Examples with ProjectNode io.prestosql.spi.plan.ProjectNode used on opensource projects

Example 16 with ProjectNode

use of io.prestosql.spi.plan.ProjectNode in project hetu-core by openlookeng.

the class QueryPlanner method aggregate.

private PlanBuilder aggregate(PlanBuilder inputSubPlan, QuerySpecification node) {
    PlanBuilder subPlan = inputSubPlan;
    if (!analysis.isAggregation(node)) {
        return subPlan;
    }
    // 1. Pre-project all scalar inputs (arguments and non-trivial group by expressions)
    Set<Expression> groupByExpressions = ImmutableSet.copyOf(analysis.getGroupByExpressions(node));
    ImmutableList.Builder<Expression> arguments = ImmutableList.builder();
    analysis.getAggregates(node).stream().map(FunctionCall::getArguments).flatMap(List::stream).filter(// lambda expression is generated at execution time
    exp -> !(exp instanceof LambdaExpression)).forEach(arguments::add);
    analysis.getAggregates(node).stream().map(FunctionCall::getOrderBy).filter(Optional::isPresent).map(Optional::get).map(OrderBy::getSortItems).flatMap(List::stream).map(SortItem::getSortKey).forEach(arguments::add);
    // filter expressions need to be projected first
    analysis.getAggregates(node).stream().map(FunctionCall::getFilter).filter(Optional::isPresent).map(Optional::get).forEach(arguments::add);
    Iterable<Expression> inputs = Iterables.concat(groupByExpressions, arguments.build());
    subPlan = handleSubqueries(subPlan, node, inputs);
    if (!Iterables.isEmpty(inputs)) {
        // avoid an empty projection if the only aggregation is COUNT (which has no arguments)
        subPlan = project(subPlan, inputs);
    }
    // 2. Aggregate
    // 2.a. Rewrite aggregate arguments
    TranslationMap argumentTranslations = new TranslationMap(subPlan.getRelationPlan(), analysis, lambdaDeclarationToSymbolMap);
    ImmutableList.Builder<Symbol> aggregationArgumentsBuilder = ImmutableList.builder();
    for (Expression argument : arguments.build()) {
        Symbol symbol = subPlan.translate(argument);
        argumentTranslations.put(argument, symbol);
        aggregationArgumentsBuilder.add(symbol);
    }
    List<Symbol> aggregationArguments = aggregationArgumentsBuilder.build();
    // 2.b. Rewrite grouping columns
    TranslationMap groupingTranslations = new TranslationMap(subPlan.getRelationPlan(), analysis, lambdaDeclarationToSymbolMap);
    Map<Symbol, Symbol> groupingSetMappings = new LinkedHashMap<>();
    for (Expression expression : groupByExpressions) {
        Symbol input = subPlan.translate(expression);
        Symbol output = planSymbolAllocator.newSymbol(expression, analysis.getTypeWithCoercions(expression), "gid");
        groupingTranslations.put(expression, output);
        groupingSetMappings.put(output, input);
    }
    // This tracks the grouping sets before complex expressions are considered (see comments below)
    // It's also used to compute the descriptors needed to implement grouping()
    List<Set<FieldId>> columnOnlyGroupingSets = ImmutableList.of(ImmutableSet.of());
    List<List<Symbol>> groupingSets = ImmutableList.of(ImmutableList.of());
    if (node.getGroupBy().isPresent()) {
        // For the purpose of "distinct", we need to canonicalize column references that may have varying
        // syntactic forms (e.g., "t.a" vs "a"). Thus we need to enumerate grouping sets based on the underlying
        // fieldId associated with each column reference expression.
        // The catch is that simple group-by expressions can be arbitrary expressions (this is a departure from the SQL specification).
        // But, they don't affect the number of grouping sets or the behavior of "distinct" . We can compute all the candidate
        // grouping sets in terms of fieldId, dedup as appropriate and then cross-join them with the complex expressions.
        Analysis.GroupingSetAnalysis groupingSetAnalysis = analysis.getGroupingSets(node);
        columnOnlyGroupingSets = enumerateGroupingSets(groupingSetAnalysis);
        if (node.getGroupBy().get().isDistinct()) {
            columnOnlyGroupingSets = columnOnlyGroupingSets.stream().distinct().collect(toImmutableList());
        }
        // add in the complex expressions an turn materialize the grouping sets in terms of plan columns
        ImmutableList.Builder<List<Symbol>> groupingSetBuilder = ImmutableList.builder();
        for (Set<FieldId> groupingSet : columnOnlyGroupingSets) {
            ImmutableList.Builder<Symbol> columns = ImmutableList.builder();
            groupingSetAnalysis.getComplexExpressions().stream().map(groupingTranslations::get).forEach(columns::add);
            groupingSet.stream().map(field -> groupingTranslations.get(new FieldReference(field.getFieldIndex()))).forEach(columns::add);
            groupingSetBuilder.add(columns.build());
        }
        groupingSets = groupingSetBuilder.build();
    }
    // 2.c. Generate GroupIdNode (multiple grouping sets) or ProjectNode (single grouping set)
    Optional<Symbol> groupIdSymbol = Optional.empty();
    if (groupingSets.size() > 1) {
        groupIdSymbol = Optional.of(planSymbolAllocator.newSymbol("groupId", BIGINT));
        GroupIdNode groupId = new GroupIdNode(idAllocator.getNextId(), subPlan.getRoot(), groupingSets, groupingSetMappings, aggregationArguments, groupIdSymbol.get());
        subPlan = new PlanBuilder(groupingTranslations, groupId);
    } else {
        Assignments.Builder assignments = Assignments.builder();
        aggregationArguments.forEach(symbol -> assignments.put(symbol, castToRowExpression(toSymbolReference(symbol))));
        groupingSetMappings.forEach((key, value) -> assignments.put(key, castToRowExpression(toSymbolReference(value))));
        ProjectNode project = new ProjectNode(idAllocator.getNextId(), subPlan.getRoot(), assignments.build());
        subPlan = new PlanBuilder(groupingTranslations, project);
    }
    TranslationMap aggregationTranslations = new TranslationMap(subPlan.getRelationPlan(), analysis, lambdaDeclarationToSymbolMap);
    aggregationTranslations.copyMappingsFrom(groupingTranslations);
    // 2.d. Rewrite aggregates
    ImmutableMap.Builder<Symbol, Aggregation> aggregationsBuilder = ImmutableMap.builder();
    boolean needPostProjectionCoercion = false;
    for (FunctionCall aggregate : analysis.getAggregates(node)) {
        Expression rewritten = argumentTranslations.rewrite(aggregate);
        Symbol newSymbol = planSymbolAllocator.newSymbol(rewritten, analysis.getType(aggregate));
        // Therefore we can end up with this implicit cast, and have to move it into a post-projection
        if (rewritten instanceof Cast) {
            rewritten = ((Cast) rewritten).getExpression();
            needPostProjectionCoercion = true;
        }
        aggregationTranslations.put(aggregate, newSymbol);
        FunctionCall functionCall = (FunctionCall) rewritten;
        aggregationsBuilder.put(newSymbol, new Aggregation(call(aggregate.getName().getSuffix(), analysis.getFunctionHandle(aggregate), analysis.getType(aggregate), functionCall.getArguments().stream().map(OriginalExpressionUtils::castToRowExpression).collect(toImmutableList())), functionCall.getArguments().stream().map(OriginalExpressionUtils::castToRowExpression).collect(toImmutableList()), functionCall.isDistinct(), functionCall.getFilter().map(SymbolUtils::from), functionCall.getOrderBy().map(OrderingSchemeUtils::fromOrderBy), Optional.empty()));
    }
    Map<Symbol, Aggregation> aggregations = aggregationsBuilder.build();
    ImmutableSet.Builder<Integer> globalGroupingSets = ImmutableSet.builder();
    for (int i = 0; i < groupingSets.size(); i++) {
        if (groupingSets.get(i).isEmpty()) {
            globalGroupingSets.add(i);
        }
    }
    ImmutableList.Builder<Symbol> groupingKeys = ImmutableList.builder();
    groupingSets.stream().flatMap(List::stream).distinct().forEach(groupingKeys::add);
    groupIdSymbol.ifPresent(groupingKeys::add);
    AggregationNode aggregationNode = new AggregationNode(idAllocator.getNextId(), subPlan.getRoot(), aggregations, groupingSets(groupingKeys.build(), groupingSets.size(), globalGroupingSets.build()), ImmutableList.of(), AggregationNode.Step.SINGLE, Optional.empty(), groupIdSymbol, AggregationNode.AggregationType.HASH, Optional.empty());
    subPlan = new PlanBuilder(aggregationTranslations, aggregationNode);
    // TODO: this is a hack, we should change type coercions to coerce the inputs to functions/operators instead of coercing the output
    if (needPostProjectionCoercion) {
        ImmutableList.Builder<Expression> alreadyCoerced = ImmutableList.builder();
        alreadyCoerced.addAll(groupByExpressions);
        groupIdSymbol.map(SymbolUtils::toSymbolReference).ifPresent(alreadyCoerced::add);
        subPlan = explicitCoercionFields(subPlan, alreadyCoerced.build(), analysis.getAggregates(node));
    }
    // 4. Project and re-write all grouping functions
    return handleGroupingOperations(subPlan, node, groupIdSymbol, columnOnlyGroupingSets);
}

Example 17 with ProjectNode

use of io.prestosql.spi.plan.ProjectNode in project hetu-core by openlookeng.

the class QueryPlanner method coerce.

/**
 * Creates a projection with any additional coercions by identity of the provided expressions.
 *
 * @return the new subplan and a mapping of each expression to the symbol representing the coercion or an existing symbol if a coercion wasn't needed
 */
public static PlanAndMappings coerce(PlanBuilder inputSubPlan, List<Expression> expressions, Analysis analysis, PlanNodeIdAllocator idAllocator, PlanSymbolAllocator planSymbolAllocator, TypeCoercion typeCoercion) {
    PlanBuilder subPlan = inputSubPlan;
    Assignments.Builder assignments = Assignments.builder();
    assignments.putAll(AssignmentUtils.identityAsSymbolReferences(subPlan.getRoot().getOutputSymbols()));
    ImmutableMap.Builder<NodeRef<Expression>, Symbol> mappings = ImmutableMap.builder();
    for (Expression expression : expressions) {
        Type coercion = analysis.getCoercion(expression);
        if (coercion != null) {
            Type type = analysis.getType(expression);
            Symbol symbol = planSymbolAllocator.newSymbol(expression, coercion);
            assignments.put(symbol, castToRowExpression(new Cast(subPlan.rewrite(expression), coercion.getTypeSignature().toString(), false, typeCoercion.isTypeOnlyCoercion(type, coercion))));
            mappings.put(NodeRef.of(expression), symbol);
        } else {
            mappings.put(NodeRef.of(expression), subPlan.translate(expression));
        }
    }
    subPlan = subPlan.withNewRoot(new ProjectNode(idAllocator.getNextId(), subPlan.getRoot(), assignments.build()));
    return new PlanAndMappings(subPlan, mappings.build());
}

Example 18 with ProjectNode

use of io.prestosql.spi.plan.ProjectNode in project hetu-core by openlookeng.

the class LogicalPlanner method createInsertPlan.

private RelationPlan createInsertPlan(Analysis analysis, Insert insertStatement) {
    Analysis.Insert insert = analysis.getInsert().get();
    TableMetadata tableMetadata = metadata.getTableMetadata(session, insert.getTarget());
    List<ColumnMetadata> visibleTableColumns = tableMetadata.getColumns().stream().filter(column -> !column.isHidden()).collect(toImmutableList());
    List<String> visibleTableColumnNames = visibleTableColumns.stream().map(ColumnMetadata::getName).collect(toImmutableList());
    RelationPlan plan = createRelationPlan(analysis, insertStatement.getQuery());
    Map<String, ColumnHandle> columns = metadata.getColumnHandles(session, insert.getTarget());
    Assignments.Builder assignments = Assignments.builder();
    for (ColumnMetadata column : tableMetadata.getColumns()) {
        if (column.isHidden()) {
            continue;
        }
        Symbol output = planSymbolAllocator.newSymbol(column.getName(), column.getType());
        int index = insert.getColumns().indexOf(columns.get(column.getName()));
        if (index < 0) {
            Expression cast = new Cast(new NullLiteral(), column.getType().getTypeSignature().toString());
            assignments.put(output, castToRowExpression(cast));
        } else {
            Symbol input = plan.getSymbol(index);
            Type tableType = column.getType();
            Type queryType = planSymbolAllocator.getTypes().get(input);
            if (queryType.equals(tableType) || typeCoercion.isTypeOnlyCoercion(queryType, tableType)) {
                assignments.put(output, castToRowExpression(toSymbolReference(input)));
            } else {
                Expression cast = noTruncationCast(toSymbolReference(input), queryType, tableType);
                assignments.put(output, castToRowExpression(cast));
            }
        }
    }
    ProjectNode projectNode = new ProjectNode(idAllocator.getNextId(), plan.getRoot(), assignments.build());
    List<Field> fields = visibleTableColumns.stream().map(column -> Field.newUnqualified(column.getName(), column.getType())).collect(toImmutableList());
    Scope scope = Scope.builder().withRelationType(RelationId.anonymous(), new RelationType(fields)).build();
    plan = new RelationPlan(projectNode, scope, projectNode.getOutputSymbols());
    Optional<NewTableLayout> newTableLayout = metadata.getInsertLayout(session, insert.getTarget());
    String catalogName = insert.getTarget().getCatalogName().getCatalogName();
    TableStatisticsMetadata statisticsMetadata = metadata.getStatisticsCollectionMetadataForWrite(session, catalogName, tableMetadata.getMetadata());
    return createTableWriterPlan(analysis, plan, new InsertReference(insert.getTarget(), analysis.isInsertOverwrite()), visibleTableColumnNames, newTableLayout, statisticsMetadata);
}

Example 19 with ProjectNode

use of io.prestosql.spi.plan.ProjectNode in project hetu-core by openlookeng.

the class LogicalPlanner method createInsertCubePlan.

private RelationPlan createInsertCubePlan(Analysis analysis, InsertCube insertCubeStatement) {
    Analysis.CubeInsert insert = analysis.getCubeInsert().get();
    TableMetadata tableMetadata = metadata.getTableMetadata(session, insert.getTarget());
    List<ColumnMetadata> visibleTableColumns = tableMetadata.getColumns().stream().filter(column -> !column.isHidden()).collect(toImmutableList());
    List<String> visibleTableColumnNames = visibleTableColumns.stream().map(ColumnMetadata::getName).collect(toImmutableList());
    RelationPlan plan = createRelationPlan(analysis, insertCubeStatement.getQuery());
    Map<String, ColumnHandle> columns = metadata.getColumnHandles(session, insert.getTarget());
    Assignments.Builder assignments = Assignments.builder();
    for (ColumnMetadata column : tableMetadata.getColumns()) {
        if (column.isHidden()) {
            continue;
        }
        Symbol output = planSymbolAllocator.newSymbol(column.getName(), column.getType());
        int index = insert.getColumns().indexOf(columns.get(column.getName()));
        if (index < 0) {
            Expression cast = new Cast(new NullLiteral(), column.getType().getTypeSignature().toString());
            assignments.put(output, castToRowExpression(cast));
        } else {
            Symbol input = plan.getSymbol(index);
            Type tableType = column.getType();
            Type queryType = planSymbolAllocator.getTypes().get(input);
            if (queryType.equals(tableType) || typeCoercion.isTypeOnlyCoercion(queryType, tableType)) {
                assignments.put(output, castToRowExpression(toSymbolReference(input)));
            } else {
                Expression cast = noTruncationCast(toSymbolReference(input), queryType, tableType);
                assignments.put(output, castToRowExpression(cast));
            }
        }
    }
    ProjectNode projectNode = new ProjectNode(idAllocator.getNextId(), plan.getRoot(), assignments.build());
    List<Field> fields = visibleTableColumns.stream().map(column -> Field.newUnqualified(column.getName(), column.getType())).collect(toImmutableList());
    Scope scope = Scope.builder().withRelationType(RelationId.anonymous(), new RelationType(fields)).build();
    plan = new RelationPlan(projectNode, scope, projectNode.getOutputSymbols());
    Optional<NewTableLayout> newTableLayout = metadata.getInsertLayout(session, insert.getTarget());
    String catalogName = insert.getTarget().getCatalogName().getCatalogName();
    TableStatisticsMetadata statisticsMetadata = metadata.getStatisticsCollectionMetadataForWrite(session, catalogName, tableMetadata.getMetadata());
    RelationPlan tableWriterPlan = createTableWriterPlan(analysis, plan, new InsertReference(insert.getTarget(), analysis.isCubeOverwrite()), visibleTableColumnNames, newTableLayout, statisticsMetadata);
    Expression rewritten = null;
    Set<Identifier> predicateColumns = new HashSet<>();
    if (insertCubeStatement.getWhere().isPresent()) {
        rewritten = new QueryPlanner(analysis, planSymbolAllocator, idAllocator, buildLambdaDeclarationToSymbolMap(analysis, planSymbolAllocator), metadata, session, namedSubPlan, uniqueIdAllocator).rewriteExpression(tableWriterPlan, insertCubeStatement.getWhere().get(), analysis, buildLambdaDeclarationToSymbolMap(analysis, planSymbolAllocator));
        predicateColumns.addAll(ExpressionUtils.getIdentifiers(rewritten));
    }
    CubeMetadata cubeMetadata = insert.getMetadata();
    if (!insertCubeStatement.isOverwrite() && !insertCubeStatement.getWhere().isPresent() && cubeMetadata.getCubeStatus() != CubeStatus.INACTIVE) {
        // Means data some data was inserted before, but trying to insert entire dataset
        throw new PrestoException(QUERY_REJECTED, "Cannot allow insert. Inserting entire dataset but cube already has partial data");
    } else if (insertCubeStatement.getWhere().isPresent()) {
        if (!canSupportPredicate(rewritten)) {
            throw new PrestoException(QUERY_REJECTED, String.format("Cannot support predicate '%s'", ExpressionFormatter.formatExpression(rewritten, Optional.empty())));
        }
        if (!insertCubeStatement.isOverwrite() && arePredicatesOverlapping(rewritten, cubeMetadata)) {
            throw new PrestoException(QUERY_REJECTED, String.format("Cannot allow insert. Cube already contains data for the given predicate '%s'", ExpressionFormatter.formatExpression(insertCubeStatement.getWhere().get(), Optional.empty())));
        }
    }
    TableHandle sourceTableHandle = insert.getSourceTable();
    // At this point it has been verified that source table has not been updated
    // so insert into cube should be allowed
    LongSupplier tableLastModifiedTimeSupplier = metadata.getTableLastModifiedTimeSupplier(session, sourceTableHandle);
    checkState(tableLastModifiedTimeSupplier != null, "Table last modified time is null");
    Map<Symbol, Type> predicateColumnsType = predicateColumns.stream().map(identifier -> new Symbol(identifier.getValue())).collect(Collectors.toMap(Function.identity(), symbol -> planSymbolAllocator.getTypes().get(symbol), (key1, ignored) -> key1));
    CubeFinishNode cubeFinishNode = new CubeFinishNode(idAllocator.getNextId(), tableWriterPlan.getRoot(), planSymbolAllocator.newSymbol("rows", BIGINT), new CubeUpdateMetadata(tableMetadata.getQualifiedName().toString(), tableLastModifiedTimeSupplier.getAsLong(), rewritten != null ? ExpressionFormatter.formatExpression(rewritten, Optional.empty()) : null, insertCubeStatement.isOverwrite()), predicateColumnsType);
    return new RelationPlan(cubeFinishNode, analysis.getScope(insertCubeStatement), cubeFinishNode.getOutputSymbols());
}

Example 20 with ProjectNode

use of io.prestosql.spi.plan.ProjectNode in project hetu-core by openlookeng.

the class QueryPlanner method explicitCoercionFields.

private PlanBuilder explicitCoercionFields(PlanBuilder subPlan, Iterable<Expression> alreadyCoerced, Iterable<? extends Expression> uncoerced) {
    TranslationMap translations = new TranslationMap(subPlan.getRelationPlan(), analysis, lambdaDeclarationToSymbolMap);
    Assignments.Builder projections = Assignments.builder();
    projections.putAll(coerce(uncoerced, subPlan, translations));
    for (Expression expression : alreadyCoerced) {
        if (expression instanceof SymbolReference) {
            // If this is an identity projection, no need to rewrite it
            // This is needed because certain synthetic identity expressions such as "group id" introduced when planning GROUPING
            // don't have a corresponding analysis, so the code below doesn't work for them
            projections.put(from(expression), castToRowExpression(expression));
            continue;
        }
        Symbol symbol = planSymbolAllocator.newSymbol(expression, analysis.getType(expression));
        Expression rewritten = subPlan.rewrite(expression);
        projections.put(symbol, castToRowExpression(rewritten));
        translations.put(expression, symbol);
    }
    return new PlanBuilder(translations, new ProjectNode(idAllocator.getNextId(), subPlan.getRoot(), projections.build()));
}