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

Example 11 with FixedSplitSource

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

the class AccumuloSplitManager method getSplits.

@Override
public ConnectorSplitSource getSplits(ConnectorTransactionHandle transactionHandle, ConnectorSession session, ConnectorTableHandle tableHandle, SplitSchedulingStrategy splitSchedulingStrategy, DynamicFilter dynamicFilter) {
    AccumuloTableHandle handle = (AccumuloTableHandle) tableHandle;
    String schemaName = handle.getSchema();
    String tableName = handle.getTable();
    String rowIdName = handle.getRowId();
    // Get non-row ID column constraints
    List<AccumuloColumnConstraint> constraints = getColumnConstraints(rowIdName, handle.getConstraint());
    // Get the row domain column range
    Optional<Domain> rDom = getRangeDomain(rowIdName, handle.getConstraint());
    // Call out to our client to retrieve all tablet split metadata using the row ID domain and the secondary index
    List<TabletSplitMetadata> tabletSplits = client.getTabletSplits(session, schemaName, tableName, rDom, constraints, handle.getSerializerInstance());
    // Pack the tablet split metadata into a connector split
    ImmutableList.Builder<ConnectorSplit> cSplits = ImmutableList.builder();
    for (TabletSplitMetadata splitMetadata : tabletSplits) {
        AccumuloSplit split = new AccumuloSplit(splitMetadata.getRanges().stream().map(SerializedRange::serialize).collect(Collectors.toList()), splitMetadata.getHostPort());
        cSplits.add(split);
    }
    return new FixedSplitSource(cSplits.build());
}

Example 12 with FixedSplitSource

use of io.trino.spi.connector.FixedSplitSource 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;
}

Example 13 with FixedSplitSource

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

the class JmxSplitManager method getSplits.

@Override
public ConnectorSplitSource getSplits(ConnectorTransactionHandle transaction, ConnectorSession session, ConnectorTableHandle table, SplitSchedulingStrategy splitSchedulingStrategy, DynamicFilter dynamicFilter) {
    JmxTableHandle tableHandle = (JmxTableHandle) table;
    // TODO is there a better way to get the node column?
    Optional<JmxColumnHandle> nodeColumnHandle = tableHandle.getColumnHandles().stream().filter(jmxColumnHandle -> jmxColumnHandle.getColumnName().equals(NODE_COLUMN_NAME)).findFirst();
    checkState(nodeColumnHandle.isPresent(), "Failed to find %s column", NODE_COLUMN_NAME);
    TupleDomain<ColumnHandle> nodeFilter = tableHandle.getNodeFilter();
    List<ConnectorSplit> splits = nodeManager.getAllNodes().stream().filter(node -> {
        NullableValue value = NullableValue.of(createUnboundedVarcharType(), utf8Slice(node.getNodeIdentifier()));
        return nodeFilter.overlaps(fromFixedValues(ImmutableMap.of(nodeColumnHandle.get(), value)));
    }).map(node -> new JmxSplit(ImmutableList.of(node.getHostAndPort()))).collect(toList());
    return new FixedSplitSource(splits);
}

Example 14 with FixedSplitSource

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

the class PhoenixSplitManager method getSplits.

@Override
public ConnectorSplitSource getSplits(ConnectorTransactionHandle transaction, ConnectorSession session, ConnectorTableHandle table, SplitSchedulingStrategy splitSchedulingStrategy, DynamicFilter dynamicFilter) {
    JdbcTableHandle tableHandle = (JdbcTableHandle) table;
    try (Connection connection = phoenixClient.getConnection(session)) {
        List<JdbcColumnHandle> columns = tableHandle.getColumns().map(columnSet -> columnSet.stream().map(JdbcColumnHandle.class::cast).collect(toList())).orElseGet(() -> phoenixClient.getColumns(session, tableHandle));
        PhoenixPreparedStatement inputQuery = (PhoenixPreparedStatement) phoenixClient.prepareStatement(session, connection, tableHandle, columns, Optional.empty());
        int maxScansPerSplit = session.getProperty(PhoenixSessionProperties.MAX_SCANS_PER_SPLIT, Integer.class);
        List<ConnectorSplit> splits = getSplits(inputQuery, maxScansPerSplit).stream().map(PhoenixInputSplit.class::cast).map(split -> new PhoenixSplit(getSplitAddresses(split), SerializedPhoenixInputSplit.serialize(split))).collect(toImmutableList());
        return new FixedSplitSource(splits);
    } catch (IOException | SQLException e) {
        throw new TrinoException(PHOENIX_SPLIT_ERROR, "Couldn't get Phoenix splits", e);
    }
}

Example 15 with FixedSplitSource

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

the class TpcdsSplitManager method getSplits.

@Override
public ConnectorSplitSource getSplits(ConnectorTransactionHandle transaction, ConnectorSession session, ConnectorTableHandle tableHandle, SplitSchedulingStrategy splitSchedulingStrategy, DynamicFilter dynamicFilter) {
    Set<Node> nodes = nodeManager.getRequiredWorkerNodes();
    checkState(!nodes.isEmpty(), "No TPCDS nodes available");
    int totalParts = nodes.size() * splitsPerNode;
    int partNumber = 0;
    // Split the data using split and skew by the number of nodes available.
    ImmutableList.Builder<ConnectorSplit> splits = ImmutableList.builder();
    for (Node node : nodes) {
        for (int i = 0; i < splitsPerNode; i++) {
            splits.add(new TpcdsSplit(partNumber, totalParts, ImmutableList.of(node.getHostAndPort()), noSexism));
            partNumber++;
        }
    }
    return new FixedSplitSource(splits.build());
}