Skip to content

Commit

Permalink
[fix](Nereids) producer to consumer should be multimap in cte (#39850)
Browse files Browse the repository at this point in the history
because consumer could refer multi times for one producer' slot, so
producer to consumer slot map should be multimap
  • Loading branch information
morrySnow authored Aug 23, 2024
1 parent 2c8d65b commit d3b48a4
Show file tree
Hide file tree
Showing 10 changed files with 100 additions and 64 deletions.
Original file line number Diff line number Diff line change
Expand Up @@ -71,6 +71,7 @@
import com.google.common.collect.ImmutableList;
import com.google.common.collect.Lists;
import com.google.common.collect.Maps;
import com.google.common.collect.Multimap;
import org.apache.logging.log4j.LogManager;
import org.apache.logging.log4j.Logger;

Expand Down Expand Up @@ -618,8 +619,8 @@ public Map<RelationId, Set<Expression>> getConsumerIdToFilters() {
return this.statementContext.getConsumerIdToFilters();
}

public void addCTEConsumerGroup(CTEId cteId, Group g, Map<Slot, Slot> producerSlotToConsumerSlot) {
List<Pair<Map<Slot, Slot>, Group>> consumerGroups =
public void addCTEConsumerGroup(CTEId cteId, Group g, Multimap<Slot, Slot> producerSlotToConsumerSlot) {
List<Pair<Multimap<Slot, Slot>, Group>> consumerGroups =
this.statementContext.getCteIdToConsumerGroup().computeIfAbsent(cteId, k -> new ArrayList<>());
consumerGroups.add(Pair.of(producerSlotToConsumerSlot, g));
}
Expand All @@ -628,12 +629,18 @@ public void addCTEConsumerGroup(CTEId cteId, Group g, Map<Slot, Slot> producerSl
* Update CTE consumer group as producer's stats update
*/
public void updateConsumerStats(CTEId cteId, Statistics statistics) {
List<Pair<Map<Slot, Slot>, Group>> consumerGroups = this.statementContext.getCteIdToConsumerGroup().get(cteId);
for (Pair<Map<Slot, Slot>, Group> p : consumerGroups) {
Map<Slot, Slot> producerSlotToConsumerSlot = p.first;
List<Pair<Multimap<Slot, Slot>, Group>> consumerGroups
= this.statementContext.getCteIdToConsumerGroup().get(cteId);
for (Pair<Multimap<Slot, Slot>, Group> p : consumerGroups) {
Multimap<Slot, Slot> producerSlotToConsumerSlot = p.first;
Statistics updatedConsumerStats = new StatisticsBuilder(statistics).build();
for (Entry<Expression, ColumnStatistic> entry : statistics.columnStatistics().entrySet()) {
updatedConsumerStats.addColumnStats(producerSlotToConsumerSlot.get(entry.getKey()), entry.getValue());
if (!(entry.getKey() instanceof Slot)) {
continue;
}
for (Slot consumer : producerSlotToConsumerSlot.get((Slot) entry.getKey())) {
updatedConsumerStats.addColumnStats(consumer, entry.getValue());
}
}
p.value().setStatistics(updatedConsumerStats);
}
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -54,6 +54,7 @@
import com.google.common.base.Suppliers;
import com.google.common.collect.ImmutableList;
import com.google.common.collect.Maps;
import com.google.common.collect.Multimap;
import com.google.common.collect.Sets;
import org.apache.logging.log4j.LogManager;
import org.apache.logging.log4j.Logger;
Expand Down Expand Up @@ -115,7 +116,7 @@ public class StatementContext implements Closeable {
private final Map<CTEId, Set<Slot>> cteIdToOutputIds = new HashMap<>();
private final Map<RelationId, Set<Expression>> consumerIdToFilters = new HashMap<>();
// Used to update consumer's stats
private final Map<CTEId, List<Pair<Map<Slot, Slot>, Group>>> cteIdToConsumerGroup = new HashMap<>();
private final Map<CTEId, List<Pair<Multimap<Slot, Slot>, Group>>> cteIdToConsumerGroup = new HashMap<>();
private final Map<CTEId, LogicalPlan> rewrittenCteProducer = new HashMap<>();
private final Map<CTEId, LogicalPlan> rewrittenCteConsumer = new HashMap<>();
private final Set<String> viewDdlSqlSet = Sets.newHashSet();
Expand Down Expand Up @@ -357,7 +358,7 @@ public Map<PlaceholderId, Expression> getIdToPlaceholderRealExpr() {
return idToPlaceholderRealExpr;
}

public Map<CTEId, List<Pair<Map<Slot, Slot>, Group>>> getCteIdToConsumerGroup() {
public Map<CTEId, List<Pair<Multimap<Slot, Slot>, Group>>> getCteIdToConsumerGroup() {
return cteIdToConsumerGroup;
}

Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -1205,12 +1205,9 @@ public PlanFragment visitPhysicalCTEConsumer(PhysicalCTEConsumer cteConsumer,
for (Slot producerSlot : cteProducer.getOutput()) {
SlotRef slotRef = context.findSlotRef(producerSlot.getExprId());
tupleDescriptor = slotRef.getDesc().getParent();
Slot consumerSlot = cteConsumer.getProducerToConsumerSlotMap().get(producerSlot);
// consumerSlot could be null if we prune partial consumers' columns
if (consumerSlot == null) {
continue;
for (Slot consumerSlot : cteConsumer.getProducerToConsumerSlotMap().get(producerSlot)) {
context.addExprIdSlotRefPair(consumerSlot.getExprId(), slotRef);
}
context.addExprIdSlotRefPair(consumerSlot.getExprId(), slotRef);
}
CTEScanNode cteScanNode = new CTEScanNode(tupleDescriptor);
context.getRuntimeTranslator().ifPresent(runtimeFilterTranslator ->
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -51,8 +51,10 @@
import com.google.common.base.Preconditions;
import com.google.common.collect.ImmutableList;
import com.google.common.collect.ImmutableSet;
import com.google.common.collect.LinkedHashMultimap;
import com.google.common.collect.Lists;
import com.google.common.collect.Maps;
import com.google.common.collect.Multimap;

import java.util.LinkedHashMap;
import java.util.List;
Expand Down Expand Up @@ -261,14 +263,15 @@ public Plan visitLogicalPartitionTopN(LogicalPartitionTopN<? extends Plan> parti
@Override
public Plan visitLogicalCTEConsumer(LogicalCTEConsumer cteConsumer, Map<ExprId, Slot> replaceMap) {
Map<Slot, Slot> consumerToProducerOutputMap = new LinkedHashMap<>();
Map<Slot, Slot> producerToConsumerOutputMap = new LinkedHashMap<>();
Multimap<Slot, Slot> producerToConsumerOutputMap = LinkedHashMultimap.create();
for (Slot producerOutputSlot : cteConsumer.getConsumerToProducerOutputMap().values()) {
Slot newProducerOutputSlot = updateExpression(producerOutputSlot, replaceMap);
Slot newConsumerOutputSlot = cteConsumer.getProducerToConsumerOutputMap().get(producerOutputSlot)
.withNullable(newProducerOutputSlot.nullable());
producerToConsumerOutputMap.put(newProducerOutputSlot, newConsumerOutputSlot);
consumerToProducerOutputMap.put(newConsumerOutputSlot, newProducerOutputSlot);
replaceMap.put(newConsumerOutputSlot.getExprId(), newConsumerOutputSlot);
for (Slot consumerOutputSlot : cteConsumer.getProducerToConsumerOutputMap().get(producerOutputSlot)) {
Slot newConsumerOutputSlot = consumerOutputSlot.withNullable(newProducerOutputSlot.nullable());
producerToConsumerOutputMap.put(newProducerOutputSlot, newConsumerOutputSlot);
consumerToProducerOutputMap.put(newConsumerOutputSlot, newProducerOutputSlot);
replaceMap.put(newConsumerOutputSlot.getExprId(), newConsumerOutputSlot);
}
}
return cteConsumer.withTwoMaps(consumerToProducerOutputMap, producerToConsumerOutputMap);
}
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -206,11 +206,11 @@ public Plan visitLogicalJoin(LogicalJoin<? extends Plan, ? extends Plan> join, O
private Map<Slot, Slot> constructReplaceMap(LogicalCTEConsumer leftConsumer, Map<Slot, Slot> leftCloneToLeft,
LogicalCTEConsumer rightConsumer, Map<Slot, Slot> rightCloneToRight) {
Map<Slot, Slot> replaced = new HashMap<>();
for (Entry<Slot, Slot> entry : leftConsumer.getProducerToConsumerOutputMap().entrySet()) {
replaced.put(leftCloneToLeft.get(entry.getKey()), entry.getValue());
for (Entry<Slot, Slot> entry : leftConsumer.getConsumerToProducerOutputMap().entrySet()) {
replaced.put(leftCloneToLeft.get(entry.getValue()), entry.getKey());
}
for (Entry<Slot, Slot> entry : rightConsumer.getProducerToConsumerOutputMap().entrySet()) {
replaced.put(rightCloneToRight.get(entry.getKey()), entry.getValue());
for (Entry<Slot, Slot> entry : rightConsumer.getConsumerToProducerOutputMap().entrySet()) {
replaced.put(rightCloneToRight.get(entry.getValue()), entry.getKey());
}
return replaced;
}
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -59,8 +59,10 @@

import com.google.common.collect.ImmutableList;
import com.google.common.collect.ImmutableSet;
import com.google.common.collect.LinkedHashMultimap;
import com.google.common.collect.Lists;
import com.google.common.collect.Maps;
import com.google.common.collect.Multimap;
import com.google.common.collect.Sets;

import java.util.Collections;
Expand Down Expand Up @@ -394,7 +396,7 @@ public Plan visitLogicalCTEConsumer(LogicalCTEConsumer cteConsumer, Context cont
return cteConsumer;
}
Map<Slot, Slot> consumerToProducerOutputMap = Maps.newHashMap();
Map<Slot, Slot> producerToConsumerOutputMap = Maps.newHashMap();
Multimap<Slot, Slot> producerToConsumerOutputMap = LinkedHashMultimap.create();
Map<Slot, Map<List<String>, SlotReference>> oriSlotToSubPathToSlot = Maps.newHashMap();
for (Map.Entry<Slot, Slot> consumerToProducer : cteConsumer.getConsumerToProducerOutputMap().entrySet()) {
Slot consumer = consumerToProducer.getKey();
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -63,6 +63,8 @@

import com.google.common.collect.ImmutableList;
import com.google.common.collect.ImmutableSet;
import com.google.common.collect.LinkedHashMultimap;
import com.google.common.collect.Multimap;

import java.util.LinkedHashMap;
import java.util.List;
Expand Down Expand Up @@ -397,7 +399,7 @@ public Plan visitLogicalCTEConsumer(LogicalCTEConsumer cteConsumer, DeepCopierCo
return context.getRelationReplaceMap().get(cteConsumer.getRelationId());
}
Map<Slot, Slot> consumerToProducerOutputMap = new LinkedHashMap<>();
Map<Slot, Slot> producerToConsumerOutputMap = new LinkedHashMap<>();
Multimap<Slot, Slot> producerToConsumerOutputMap = LinkedHashMultimap.create();
for (Slot consumerOutput : cteConsumer.getOutput()) {
Slot newOutput = (Slot) ExpressionDeepCopier.INSTANCE.deepCopy(consumerOutput, context);
consumerToProducerOutputMap.put(newOutput, cteConsumer.getProducerSlot(consumerOutput));
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -33,6 +33,10 @@

import com.google.common.base.Preconditions;
import com.google.common.collect.ImmutableList;
import com.google.common.collect.ImmutableMap;
import com.google.common.collect.ImmutableMultimap;
import com.google.common.collect.LinkedHashMultimap;
import com.google.common.collect.Multimap;

import java.util.LinkedHashMap;
import java.util.List;
Expand All @@ -50,20 +54,15 @@ public class LogicalCTEConsumer extends LogicalRelation implements BlockFuncDeps
private final String name;
private final CTEId cteId;
private final Map<Slot, Slot> consumerToProducerOutputMap;
private final Map<Slot, Slot> producerToConsumerOutputMap;
private final Multimap<Slot, Slot> producerToConsumerOutputMap;

/**
* Logical CTE consumer.
*/
public LogicalCTEConsumer(RelationId relationId, CTEId cteId, String name,
Map<Slot, Slot> consumerToProducerOutputMap, Map<Slot, Slot> producerToConsumerOutputMap) {
super(relationId, PlanType.LOGICAL_CTE_CONSUMER, Optional.empty(), Optional.empty());
this.cteId = Objects.requireNonNull(cteId, "cteId should not null");
this.name = Objects.requireNonNull(name, "name should not null");
this.consumerToProducerOutputMap = Objects.requireNonNull(consumerToProducerOutputMap,
"consumerToProducerOutputMap should not null");
this.producerToConsumerOutputMap = Objects.requireNonNull(producerToConsumerOutputMap,
"producerToConsumerOutputMap should not null");
Map<Slot, Slot> consumerToProducerOutputMap, Multimap<Slot, Slot> producerToConsumerOutputMap) {
this(relationId, cteId, name, consumerToProducerOutputMap, producerToConsumerOutputMap,
Optional.empty(), Optional.empty());
}

/**
Expand All @@ -73,24 +72,31 @@ public LogicalCTEConsumer(RelationId relationId, CTEId cteId, String name, Logic
super(relationId, PlanType.LOGICAL_CTE_CONSUMER, Optional.empty(), Optional.empty());
this.cteId = Objects.requireNonNull(cteId, "cteId should not null");
this.name = Objects.requireNonNull(name, "name should not null");
this.consumerToProducerOutputMap = new LinkedHashMap<>();
this.producerToConsumerOutputMap = new LinkedHashMap<>();
initOutputMaps(producerPlan);
ImmutableMap.Builder<Slot, Slot> cToPBuilder = ImmutableMap.builder();
ImmutableMultimap.Builder<Slot, Slot> pToCBuilder = ImmutableMultimap.builder();
List<Slot> producerOutput = producerPlan.getOutput();
for (Slot producerOutputSlot : producerOutput) {
Slot consumerSlot = generateConsumerSlot(this.name, producerOutputSlot);
cToPBuilder.put(consumerSlot, producerOutputSlot);
pToCBuilder.put(producerOutputSlot, consumerSlot);
}
consumerToProducerOutputMap = cToPBuilder.build();
producerToConsumerOutputMap = pToCBuilder.build();
}

/**
* Logical CTE consumer.
*/
public LogicalCTEConsumer(RelationId relationId, CTEId cteId, String name,
Map<Slot, Slot> consumerToProducerOutputMap, Map<Slot, Slot> producerToConsumerOutputMap,
Map<Slot, Slot> consumerToProducerOutputMap, Multimap<Slot, Slot> producerToConsumerOutputMap,
Optional<GroupExpression> groupExpression, Optional<LogicalProperties> logicalProperties) {
super(relationId, PlanType.LOGICAL_CTE_CONSUMER, groupExpression, logicalProperties);
this.cteId = Objects.requireNonNull(cteId, "cteId should not null");
this.name = Objects.requireNonNull(name, "name should not null");
this.consumerToProducerOutputMap = Objects.requireNonNull(consumerToProducerOutputMap,
"consumerToProducerOutputMap should not null");
this.producerToConsumerOutputMap = Objects.requireNonNull(producerToConsumerOutputMap,
"producerToConsumerOutputMap should not null");
this.consumerToProducerOutputMap = ImmutableMap.copyOf(Objects.requireNonNull(consumerToProducerOutputMap,
"consumerToProducerOutputMap should not null"));
this.producerToConsumerOutputMap = ImmutableMultimap.copyOf(Objects.requireNonNull(producerToConsumerOutputMap,
"producerToConsumerOutputMap should not null"));
}

/**
Expand All @@ -107,20 +113,11 @@ public static SlotReference generateConsumerSlot(String cteName, Slot producerOu
slotRef != null ? Optional.of(slotRef.getInternalName()) : Optional.empty());
}

private void initOutputMaps(LogicalPlan childPlan) {
List<Slot> producerOutput = childPlan.getOutput();
for (Slot producerOutputSlot : producerOutput) {
Slot consumerSlot = generateConsumerSlot(this.name, producerOutputSlot);
producerToConsumerOutputMap.put(producerOutputSlot, consumerSlot);
consumerToProducerOutputMap.put(consumerSlot, producerOutputSlot);
}
}

public Map<Slot, Slot> getConsumerToProducerOutputMap() {
return consumerToProducerOutputMap;
}

public Map<Slot, Slot> getProducerToConsumerOutputMap() {
public Multimap<Slot, Slot> getProducerToConsumerOutputMap() {
return producerToConsumerOutputMap;
}

Expand All @@ -129,7 +126,8 @@ public <R, C> R accept(PlanVisitor<R, C> visitor, C context) {
return visitor.visitLogicalCTEConsumer(this, context);
}

public Plan withTwoMaps(Map<Slot, Slot> consumerToProducerOutputMap, Map<Slot, Slot> producerToConsumerOutputMap) {
public Plan withTwoMaps(Map<Slot, Slot> consumerToProducerOutputMap,
Multimap<Slot, Slot> producerToConsumerOutputMap) {
return new LogicalCTEConsumer(relationId, cteId, name,
consumerToProducerOutputMap, producerToConsumerOutputMap);
}
Expand Down Expand Up @@ -162,7 +160,8 @@ public List<Slot> computeOutput() {
@Override
public Plan pruneOutputs(List<NamedExpression> prunedOutputs) {
Map<Slot, Slot> consumerToProducerOutputMap = new LinkedHashMap<>(this.consumerToProducerOutputMap.size());
Map<Slot, Slot> producerToConsumerOutputMap = new LinkedHashMap<>(this.consumerToProducerOutputMap.size());
Multimap<Slot, Slot> producerToConsumerOutputMap = LinkedHashMultimap.create(
this.consumerToProducerOutputMap.size(), this.consumerToProducerOutputMap.size());
for (Entry<Slot, Slot> consumerToProducerSlot : this.consumerToProducerOutputMap.entrySet()) {
if (prunedOutputs.contains(consumerToProducerSlot.getKey())) {
consumerToProducerOutputMap.put(consumerToProducerSlot.getKey(), consumerToProducerSlot.getValue());
Expand Down
Loading

0 comments on commit d3b48a4

Please sign in to comment.