use of io.trino.sql.planner.plan.AggregationNode.Step.SINGLE in project trino by trinodb.
the class DecorrelateInnerUnnestWithGlobalAggregation method apply.
@Override
public Result apply(CorrelatedJoinNode correlatedJoinNode, Captures captures, Context context) {
// find global aggregation in subquery
List<PlanNode> globalAggregations = PlanNodeSearcher.searchFrom(correlatedJoinNode.getSubquery(), context.getLookup()).where(DecorrelateInnerUnnestWithGlobalAggregation::isGlobalAggregation).recurseOnlyWhen(node -> node instanceof ProjectNode || isGlobalAggregation(node)).findAll();
if (globalAggregations.isEmpty()) {
return Result.empty();
}
// if there are multiple global aggregations, the one that is closest to the source is the "reducing" aggregation, because it reduces multiple input rows to single output row
AggregationNode reducingAggregation = (AggregationNode) globalAggregations.get(globalAggregations.size() - 1);
// find unnest in subquery
Optional<UnnestNode> subqueryUnnest = PlanNodeSearcher.searchFrom(reducingAggregation.getSource(), context.getLookup()).where(node -> isSupportedUnnest(node, correlatedJoinNode.getCorrelation(), context.getLookup())).recurseOnlyWhen(node -> node instanceof ProjectNode || isGroupedAggregation(node)).findFirst();
if (subqueryUnnest.isEmpty()) {
return Result.empty();
}
UnnestNode unnestNode = subqueryUnnest.get();
// assign unique id to input rows to restore semantics of aggregations after rewrite
PlanNode input = new AssignUniqueId(context.getIdAllocator().getNextId(), correlatedJoinNode.getInput(), context.getSymbolAllocator().newSymbol("unique", BIGINT));
// 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());
}
// rewrite correlated join to UnnestNode
Symbol ordinalitySymbol = unnestNode.getOrdinalitySymbol().orElseGet(() -> context.getSymbolAllocator().newSymbol("ordinality", BIGINT));
UnnestNode rewrittenUnnest = new UnnestNode(context.getIdAllocator().getNextId(), input, input.getOutputSymbols(), unnestNode.getMappings(), Optional.of(ordinalitySymbol), LEFT, Optional.empty());
// append mask symbol based on ordinality to distinguish between the unnested rows and synthetic null rows
Symbol mask = context.getSymbolAllocator().newSymbol("mask", BOOLEAN);
ProjectNode sourceWithMask = new ProjectNode(context.getIdAllocator().getNextId(), rewrittenUnnest, Assignments.builder().putIdentities(rewrittenUnnest.getOutputSymbols()).put(mask, new IsNotNullPredicate(ordinalitySymbol.toSymbolReference())).build());
// restore all projections, grouped aggregations and global aggregations from the subquery
PlanNode result = rewriteNodeSequence(context.getLookup().resolve(correlatedJoinNode.getSubquery()), input.getOutputSymbols(), mask, sourceWithMask, reducingAggregation.getId(), unnestNode.getId(), context.getSymbolAllocator(), context.getIdAllocator(), context.getLookup());
// restrict outputs
return Result.ofPlanNode(restrictOutputs(context.getIdAllocator(), result, ImmutableSet.copyOf(correlatedJoinNode.getOutputSymbols())).orElse(result));
}
use of io.trino.sql.planner.plan.AggregationNode.Step.SINGLE in project trino by trinodb.
the class SingleDistinctAggregationToGroupBy method apply.
@Override
public Result apply(AggregationNode aggregation, Captures captures, Context context) {
List<Set<Expression>> argumentSets = extractArgumentSets(aggregation).collect(Collectors.toList());
Set<Symbol> symbols = Iterables.getOnlyElement(argumentSets).stream().map(Symbol::from).collect(Collectors.toSet());
return Result.ofPlanNode(new AggregationNode(aggregation.getId(), new AggregationNode(context.getIdAllocator().getNextId(), aggregation.getSource(), ImmutableMap.of(), singleGroupingSet(ImmutableList.<Symbol>builder().addAll(aggregation.getGroupingKeys()).addAll(symbols).build()), ImmutableList.of(), SINGLE, Optional.empty(), Optional.empty()), // remove DISTINCT flag from function calls
aggregation.getAggregations().entrySet().stream().collect(Collectors.toMap(Map.Entry::getKey, e -> removeDistinct(e.getValue()))), aggregation.getGroupingSets(), emptyList(), aggregation.getStep(), aggregation.getHashSymbol(), aggregation.getGroupIdSymbol()));
}
use of io.trino.sql.planner.plan.AggregationNode.Step.SINGLE in project trino by trinodb.
the class TransformFilteringSemiJoinToInnerJoin method apply.
@Override
public Result apply(FilterNode filterNode, Captures captures, Context context) {
SemiJoinNode semiJoin = captures.get(SEMI_JOIN);
// Do not transform semi-join in context of DELETE
if (PlanNodeSearcher.searchFrom(semiJoin.getSource(), context.getLookup()).where(node -> node instanceof TableScanNode && ((TableScanNode) node).isUpdateTarget()).matches()) {
return Result.empty();
}
Symbol semiJoinSymbol = semiJoin.getSemiJoinOutput();
Predicate<Expression> isSemiJoinSymbol = expression -> expression.equals(semiJoinSymbol.toSymbolReference());
List<Expression> conjuncts = extractConjuncts(filterNode.getPredicate());
if (conjuncts.stream().noneMatch(isSemiJoinSymbol)) {
return Result.empty();
}
Expression filteredPredicate = and(conjuncts.stream().filter(not(isSemiJoinSymbol)).collect(toImmutableList()));
Expression simplifiedPredicate = inlineSymbols(symbol -> {
if (symbol.equals(semiJoinSymbol)) {
return TRUE_LITERAL;
}
return symbol.toSymbolReference();
}, filteredPredicate);
Optional<Expression> joinFilter = simplifiedPredicate.equals(TRUE_LITERAL) ? Optional.empty() : Optional.of(simplifiedPredicate);
PlanNode filteringSourceDistinct = new AggregationNode(context.getIdAllocator().getNextId(), semiJoin.getFilteringSource(), ImmutableMap.of(), singleGroupingSet(ImmutableList.of(semiJoin.getFilteringSourceJoinSymbol())), ImmutableList.of(), SINGLE, Optional.empty(), Optional.empty());
JoinNode innerJoin = new JoinNode(semiJoin.getId(), INNER, semiJoin.getSource(), filteringSourceDistinct, ImmutableList.of(new EquiJoinClause(semiJoin.getSourceJoinSymbol(), semiJoin.getFilteringSourceJoinSymbol())), semiJoin.getSource().getOutputSymbols(), ImmutableList.of(), false, joinFilter, Optional.empty(), Optional.empty(), Optional.empty(), Optional.empty(), semiJoin.getDynamicFilterId().map(id -> ImmutableMap.of(id, semiJoin.getFilteringSourceJoinSymbol())).orElse(ImmutableMap.of()), Optional.empty());
ProjectNode project = new ProjectNode(context.getIdAllocator().getNextId(), innerJoin, Assignments.builder().putIdentities(innerJoin.getOutputSymbols()).put(semiJoinSymbol, TRUE_LITERAL).build());
return Result.ofPlanNode(project);
}
use of io.trino.sql.planner.plan.AggregationNode.Step.SINGLE in project trino by trinodb.
the class TestHistogram method testManyValuesInducingRehash.
private static void testManyValuesInducingRehash(TestingAggregationFunction aggregationFunction) {
double distinctFraction = 0.1f;
int numGroups = 50000;
int itemCount = 30;
Random random = new Random();
GroupedAggregator groupedAggregator = aggregationFunction.createAggregatorFactory(SINGLE, ImmutableList.of(0), OptionalInt.empty()).createGroupedAggregator();
for (int j = 0; j < numGroups; j++) {
Map<String, Long> expectedValues = new HashMap<>();
List<String> valueList = new ArrayList<>();
for (int i = 0; i < itemCount; i++) {
String str = String.valueOf(i % 10);
String item = IntStream.range(0, itemCount).mapToObj(x -> str).collect(Collectors.joining());
boolean distinctValue = random.nextDouble() < distinctFraction;
if (distinctValue) {
// produce a unique value for the histogram
item = j + "-" + item;
valueList.add(item);
} else {
valueList.add(item);
}
expectedValues.compute(item, (k, v) -> v == null ? 1L : ++v);
}
Block block = createStringsBlock(valueList);
AggregationTestInputBuilder testInputBuilder = new AggregationTestInputBuilder(new Block[] { block }, aggregationFunction);
AggregationTestInput test1 = testInputBuilder.build();
test1.runPagesOnAggregatorWithAssertion(j, aggregationFunction.getFinalType(), groupedAggregator, new AggregationTestOutput(expectedValues));
}
}
use of io.trino.sql.planner.plan.AggregationNode.Step.SINGLE in project trino by trinodb.
the class TestValidateStreamingAggregations method testValidateSuccessful.
@Test
public void testValidateSuccessful() {
validatePlan(p -> p.aggregation(a -> a.step(SINGLE).singleGroupingSet(p.symbol("nationkey")).source(p.tableScan(nationTableHandle, ImmutableList.of(p.symbol("nationkey", BIGINT)), ImmutableMap.of(p.symbol("nationkey", BIGINT), new TpchColumnHandle("nationkey", BIGINT))))));
validatePlan(p -> p.aggregation(a -> a.step(SINGLE).singleGroupingSet(p.symbol("unique"), p.symbol("nationkey")).preGroupedSymbols(p.symbol("unique"), p.symbol("nationkey")).source(p.assignUniqueId(p.symbol("unique"), p.tableScan(nationTableHandle, ImmutableList.of(p.symbol("nationkey", BIGINT)), ImmutableMap.of(p.symbol("nationkey", BIGINT), new TpchColumnHandle("nationkey", BIGINT)))))));
}
Aggregations