Examples with PartitioningScheme io.prestosql.sql.planner.PartitioningScheme used on opensource projects

Example 1 with PartitioningScheme

use of io.prestosql.sql.planner.PartitioningScheme in project hetu-core by openlookeng.

the class TestUtil method createExchangePlanFragment.

private static PlanFragment createExchangePlanFragment(RowExpression expr) {
    Symbol testSymbol = new Symbol("a");
    Map<Symbol, ColumnHandle> scanAssignments = ImmutableMap.<Symbol, ColumnHandle>builder().put(testSymbol, new TestingMetadata.TestingColumnHandle("a")).build();
    Map<Symbol, ColumnHandle> assignments = Maps.filterKeys(scanAssignments, Predicates.in(ImmutableList.of(testSymbol)));
    TableScanNode tableScanNode = new TableScanNode(new PlanNodeId(UUID.randomUUID().toString()), makeTableHandle(TupleDomain.none()), ImmutableList.copyOf(assignments.keySet()), assignments, TupleDomain.none(), Optional.empty(), ReuseExchangeOperator.STRATEGY.REUSE_STRATEGY_DEFAULT, new UUID(0, 0), 0, false);
    PlanBuilder planBuilder = new PlanBuilder(new PlanNodeIdAllocator(), dummyMetadata());
    FilterNode filterNode = planBuilder.filter(expr, tableScanNode);
    PlanNode planNode = new LimitNode(new PlanNodeId("limit"), filterNode, 1, false);
    ImmutableMap.Builder<Symbol, Type> types = ImmutableMap.builder();
    for (Symbol symbol : planNode.getOutputSymbols()) {
        types.put(symbol, VARCHAR);
    }
    return new PlanFragment(new PlanFragmentId("limit_fragment_id"), planNode, types.build(), SOURCE_DISTRIBUTION, ImmutableList.of(planNode.getId()), new PartitioningScheme(Partitioning.create(SINGLE_DISTRIBUTION, ImmutableList.of()), planNode.getOutputSymbols()), ungroupedExecution(), StatsAndCosts.empty(), Optional.empty(), Optional.empty(), Optional.empty());
}

Example 2 with PartitioningScheme

use of io.prestosql.sql.planner.PartitioningScheme in project hetu-core by openlookeng.

the class PlanPrinter method formatFragment.

private static String formatFragment(Function<TableScanNode, TableInfo> tableInfoSupplier, ValuePrinter valuePrinter, PlanFragment fragment, Optional<StageInfo> stageInfo, Optional<Map<PlanNodeId, PlanNodeStats>> planNodeStats, boolean verbose, List<PlanFragment> allFragments, Metadata metadata) {
    StringBuilder builder = new StringBuilder();
    builder.append(format("Fragment %s [%s]\n", fragment.getId(), fragment.getPartitioning()));
    if (stageInfo.isPresent()) {
        StageStats stageStats = stageInfo.get().getStageStats();
        double avgPositionsPerTask = stageInfo.get().getTasks().stream().mapToLong(task -> task.getStats().getProcessedInputPositions()).average().orElse(Double.NaN);
        double squaredDifferences = stageInfo.get().getTasks().stream().mapToDouble(task -> Math.pow(task.getStats().getProcessedInputPositions() - avgPositionsPerTask, 2)).sum();
        double sdAmongTasks = Math.sqrt(squaredDifferences / stageInfo.get().getTasks().size());
        builder.append(indentString(1)).append(format("CPU: %s, Scheduled: %s, Input: %s (%s); per task: avg.: %s std.dev.: %s, Output: %s (%s)\n", stageStats.getTotalCpuTime().convertToMostSuccinctTimeUnit(), stageStats.getTotalScheduledTime().convertToMostSuccinctTimeUnit(), formatPositions(stageStats.getProcessedInputPositions()), stageStats.getProcessedInputDataSize(), formatDouble(avgPositionsPerTask), formatDouble(sdAmongTasks), formatPositions(stageStats.getOutputPositions()), stageStats.getOutputDataSize()));
    }
    PartitioningScheme partitioningScheme = fragment.getPartitioningScheme();
    builder.append(indentString(1)).append(format("Output layout: [%s]\n", Joiner.on(", ").join(partitioningScheme.getOutputLayout())));
    boolean replicateNullsAndAny = partitioningScheme.isReplicateNullsAndAny();
    List<String> arguments = partitioningScheme.getPartitioning().getArguments().stream().map(argument -> {
        if (argument.isConstant()) {
            NullableValue constant = argument.getConstant();
            String printableValue = valuePrinter.castToVarchar(constant.getType(), constant.getValue());
            return constant.getType().getDisplayName() + "(" + printableValue + ")";
        }
        return argument.getColumn().toString();
    }).collect(toImmutableList());
    builder.append(indentString(1));
    if (replicateNullsAndAny) {
        builder.append(format("Output partitioning: %s (replicate nulls and any) [%s]%s\n", partitioningScheme.getPartitioning().getHandle(), Joiner.on(", ").join(arguments), formatHash(partitioningScheme.getHashColumn())));
    } else {
        builder.append(format("Output partitioning: %s [%s]%s\n", partitioningScheme.getPartitioning().getHandle(), Joiner.on(", ").join(arguments), formatHash(partitioningScheme.getHashColumn())));
    }
    builder.append(indentString(1)).append(format("Stage Execution Strategy: %s\n", fragment.getStageExecutionDescriptor().getStageExecutionStrategy()));
    TypeProvider typeProvider = TypeProvider.copyOf(allFragments.stream().flatMap(f -> f.getSymbols().entrySet().stream()).distinct().collect(toImmutableMap(Map.Entry::getKey, Map.Entry::getValue)));
    builder.append(new PlanPrinter(fragment.getRoot(), typeProvider, Optional.of(fragment.getStageExecutionDescriptor()), tableInfoSupplier, valuePrinter, fragment.getStatsAndCosts(), planNodeStats, metadata).toText(verbose, 1)).append("\n");
    return builder.toString();
}

Example 3 with PartitioningScheme

use of io.prestosql.sql.planner.PartitioningScheme in project hetu-core by openlookeng.

the class TestSqlStageExecution method createExchangePlanFragment.

private static PlanFragment createExchangePlanFragment() {
    PlanNode planNode = new RemoteSourceNode(new PlanNodeId("exchange"), ImmutableList.of(new PlanFragmentId("source")), ImmutableList.of(new Symbol("column")), Optional.empty(), REPARTITION);
    ImmutableMap.Builder<Symbol, Type> types = ImmutableMap.builder();
    for (Symbol symbol : planNode.getOutputSymbols()) {
        types.put(symbol, VARCHAR);
    }
    return new PlanFragment(new PlanFragmentId("exchange_fragment_id"), planNode, types.build(), SOURCE_DISTRIBUTION, ImmutableList.of(planNode.getId()), new PartitioningScheme(Partitioning.create(SINGLE_DISTRIBUTION, ImmutableList.of()), planNode.getOutputSymbols()), ungroupedExecution(), StatsAndCosts.empty(), Optional.empty(), Optional.empty(), Optional.empty());
}

Example 4 with PartitioningScheme

use of io.prestosql.sql.planner.PartitioningScheme in project hetu-core by openlookeng.

the class PushProjectionThroughExchange method apply.

@Override
public Result apply(ProjectNode project, Captures captures, Context context) {
    ExchangeNode exchange = captures.get(CHILD);
    Set<Symbol> partitioningColumns = exchange.getPartitioningScheme().getPartitioning().getColumns();
    ImmutableList.Builder<PlanNode> newSourceBuilder = ImmutableList.builder();
    ImmutableList.Builder<List<Symbol>> inputsBuilder = ImmutableList.builder();
    for (int i = 0; i < exchange.getSources().size(); i++) {
        Map<Symbol, VariableReferenceExpression> outputToInputMap = extractExchangeOutputToInput(exchange, i, context.getSymbolAllocator().getTypes());
        Assignments.Builder projections = Assignments.builder();
        ImmutableList.Builder<Symbol> inputs = ImmutableList.builder();
        // Need to retain the partition keys for the exchange
        partitioningColumns.stream().map(outputToInputMap::get).forEach(nameReference -> {
            Symbol symbol = new Symbol(nameReference.getName());
            projections.put(symbol, nameReference);
            inputs.add(symbol);
        });
        if (exchange.getPartitioningScheme().getHashColumn().isPresent()) {
            // Need to retain the hash symbol for the exchange
            projections.put(exchange.getPartitioningScheme().getHashColumn().get(), toVariableReference(exchange.getPartitioningScheme().getHashColumn().get(), context.getSymbolAllocator().getTypes()));
            inputs.add(exchange.getPartitioningScheme().getHashColumn().get());
        }
        if (exchange.getOrderingScheme().isPresent()) {
            // need to retain ordering columns for the exchange
            exchange.getOrderingScheme().get().getOrderBy().stream().filter(symbol -> !partitioningColumns.contains(symbol)).map(outputToInputMap::get).forEach(nameReference -> {
                Symbol symbol = new Symbol(nameReference.getName());
                projections.put(symbol, nameReference);
                inputs.add(symbol);
            });
        }
        for (Map.Entry<Symbol, RowExpression> projection : project.getAssignments().entrySet()) {
            checkArgument(!isExpression(projection.getValue()), "Cannot contain OriginalExpression after AddExchange");
            Map<VariableReferenceExpression, VariableReferenceExpression> variableOutputToInputMap = new LinkedHashMap<>();
            outputToInputMap.forEach(((symbol, variable) -> variableOutputToInputMap.put(new VariableReferenceExpression(symbol.getName(), context.getSymbolAllocator().getTypes().get(symbol)), variable)));
            RowExpression translatedExpression = RowExpressionVariableInliner.inlineVariables(variableOutputToInputMap, projection.getValue());
            Symbol symbol = context.getSymbolAllocator().newSymbol(translatedExpression);
            projections.put(symbol, translatedExpression);
            inputs.add(symbol);
        }
        newSourceBuilder.add(new ProjectNode(context.getIdAllocator().getNextId(), exchange.getSources().get(i), projections.build()));
        inputsBuilder.add(inputs.build());
    }
    // Construct the output symbols in the same order as the sources
    ImmutableList.Builder<Symbol> outputBuilder = ImmutableList.builder();
    partitioningColumns.forEach(outputBuilder::add);
    exchange.getPartitioningScheme().getHashColumn().ifPresent(outputBuilder::add);
    if (exchange.getOrderingScheme().isPresent()) {
        exchange.getOrderingScheme().get().getOrderBy().stream().filter(symbol -> !partitioningColumns.contains(symbol)).forEach(outputBuilder::add);
    }
    for (Map.Entry<Symbol, RowExpression> projection : project.getAssignments().entrySet()) {
        outputBuilder.add(projection.getKey());
    }
    // outputBuilder contains all partition and hash symbols so simply swap the output layout
    PartitioningScheme partitioningScheme = new PartitioningScheme(exchange.getPartitioningScheme().getPartitioning(), outputBuilder.build(), exchange.getPartitioningScheme().getHashColumn(), exchange.getPartitioningScheme().isReplicateNullsAndAny(), exchange.getPartitioningScheme().getBucketToPartition());
    PlanNode result = new ExchangeNode(exchange.getId(), exchange.getType(), exchange.getScope(), partitioningScheme, newSourceBuilder.build(), inputsBuilder.build(), exchange.getOrderingScheme(), AggregationNode.AggregationType.HASH);
    // we need to strip unnecessary symbols (hash, partitioning columns).
    return Result.ofPlanNode(restrictOutputs(context.getIdAllocator(), result, ImmutableSet.copyOf(project.getOutputSymbols()), true, context.getSymbolAllocator().getTypes()).orElse(result));
}

Example 5 with PartitioningScheme

use of io.prestosql.sql.planner.PartitioningScheme in project hetu-core by openlookeng.

the class PushPartialAggregationThroughExchange method pushPartial.

private PlanNode pushPartial(AggregationNode aggregation, ExchangeNode exchange, Context context) {
    List<PlanNode> partials = new ArrayList<>();
    for (int i = 0; i < exchange.getSources().size(); i++) {
        PlanNode source = exchange.getSources().get(i);
        SymbolMapper.Builder mappingsBuilder = SymbolMapper.builder();
        for (int outputIndex = 0; outputIndex < exchange.getOutputSymbols().size(); outputIndex++) {
            Symbol output = exchange.getOutputSymbols().get(outputIndex);
            Symbol input = exchange.getInputs().get(i).get(outputIndex);
            if (!output.equals(input)) {
                mappingsBuilder.put(output, input);
            }
        }
        SymbolMapper symbolMapper = mappingsBuilder.build();
        if (symbolMapper.getTypes() == null) {
            symbolMapper.setTypes(context.getSymbolAllocator().getTypes());
        }
        AggregationNode mappedPartial = symbolMapper.map(aggregation, source, context.getIdAllocator());
        Assignments.Builder assignments = Assignments.builder();
        for (Symbol output : aggregation.getOutputSymbols()) {
            Symbol input = symbolMapper.map(output);
            assignments.put(output, VariableReferenceSymbolConverter.toVariableReference(input, context.getSymbolAllocator().getTypes()));
        }
        partials.add(new ProjectNode(context.getIdAllocator().getNextId(), mappedPartial, assignments.build()));
    }
    for (PlanNode node : partials) {
        verify(aggregation.getOutputSymbols().equals(node.getOutputSymbols()));
    }
    // Since this exchange source is now guaranteed to have the same symbols as the inputs to the the partial
    // aggregation, we don't need to rewrite symbols in the partitioning function
    PartitioningScheme partitioning = new PartitioningScheme(exchange.getPartitioningScheme().getPartitioning(), aggregation.getOutputSymbols(), exchange.getPartitioningScheme().getHashColumn(), exchange.getPartitioningScheme().isReplicateNullsAndAny(), exchange.getPartitioningScheme().getBucketToPartition());
    return new ExchangeNode(context.getIdAllocator().getNextId(), exchange.getType(), exchange.getScope(), partitioning, partials, ImmutableList.copyOf(Collections.nCopies(partials.size(), aggregation.getOutputSymbols())), Optional.empty(), aggregation.getAggregationType());
}