadata> finishInsert(ConnectorSession session, ConnectorInsertTableHandle insertHandle, Collection<Slice> fragments, Collection<ComputedStatistics> computedStatistics) {
    if (fragments.isEmpty()) {
        transaction.commitTransaction();
        return Optional.empty();
    }
    IcebergWritableTableHandle table = (IcebergWritableTableHandle) insertHandle;
    org.apache.iceberg.Table icebergTable = transaction.table();
    List<CommitTaskData> commitTasks = fragments.stream().map(slice -> commitTaskCodec.fromJson(slice.getBytes())).collect(toImmutableList());
    Type[] partitionColumnTypes = icebergTable.spec().fields().stream().map(field -> field.transform().getResultType(icebergTable.schema().findType(field.sourceId()))).toArray(Type[]::new);
    AppendFiles appendFiles = transaction.newFastAppend();
    for (CommitTaskData task : commitTasks) {
        DataFiles.Builder builder = DataFiles.builder(icebergTable.spec()).withPath(task.getPath()).withFileSizeInBytes(task.getFileSizeInBytes()).withFormat(table.getFileFormat()).withMetrics(task.getMetrics().metrics());
        if (!icebergTable.spec().fields().isEmpty()) {
            String partitionDataJson = task.getPartitionDataJson().orElseThrow(() -> new VerifyException("No partition data for partitioned table"));
            builder.withPartition(PartitionData.fromJson(partitionDataJson, partitionColumnTypes));
        }
        appendFiles.appendFile(builder.build());
    }
    appendFiles.commit();
    transaction.commitTransaction();
    return Optional.of(new HiveWrittenPartitions(commitTasks.stream().map(CommitTaskData::getPath).collect(toImmutableList())));
}

Example 4 with ComputedStatistics

use of com.facebook.presto.spi.statistics.ComputedStatistics in project presto by prestodb.

the class HiveMetadata method finishStatisticsCollection.

@Override
public void finishStatisticsCollection(ConnectorSession session, ConnectorTableHandle tableHandle, Collection<ComputedStatistics> computedStatistics) {
    HiveTableHandle handle = (HiveTableHandle) tableHandle;
    SchemaTableName tableName = handle.getSchemaTableName();
    MetastoreContext metastoreContext = getMetastoreContext(session);
    Table table = metastore.getTable(metastoreContext, tableName.getSchemaName(), tableName.getTableName()).orElseThrow(() -> new TableNotFoundException(handle.getSchemaTableName()));
    List<Column> partitionColumns = table.getPartitionColumns();
    List<String> partitionColumnNames = partitionColumns.stream().map(Column::getName).collect(toImmutableList());
    List<HiveColumnHandle> hiveColumnHandles = hiveColumnHandles(table);
    Map<String, Type> columnTypes = hiveColumnHandles.stream().filter(columnHandle -> !columnHandle.isHidden()).collect(toImmutableMap(HiveColumnHandle::getName, column -> column.getHiveType().getType(typeManager)));
    Map<List<String>, ComputedStatistics> computedStatisticsMap = createComputedStatisticsToPartitionMap(computedStatistics, partitionColumnNames, columnTypes);
    if (partitionColumns.isEmpty()) {
        // commit analyze to unpartitioned table
        metastore.setTableStatistics(metastoreContext, table, createPartitionStatistics(session, columnTypes, computedStatisticsMap.get(ImmutableList.<String>of())));
    } else {
        List<List<String>> partitionValuesList;
        if (handle.getAnalyzePartitionValues().isPresent()) {
            partitionValuesList = handle.getAnalyzePartitionValues().get();
        } else {
            partitionValuesList = metastore.getPartitionNames(metastoreContext, handle.getSchemaName(), handle.getTableName()).orElseThrow(() -> new TableNotFoundException(((HiveTableHandle) tableHandle).getSchemaTableName())).stream().map(MetastoreUtil::toPartitionValues).collect(toImmutableList());
        }
        ImmutableMap.Builder<List<String>, PartitionStatistics> partitionStatistics = ImmutableMap.builder();
        Map<String, Set<ColumnStatisticType>> columnStatisticTypes = hiveColumnHandles.stream().filter(columnHandle -> !partitionColumnNames.contains(columnHandle.getName())).filter(column -> !column.isHidden()).collect(toImmutableMap(HiveColumnHandle::getName, column -> ImmutableSet.copyOf(metastore.getSupportedColumnStatistics(metastoreContext, typeManager.getType(column.getTypeSignature())))));
        Supplier<PartitionStatistics> emptyPartitionStatistics = Suppliers.memoize(() -> createEmptyPartitionStatistics(columnTypes, columnStatisticTypes));
        int usedComputedStatistics = 0;
        for (List<String> partitionValues : partitionValuesList) {
            ComputedStatistics collectedStatistics = computedStatisticsMap.get(partitionValues);
            if (collectedStatistics == null) {
                partitionStatistics.put(partitionValues, emptyPartitionStatistics.get());
            } else {
                usedComputedStatistics++;
                partitionStatistics.put(partitionValues, createPartitionStatistics(session, columnTypes, collectedStatistics));
            }
        }
        verify(usedComputedStatistics == computedStatistics.size(), "All computed statistics must be used");
        metastore.setPartitionStatistics(metastoreContext, table, partitionStatistics.build());
    }
}

Example 5 with ComputedStatistics

use of com.facebook.presto.spi.statistics.ComputedStatistics in project presto by prestodb.

the class StatisticsWriterOperator method getOutput.

@Override
public Page getOutput() {
    if (state != State.FINISHING) {
        return null;
    }
    state = State.FINISHED;
    Collection<ComputedStatistics> computedStatistics = computedStatisticsBuilder.build();
    statisticsWriter.writeStatistics(computedStatistics);
    // output page will only be constructed once,
    // so a new PageBuilder is constructed (instead of using PageBuilder.reset)
    PageBuilder page = new PageBuilder(1, TYPES);
    page.declarePosition();
    BlockBuilder rowsBuilder = page.getBlockBuilder(0);
    if (rowCountEnabled) {
        BIGINT.writeLong(rowsBuilder, getRowCount(computedStatistics));
    } else {
        rowsBuilder.appendNull();
    }
    return page.build();
}