Examples with GROUPED_SCHEDULING io.prestosql.spi.connector.ConnectorSplitManager.SplitSchedulingStrategy.GROUPED_SCHEDULING used on opensource projects

Example 1 with GROUPED_SCHEDULING

use of io.prestosql.spi.connector.ConnectorSplitManager.SplitSchedulingStrategy.GROUPED_SCHEDULING in project hetu-core by openlookeng.

the class HiveSplitManager method getSplits.

@Override
public ConnectorSplitSource getSplits(ConnectorTransactionHandle transaction, ConnectorSession session, ConnectorTableHandle tableHandle, SplitSchedulingStrategy splitSchedulingStrategy, Supplier<List<Set<DynamicFilter>>> dynamicFilterSupplier, Optional<QueryType> queryType, Map<String, Object> queryInfo, Set<TupleDomain<ColumnMetadata>> userDefinedCachePredicates, boolean partOfReuse) {
    HiveTableHandle hiveTable = (HiveTableHandle) tableHandle;
    SchemaTableName tableName = hiveTable.getSchemaTableName();
    // get table metadata
    SemiTransactionalHiveMetastore metastore = metastoreProvider.apply((HiveTransactionHandle) transaction);
    Table table = metastore.getTable(new HiveIdentity(session), tableName.getSchemaName(), tableName.getTableName()).orElseThrow(() -> new TableNotFoundException(tableName));
    if (table.getStorage().getStorageFormat().getInputFormat().contains("carbon")) {
        throw new PrestoException(NOT_SUPPORTED, "Hive connector can't read carbondata tables");
    }
    // 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(session, metastore, new HiveIdentity(session), hiveTable);
    // short circuit if we don't have any partitions
    if (partitions.isEmpty()) {
        return new FixedSplitSource(ImmutableList.of());
    }
    // get buckets from first partition (arbitrary)
    Optional<HiveBucketing.HiveBucketFilter> bucketFilter = hiveTable.getBucketFilter();
    // validate bucket bucketed execution
    Optional<HiveBucketHandle> bucketHandle = hiveTable.getBucketHandle();
    if ((splitSchedulingStrategy == GROUPED_SCHEDULING) && !bucketHandle.isPresent()) {
        throw new PrestoException(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));
    HiveSplitLoader hiveSplitLoader = new BackgroundHiveSplitLoader(table, hivePartitions, hiveTable.getCompactEffectivePredicate(), BackgroundHiveSplitLoader.BucketSplitInfo.createBucketSplitInfo(bucketHandle, bucketFilter), session, hdfsEnvironment, namenodeStats, directoryLister, executor, splitLoaderConcurrency, recursiveDfsWalkerEnabled, metastore.getValidWriteIds(session, hiveTable, queryType.map(t -> t == QueryType.VACUUM).orElse(false)).map(validTxnWriteIdList -> validTxnWriteIdList.getTableValidWriteIdList(table.getDatabaseName() + "." + table.getTableName())), dynamicFilterSupplier, queryType, queryInfo, typeManager);
    HiveSplitSource splitSource;
    HiveStorageFormat hiveStorageFormat = HiveMetadata.extractHiveStorageFormat(table);
    switch(splitSchedulingStrategy) {
        case UNGROUPED_SCHEDULING:
            splitSource = HiveSplitSource.allAtOnce(session, table.getDatabaseName(), table.getTableName(), // For reuse, we should make sure to have same split size all time for a table.
            partOfReuse ? 0 : maxInitialSplits, maxOutstandingSplits, maxOutstandingSplitsSize, maxSplitsPerSecond, hiveSplitLoader, executor, new CounterStat(), dynamicFilterSupplier, userDefinedCachePredicates, typeManager, hiveConfig, hiveStorageFormat);
            break;
        case GROUPED_SCHEDULING:
            splitSource = HiveSplitSource.bucketed(session, table.getDatabaseName(), table.getTableName(), // For reuse, we should make sure to have same split size all time for a table.
            partOfReuse ? 0 : maxInitialSplits, maxOutstandingSplits, maxOutstandingSplitsSize, maxSplitsPerSecond, hiveSplitLoader, executor, new CounterStat(), dynamicFilterSupplier, userDefinedCachePredicates, typeManager, hiveConfig, hiveStorageFormat);
            break;
        default:
            throw new IllegalArgumentException("Unknown splitSchedulingStrategy: " + splitSchedulingStrategy);
    }
    hiveSplitLoader.start(splitSource);
    if (queryType.isPresent() && queryType.get() == QueryType.VACUUM) {
        HdfsContext hdfsContext = new HdfsContext(session, table.getDatabaseName(), table.getTableName());
        return new HiveVacuumSplitSource(splitSource, (HiveVacuumTableHandle) queryInfo.get("vacuumHandle"), hdfsEnvironment, hdfsContext, session);
    }
    return splitSource;
}

Example 2 with GROUPED_SCHEDULING

use of io.prestosql.spi.connector.ConnectorSplitManager.SplitSchedulingStrategy.GROUPED_SCHEDULING in project boostkit-bigdata by kunpengcompute.

the class HiveSplitManager method getSplits.

@Override
public ConnectorSplitSource getSplits(ConnectorTransactionHandle transaction, ConnectorSession session, ConnectorTableHandle tableHandle, SplitSchedulingStrategy splitSchedulingStrategy, Supplier<List<Set<DynamicFilter>>> dynamicFilterSupplier, Optional<QueryType> queryType, Map<String, Object> queryInfo, Set<TupleDomain<ColumnMetadata>> userDefinedCachePredicates, boolean partOfReuse) {
    HiveTableHandle hiveTable = (HiveTableHandle) tableHandle;
    SchemaTableName tableName = hiveTable.getSchemaTableName();
    // get table metadata
    SemiTransactionalHiveMetastore metastore = metastoreProvider.apply((HiveTransactionHandle) transaction);
    Table table = metastore.getTable(new HiveIdentity(session), tableName.getSchemaName(), tableName.getTableName()).orElseThrow(() -> new TableNotFoundException(tableName));
    if (table.getStorage().getStorageFormat().getInputFormat().contains("carbon")) {
        throw new PrestoException(NOT_SUPPORTED, "Hive connector can't read carbondata tables");
    }
    // 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(session, metastore, new HiveIdentity(session), hiveTable);
    // short circuit if we don't have any partitions
    if (partitions.isEmpty()) {
        return new FixedSplitSource(ImmutableList.of());
    }
    // get buckets from first partition (arbitrary)
    Optional<HiveBucketing.HiveBucketFilter> bucketFilter = hiveTable.getBucketFilter();
    // validate bucket bucketed execution
    Optional<HiveBucketHandle> bucketHandle = hiveTable.getBucketHandle();
    if ((splitSchedulingStrategy == GROUPED_SCHEDULING) && !bucketHandle.isPresent()) {
        throw new PrestoException(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));
    HiveSplitLoader hiveSplitLoader = new BackgroundHiveSplitLoader(table, hivePartitions, hiveTable.getCompactEffectivePredicate(), BackgroundHiveSplitLoader.BucketSplitInfo.createBucketSplitInfo(bucketHandle, bucketFilter), session, hdfsEnvironment, namenodeStats, directoryLister, executor, splitLoaderConcurrency, recursiveDfsWalkerEnabled, metastore.getValidWriteIds(session, hiveTable, queryType.map(t -> t == QueryType.VACUUM).orElse(false)).map(validTxnWriteIdList -> validTxnWriteIdList.getTableValidWriteIdList(table.getDatabaseName() + "." + table.getTableName())), dynamicFilterSupplier, queryType, queryInfo, typeManager);
    HiveSplitSource splitSource;
    HiveStorageFormat hiveStorageFormat = HiveMetadata.extractHiveStorageFormat(table);
    switch(splitSchedulingStrategy) {
        case UNGROUPED_SCHEDULING:
            splitSource = HiveSplitSource.allAtOnce(session, table.getDatabaseName(), table.getTableName(), // For reuse, we should make sure to have same split size all time for a table.
            partOfReuse ? 0 : maxInitialSplits, maxOutstandingSplits, maxOutstandingSplitsSize, maxSplitsPerSecond, hiveSplitLoader, executor, new CounterStat(), dynamicFilterSupplier, userDefinedCachePredicates, typeManager, hiveConfig, hiveStorageFormat);
            break;
        case GROUPED_SCHEDULING:
            splitSource = HiveSplitSource.bucketed(session, table.getDatabaseName(), table.getTableName(), // For reuse, we should make sure to have same split size all time for a table.
            partOfReuse ? 0 : maxInitialSplits, maxOutstandingSplits, maxOutstandingSplitsSize, maxSplitsPerSecond, hiveSplitLoader, executor, new CounterStat(), dynamicFilterSupplier, userDefinedCachePredicates, typeManager, hiveConfig, hiveStorageFormat);
            break;
        default:
            throw new IllegalArgumentException("Unknown splitSchedulingStrategy: " + splitSchedulingStrategy);
    }
    hiveSplitLoader.start(splitSource);
    if (queryType.isPresent() && queryType.get() == QueryType.VACUUM) {
        HdfsContext hdfsContext = new HdfsContext(session, table.getDatabaseName(), table.getTableName());
        return new HiveVacuumSplitSource(splitSource, (HiveVacuumTableHandle) queryInfo.get("vacuumHandle"), hdfsEnvironment, hdfsContext, session);
    }
    return splitSource;
}