Examples with Type io.trino.sql.planner.plan.JoinNode.Type used on opensource projects

Example 1 with Type

use of io.trino.sql.planner.plan.JoinNode.Type in project trino by trinodb.

the class DecorrelateUnnest method apply.

@Override
public Result apply(CorrelatedJoinNode correlatedJoinNode, Captures captures, Context context) {
    // determine shape of the subquery
    PlanNode searchRoot = correlatedJoinNode.getSubquery();
    // 1. find EnforceSingleRowNode in the subquery
    Optional<EnforceSingleRowNode> enforceSingleRow = PlanNodeSearcher.searchFrom(searchRoot, context.getLookup()).where(EnforceSingleRowNode.class::isInstance).recurseOnlyWhen(planNode -> false).findFirst();
    if (enforceSingleRow.isPresent()) {
        searchRoot = enforceSingleRow.get().getSource();
    }
    // 2. find correlated UnnestNode in the subquery
    Optional<UnnestNode> subqueryUnnest = PlanNodeSearcher.searchFrom(searchRoot, context.getLookup()).where(node -> isSupportedUnnest(node, correlatedJoinNode.getCorrelation(), context.getLookup())).recurseOnlyWhen(node -> node instanceof ProjectNode || (node instanceof LimitNode && ((LimitNode) node).getCount() > 0) || (node instanceof TopNNode && ((TopNNode) node).getCount() > 0)).findFirst();
    if (subqueryUnnest.isEmpty()) {
        return Result.empty();
    }
    UnnestNode unnestNode = subqueryUnnest.get();
    // assign unique id to input rows
    Symbol uniqueSymbol = context.getSymbolAllocator().newSymbol("unique", BIGINT);
    PlanNode input = new AssignUniqueId(context.getIdAllocator().getNextId(), correlatedJoinNode.getInput(), uniqueSymbol);
    // pre-project unnest symbols if they were pre-projected in subquery
    // The correlated UnnestNode either unnests correlation symbols directly, or unnests symbols produced by a projection that uses only correlation symbols.
    // Here, any underlying projection that was a source of the correlated UnnestNode, is appended as a source of the rewritten UnnestNode.
    // If the projection is not necessary for UnnestNode (i.e. it does not produce any unnest symbols), it should be pruned afterwards.
    PlanNode unnestSource = context.getLookup().resolve(unnestNode.getSource());
    if (unnestSource instanceof ProjectNode) {
        ProjectNode sourceProjection = (ProjectNode) unnestSource;
        input = new ProjectNode(sourceProjection.getId(), input, Assignments.builder().putIdentities(input.getOutputSymbols()).putAll(sourceProjection.getAssignments()).build());
    }
    // determine join type for rewritten UnnestNode
    Type unnestJoinType = LEFT;
    if (enforceSingleRow.isEmpty() && correlatedJoinNode.getType() == CorrelatedJoinNode.Type.INNER && unnestNode.getJoinType() == INNER) {
        unnestJoinType = INNER;
    }
    // make sure that the rewritten node is with ordinality, which might be necessary to restore inner unnest semantics after rewrite.
    Symbol ordinalitySymbol = unnestNode.getOrdinalitySymbol().orElseGet(() -> context.getSymbolAllocator().newSymbol("ordinality", BIGINT));
    // rewrite correlated join to UnnestNode.
    UnnestNode rewrittenUnnest = new UnnestNode(context.getIdAllocator().getNextId(), input, input.getOutputSymbols(), unnestNode.getMappings(), Optional.of(ordinalitySymbol), unnestJoinType, Optional.empty());
    // restore all nodes from the subquery
    PlanNode rewrittenPlan = Rewriter.rewriteNodeSequence(correlatedJoinNode.getSubquery(), input.getOutputSymbols(), ordinalitySymbol, uniqueSymbol, rewrittenUnnest, context.getSession(), metadata, context.getLookup(), context.getIdAllocator(), context.getSymbolAllocator());
    // between unnested rows and synthetic rows added by left unnest.
    if (unnestNode.getJoinType() == INNER && rewrittenUnnest.getJoinType() == LEFT) {
        Assignments.Builder assignments = Assignments.builder().putIdentities(correlatedJoinNode.getInput().getOutputSymbols());
        for (Symbol subquerySymbol : correlatedJoinNode.getSubquery().getOutputSymbols()) {
            assignments.put(subquerySymbol, new IfExpression(new IsNullPredicate(ordinalitySymbol.toSymbolReference()), new Cast(new NullLiteral(), toSqlType(context.getSymbolAllocator().getTypes().get(subquerySymbol))), subquerySymbol.toSymbolReference()));
        }
        rewrittenPlan = new ProjectNode(context.getIdAllocator().getNextId(), rewrittenPlan, assignments.build());
    }
    // restrict outputs
    return Result.ofPlanNode(restrictOutputs(context.getIdAllocator(), rewrittenPlan, ImmutableSet.copyOf(correlatedJoinNode.getOutputSymbols())).orElse(rewrittenPlan));
}

Example 2 with Type

use of io.trino.sql.planner.plan.JoinNode.Type in project trino by trinodb.

the class TestDetermineJoinDistributionType method testReplicatesWhenSourceIsSmall.

@Test
public void testReplicatesWhenSourceIsSmall() {
    // variable width so that average row size is respected
    VarcharType symbolType = createUnboundedVarcharType();
    int aRows = 10_000;
    int bRows = 10;
    // output size exceeds AUTOMATIC_RESTRICTED limit
    PlanNodeStatsEstimate aStatsEstimate = PlanNodeStatsEstimate.builder().setOutputRowCount(aRows).addSymbolStatistics(ImmutableMap.of(new Symbol("A1"), new SymbolStatsEstimate(0, 100, 0, 640000d * 10000, 10))).build();
    // output size exceeds AUTOMATIC_RESTRICTED limit
    PlanNodeStatsEstimate bStatsEstimate = PlanNodeStatsEstimate.builder().setOutputRowCount(bRows).addSymbolStatistics(ImmutableMap.of(new Symbol("B1"), new SymbolStatsEstimate(0, 100, 0, 640000d * 10000, 10))).build();
    // output size does not  exceed AUTOMATIC_RESTRICTED limit
    PlanNodeStatsEstimate bSourceStatsEstimate = PlanNodeStatsEstimate.builder().setOutputRowCount(bRows).addSymbolStatistics(ImmutableMap.of(new Symbol("B1"), new SymbolStatsEstimate(0, 100, 0, 64, 10))).build();
    // immediate join sources exceeds AUTOMATIC_RESTRICTED limit but build tables are small
    // therefore replicated distribution type is chosen
    assertDetermineJoinDistributionType().setSystemProperty(JOIN_DISTRIBUTION_TYPE, JoinDistributionType.AUTOMATIC.name()).setSystemProperty(JOIN_MAX_BROADCAST_TABLE_SIZE, "100MB").overrideStats("valuesA", aStatsEstimate).overrideStats("filterB", bStatsEstimate).overrideStats("valuesB", bSourceStatsEstimate).on(p -> {
        Symbol a1 = p.symbol("A1", symbolType);
        Symbol b1 = p.symbol("B1", symbolType);
        return p.join(INNER, p.values(new PlanNodeId("valuesA"), aRows, a1), p.filter(new PlanNodeId("filterB"), TRUE_LITERAL, p.values(new PlanNodeId("valuesB"), bRows, b1)), ImmutableList.of(new JoinNode.EquiJoinClause(a1, b1)), ImmutableList.of(a1), ImmutableList.of(b1), Optional.empty());
    }).matches(join(INNER, ImmutableList.of(equiJoinClause("A1", "B1")), Optional.empty(), Optional.of(REPLICATED), values(ImmutableMap.of("A1", 0)), filter("true", values(ImmutableMap.of("B1", 0)))));
    // same but with join sides reversed
    assertDetermineJoinDistributionType().setSystemProperty(JOIN_DISTRIBUTION_TYPE, JoinDistributionType.AUTOMATIC.name()).setSystemProperty(JOIN_MAX_BROADCAST_TABLE_SIZE, "100MB").overrideStats("valuesA", aStatsEstimate).overrideStats("filterB", bStatsEstimate).overrideStats("valuesB", bSourceStatsEstimate).on(p -> {
        Symbol a1 = p.symbol("A1", symbolType);
        Symbol b1 = p.symbol("B1", symbolType);
        return p.join(INNER, p.filter(new PlanNodeId("filterB"), TRUE_LITERAL, p.values(new PlanNodeId("valuesB"), bRows, b1)), p.values(new PlanNodeId("valuesA"), aRows, a1), ImmutableList.of(new JoinNode.EquiJoinClause(b1, a1)), ImmutableList.of(b1), ImmutableList.of(a1), Optional.empty());
    }).matches(join(INNER, ImmutableList.of(equiJoinClause("A1", "B1")), Optional.empty(), Optional.of(REPLICATED), values(ImmutableMap.of("A1", 0)), filter("true", values(ImmutableMap.of("B1", 0)))));
    // only probe side (with small tables) source stats are available, join sides should be flipped
    assertDetermineJoinDistributionType().setSystemProperty(JOIN_DISTRIBUTION_TYPE, JoinDistributionType.AUTOMATIC.name()).setSystemProperty(JOIN_MAX_BROADCAST_TABLE_SIZE, "100MB").overrideStats("valuesA", PlanNodeStatsEstimate.unknown()).overrideStats("filterB", PlanNodeStatsEstimate.unknown()).overrideStats("valuesB", bSourceStatsEstimate).on(p -> {
        Symbol a1 = p.symbol("A1", symbolType);
        Symbol b1 = p.symbol("B1", symbolType);
        return p.join(LEFT, p.filter(new PlanNodeId("filterB"), TRUE_LITERAL, p.values(new PlanNodeId("valuesB"), bRows, b1)), p.values(new PlanNodeId("valuesA"), aRows, a1), ImmutableList.of(new JoinNode.EquiJoinClause(b1, a1)), ImmutableList.of(b1), ImmutableList.of(a1), Optional.empty());
    }).matches(join(RIGHT, ImmutableList.of(equiJoinClause("A1", "B1")), Optional.empty(), Optional.of(PARTITIONED), values(ImmutableMap.of("A1", 0)), filter("true", values(ImmutableMap.of("B1", 0)))));
}