Examples with HiveBucketFilter io.trino.plugin.hive.util.HiveBucketing.HiveBucketFilter used on opensource projects

Example 1 with HiveBucketFilter

use of io.trino.plugin.hive.util.HiveBucketing.HiveBucketFilter in project trino by trinodb.

the class HivePartitionManager method getPartitions.

public HivePartitionResult getPartitions(SemiTransactionalHiveMetastore metastore, ConnectorTableHandle tableHandle, Constraint constraint) {
    HiveTableHandle hiveTableHandle = (HiveTableHandle) tableHandle;
    TupleDomain<ColumnHandle> effectivePredicate = constraint.getSummary().intersect(hiveTableHandle.getEnforcedConstraint());
    SchemaTableName tableName = hiveTableHandle.getSchemaTableName();
    Optional<HiveBucketHandle> hiveBucketHandle = hiveTableHandle.getBucketHandle();
    List<HiveColumnHandle> partitionColumns = hiveTableHandle.getPartitionColumns();
    if (effectivePredicate.isNone()) {
        return new HivePartitionResult(partitionColumns, Optional.empty(), ImmutableList.of(), TupleDomain.none(), TupleDomain.none(), hiveBucketHandle, Optional.empty());
    }
    Optional<HiveBucketFilter> bucketFilter = getHiveBucketFilter(hiveTableHandle, effectivePredicate);
    TupleDomain<HiveColumnHandle> compactEffectivePredicate = effectivePredicate.transformKeys(HiveColumnHandle.class::cast).simplify(domainCompactionThreshold);
    if (partitionColumns.isEmpty()) {
        return new HivePartitionResult(partitionColumns, Optional.empty(), ImmutableList.of(new HivePartition(tableName)), effectivePredicate, compactEffectivePredicate, hiveBucketHandle, bucketFilter);
    }
    List<Type> partitionTypes = partitionColumns.stream().map(HiveColumnHandle::getType).collect(toList());
    Optional<List<String>> partitionNames = Optional.empty();
    Iterable<HivePartition> partitionsIterable;
    Predicate<Map<ColumnHandle, NullableValue>> predicate = constraint.predicate().orElse(value -> true);
    if (hiveTableHandle.getPartitions().isPresent()) {
        partitionsIterable = hiveTableHandle.getPartitions().get().stream().filter(partition -> partitionMatches(partitionColumns, effectivePredicate, predicate, partition)).collect(toImmutableList());
    } else {
        List<String> partitionNamesList = hiveTableHandle.getPartitionNames().orElseGet(() -> getFilteredPartitionNames(metastore, tableName, partitionColumns, compactEffectivePredicate));
        partitionsIterable = () -> partitionNamesList.stream().map(partitionName -> parseValuesAndFilterPartition(tableName, partitionName, partitionColumns, partitionTypes, effectivePredicate, predicate)).filter(Optional::isPresent).map(Optional::get).iterator();
        partitionNames = Optional.of(partitionNamesList);
    }
    return new HivePartitionResult(partitionColumns, partitionNames, partitionsIterable, effectivePredicate, compactEffectivePredicate, hiveBucketHandle, bucketFilter);
}

Example 2 with HiveBucketFilter

use of io.trino.plugin.hive.util.HiveBucketing.HiveBucketFilter in project trino by trinodb.

the class HiveSplitManager method getSplits.

@Override
public ConnectorSplitSource getSplits(ConnectorTransactionHandle transaction, ConnectorSession session, ConnectorTableHandle tableHandle, SplitSchedulingStrategy splitSchedulingStrategy, DynamicFilter dynamicFilter) {
    HiveTableHandle hiveTable = (HiveTableHandle) tableHandle;
    SchemaTableName tableName = hiveTable.getSchemaTableName();
    // get table metadata
    SemiTransactionalHiveMetastore metastore = transactionManager.get(transaction, session.getIdentity()).getMetastore();
    Table table = metastore.getTable(tableName.getSchemaName(), tableName.getTableName()).orElseThrow(() -> new TableNotFoundException(tableName));
    // verify table is not marked as non-readable
    String tableNotReadable = table.getParameters().get(OBJECT_NOT_READABLE);
    if (!isNullOrEmpty(tableNotReadable)) {
        throw new HiveNotReadableException(tableName, Optional.empty(), tableNotReadable);
    }
    // get partitions
    List<HivePartition> partitions = partitionManager.getOrLoadPartitions(metastore, hiveTable);
    // short circuit if we don't have any partitions
    if (partitions.isEmpty()) {
        if (hiveTable.isRecordScannedFiles()) {
            return new FixedSplitSource(ImmutableList.of(), ImmutableList.of());
        }
        return new FixedSplitSource(ImmutableList.of());
    }
    // get buckets from first partition (arbitrary)
    Optional<HiveBucketFilter> bucketFilter = hiveTable.getBucketFilter();
    // validate bucket bucketed execution
    Optional<HiveBucketHandle> bucketHandle = hiveTable.getBucketHandle();
    if ((splitSchedulingStrategy == GROUPED_SCHEDULING) && bucketHandle.isEmpty()) {
        throw new TrinoException(GENERIC_INTERNAL_ERROR, "SchedulingPolicy is bucketed, but BucketHandle is not present");
    }
    // sort partitions
    partitions = Ordering.natural().onResultOf(HivePartition::getPartitionId).reverse().sortedCopy(partitions);
    Iterable<HivePartitionMetadata> hivePartitions = getPartitionMetadata(session, metastore, table, tableName, partitions, bucketHandle.map(HiveBucketHandle::toTableBucketProperty));
    // Only one thread per partition is usable when a table is not transactional
    int concurrency = isTransactionalTable(table.getParameters()) ? splitLoaderConcurrency : min(splitLoaderConcurrency, partitions.size());
    HiveSplitLoader hiveSplitLoader = new BackgroundHiveSplitLoader(table, hiveTable.getTransaction(), hivePartitions, hiveTable.getCompactEffectivePredicate(), dynamicFilter, getDynamicFilteringWaitTimeout(session), typeManager, createBucketSplitInfo(bucketHandle, bucketFilter), session, hdfsEnvironment, namenodeStats, directoryLister, executor, concurrency, recursiveDfsWalkerEnabled, !hiveTable.getPartitionColumns().isEmpty() && isIgnoreAbsentPartitions(session), isOptimizeSymlinkListing(session), metastore.getValidWriteIds(session, hiveTable).map(validTxnWriteIdList -> validTxnWriteIdList.getTableValidWriteIdList(table.getDatabaseName() + "." + table.getTableName())), hiveTable.getMaxScannedFileSize());
    HiveSplitSource splitSource;
    switch(splitSchedulingStrategy) {
        case UNGROUPED_SCHEDULING:
            splitSource = HiveSplitSource.allAtOnce(session, table.getDatabaseName(), table.getTableName(), maxInitialSplits, maxOutstandingSplits, maxOutstandingSplitsSize, maxSplitsPerSecond, hiveSplitLoader, executor, highMemorySplitSourceCounter, hiveTable.isRecordScannedFiles());
            break;
        case GROUPED_SCHEDULING:
            splitSource = HiveSplitSource.bucketed(session, table.getDatabaseName(), table.getTableName(), maxInitialSplits, maxOutstandingSplits, maxOutstandingSplitsSize, maxSplitsPerSecond, hiveSplitLoader, executor, highMemorySplitSourceCounter, hiveTable.isRecordScannedFiles());
            break;
        default:
            throw new IllegalArgumentException("Unknown splitSchedulingStrategy: " + splitSchedulingStrategy);
    }
    hiveSplitLoader.start(splitSource);
    return splitSource;
}