github-actions[bot] commented on code in PR #64559: URL: https://github.com/apache/doris/pull/64559#discussion_r3459946199
########## fe/fe-core/src/main/java/org/apache/doris/nereids/stats/MemoStatsAndCostRecomputer.java: ########## @@ -0,0 +1,836 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to you under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, +// software distributed under the License is distributed on an +// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, either express or implied. See the License for the +// specific language governing permissions and limitations +// under the License. + +package org.apache.doris.nereids.stats; + +import org.apache.doris.common.Pair; +import org.apache.doris.nereids.CascadesContext; +import org.apache.doris.nereids.cost.Cost; +import org.apache.doris.nereids.cost.CostCalculator; +import org.apache.doris.nereids.memo.Group; +import org.apache.doris.nereids.memo.GroupExpression; +import org.apache.doris.nereids.properties.PhysicalProperties; +import org.apache.doris.nereids.trees.expressions.CTEId; +import org.apache.doris.nereids.trees.plans.Plan; +import org.apache.doris.nereids.trees.plans.algebra.Join; +import org.apache.doris.nereids.trees.plans.logical.LogicalCTEConsumer; +import org.apache.doris.nereids.trees.plans.logical.LogicalCTEProducer; +import org.apache.doris.nereids.trees.plans.logical.LogicalProject; +import org.apache.doris.nereids.trees.plans.physical.PhysicalCTEConsumer; +import org.apache.doris.nereids.trees.plans.physical.PhysicalCTEProducer; +import org.apache.doris.nereids.trees.plans.physical.PhysicalProject; +import org.apache.doris.qe.ConnectContext; +import org.apache.doris.statistics.Statistics; + +import com.google.common.collect.Lists; + +import java.util.ArrayList; +import java.util.Collections; +import java.util.HashMap; +import java.util.HashSet; +import java.util.LinkedHashMap; +import java.util.List; +import java.util.Map; +import java.util.Optional; +import java.util.Set; + +/** + * Re-estimate memo logical row counts and rebuild physical costs. + * and rebuild physical cost state. + */ +public final class MemoStatsAndCostRecomputer { + private static final double CHOSEN_PROJECT_STATS_DIVERGENCE_RATIO_THRESHOLD = 1_000D; + private final CascadesContext cascadesContext; + private final Map<CTEId, Statistics> cteIdToStats = new HashMap<>(); + private final LogicalExpressionRowCountSyncPolicy logicalExpressionRowCountSyncPolicy; + + private MemoStatsAndCostRecomputer(CascadesContext cascadesContext, + LogicalExpressionRowCountSyncPolicy logicalExpressionRowCountSyncPolicy) { + this.cascadesContext = cascadesContext; + this.logicalExpressionRowCountSyncPolicy = logicalExpressionRowCountSyncPolicy; + } + + /** + * recompute + */ + public static void recompute(Group rootGroup, PhysicalProperties physicalProperties, + CascadesContext cascadesContext) { + recompute(rootGroup, physicalProperties, cascadesContext, + LogicalExpressionRowCountSyncPolicy.KEEP_INDIVIDUAL_EXPRESSION_ROW_COUNT); + } + + /** + * recompute with configurable logical expression row count sync behavior. + */ + public static void recompute(Group rootGroup, PhysicalProperties physicalProperties, + CascadesContext cascadesContext, + LogicalExpressionRowCountSyncPolicy logicalExpressionRowCountSyncPolicy) { + MemoStatsAndCostRecomputer recomputer = new MemoStatsAndCostRecomputer(cascadesContext, + logicalExpressionRowCountSyncPolicy); + recomputer.seedProducerStats(rootGroup, new HashSet<>()); + recomputer.reestimateLogicalStatsBottomUp(rootGroup, new HashSet<>()); + // Run a second pass so CTE consumers and their ancestors can settle on producer stats refreshed above. + recomputer.reestimateLogicalStatsBottomUp(rootGroup, new HashSet<>()); + recomputer.recomputePhysicalCostsBottomUp(rootGroup, new HashSet<>()); + } + + private void seedProducerStats(Group group, Set<Group> visited) { + if (!visited.add(group)) { + return; + } + Statistics statistics = group.getStatistics(); + if (statistics != null) { + recordProducerStats(group, statistics); + } + for (Group child : getTraversalChildren(group)) { + seedProducerStats(child, visited); + } + } + + private void reestimateLogicalStatsBottomUp(Group group, Set<Group> visited) { + if (!visited.add(group)) { + return; + } + for (Group child : getTraversalChildren(group)) { + reestimateLogicalStatsBottomUp(child, visited); + } + reestimateCurrentGroup(group); + refreshEnforcerRowCount(group); + } + + private void reestimateCurrentGroup(Group group) { + List<GroupExpression> estimableExpressions = getEstimableLogicalExpressions(group); + if (estimableExpressions.isEmpty()) { + if (group.getLogicalExpressions().isEmpty()) { + reestimatePhysicalOnlyGroup(group); + } + return; + } + Statistics originalStatistics = group.getStatistics(); + boolean originalStatsReliable = group.isStatsReliable(); + Map<GroupExpression, Statistics> candidateStatisticsByExpression = new LinkedHashMap<>(); + Map<GroupExpression, Boolean> candidateStatsReliableByExpression = new LinkedHashMap<>(); + for (GroupExpression logicalExpression : estimableExpressions) { + List<Statistics> originalChildStatistics = replaceChildStatisticsForLogicalEstimation(logicalExpression); + group.setStatistics(null); + try { + estimateStats(logicalExpression); + } finally { + restoreChildStatistics(logicalExpression, originalChildStatistics); + } + Statistics estimatedStatistics = group.getStatistics(); + if (estimatedStatistics == null || !isValidCandidateStatistics(estimatedStatistics)) { + continue; + } + logicalExpression.setEstOutputRowCount(estimatedStatistics.getRowCount()); + candidateStatisticsByExpression.put(logicalExpression, new Statistics(estimatedStatistics)); + candidateStatsReliableByExpression.put(logicalExpression, group.isStatsReliable()); + } + if (candidateStatisticsByExpression.isEmpty()) { + group.setStatistics(originalStatistics); + group.setStatsReliable(originalStatsReliable); + return; + } + LogicalRowCountAggregationPolicy aggregationPolicy = getLogicalRowCountAggregationPolicy(); + Map<GroupExpression, Statistics> selectedCandidateStatisticsByExpression = filterCandidateStatisticsByPolicy( + aggregationPolicy, candidateStatisticsByExpression); + List<Statistics> candidateStatistics = new ArrayList<>(selectedCandidateStatisticsByExpression.values()); + double aggregatedRowCount = aggregationPolicy.aggregate(candidateStatistics); + Statistics updatedStatistics = resolveUpdatedGroupStatistics(group, selectedCandidateStatisticsByExpression, + candidateStatistics, aggregatedRowCount, originalStatistics); + boolean resolvedStatsReliable = resolveUpdatedGroupStatsReliability(group, + selectedCandidateStatisticsByExpression, candidateStatsReliableByExpression, + aggregatedRowCount); + group.setStatsReliable(resolvedStatsReliable); + group.setStatistics(updatedStatistics); + repairInvalidLogicalExpressionRowCounts(group, aggregatedRowCount); + refreshPhysicalExpressionRowCount(group, updatedStatistics.getRowCount()); + recordProducerStats(group, updatedStatistics); + if (shouldSyncLogicalExpressionRowCount()) { + syncLogicalExpressionRowCount(group, updatedStatistics.getRowCount()); + } + } + + private void reestimatePhysicalOnlyGroup(Group group) { + List<GroupExpression> estimableExpressions = getEstimablePhysicalExpressions(group); + if (estimableExpressions.isEmpty()) { + return; + } + Statistics originalStatistics = group.getStatistics(); + boolean originalStatsReliable = group.isStatsReliable(); + Map<GroupExpression, Statistics> candidateStatisticsByExpression = new LinkedHashMap<>(); + Map<GroupExpression, Boolean> candidateStatsReliableByExpression = new LinkedHashMap<>(); + for (GroupExpression physicalExpression : estimableExpressions) { + group.setStatistics(null); + estimateStats(physicalExpression); + Statistics estimatedStatistics = group.getStatistics(); + if (estimatedStatistics == null || !isValidCandidateStatistics(estimatedStatistics)) { + continue; + } + physicalExpression.setEstOutputRowCount(estimatedStatistics.getRowCount()); + candidateStatisticsByExpression.put(physicalExpression, new Statistics(estimatedStatistics)); + candidateStatsReliableByExpression.put(physicalExpression, group.isStatsReliable()); + } + if (candidateStatisticsByExpression.isEmpty()) { + group.setStatistics(originalStatistics); + group.setStatsReliable(originalStatsReliable); + return; + } + Statistics updatedStatistics = choosePhysicalOnlyGroupStatistics(group, candidateStatisticsByExpression, + originalStatistics); + boolean resolvedStatsReliable = resolvePhysicalOnlyGroupStatsReliability(group, + candidateStatisticsByExpression, candidateStatsReliableByExpression, + originalStatistics); + group.setStatsReliable(resolvedStatsReliable); + group.setStatistics(updatedStatistics); + refreshPhysicalExpressionRowCount(group, updatedStatistics.getRowCount()); + recordProducerStats(group, updatedStatistics); + } + + private boolean isValidCandidateStatistics(Statistics statistics) { + return Double.isFinite(statistics.getRowCount()) && statistics.getRowCount() >= 0; + } + + private void estimateStats(GroupExpression groupExpression) { + ConnectContext connectContext = cascadesContext.getConnectContext(); + StatsCalculator statsCalculator = new StatsCalculator( + groupExpression, + connectContext.getSessionVariable().getForbidUnknownColStats(), + connectContext.getTotalColumnStatisticMap(), + connectContext.getSessionVariable().isPlayNereidsDump(), + cteIdToStats, + cascadesContext); + statsCalculator.estimate(); + } + + private List<Statistics> replaceChildStatisticsForLogicalEstimation(GroupExpression logicalExpression) { Review Comment: `replaceChildStatisticsForLogicalEstimation()` is still a no-op, so each logical alternative in a reduced DPHyp group is re-estimated from the child groups' current aggregate statistics rather than from the child alternatives this expression actually uses. For example, if `Group{A,B}` contains both an `A join B` alternative with a small row count and a `B join A` alternative with a large row count, a parent `LogicalJoin(Group{A,B}, C)` will call `StatsCalculator.visitLogicalJoin()` and read the same group-level child stats for both candidates. That makes the recomputed parent stats and physical costs independent of the actual subjoin chosen by the DPHyp alternative, so `copyOutBestLogicalPlan()` can still select using statistics from a different join order. Please preserve per-`GroupExpression` statistics and restore the matching child stats while estimating each parent candidate, or otherwise recurse through the selected child expression when rebuilding candidate stats. ########## fe/fe-core/src/main/java/org/apache/doris/nereids/stats/MemoStatsAndCostRecomputer.java: ########## @@ -0,0 +1,836 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to you under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, +// software distributed under the License is distributed on an +// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, either express or implied. See the License for the +// specific language governing permissions and limitations +// under the License. + +package org.apache.doris.nereids.stats; + +import org.apache.doris.common.Pair; +import org.apache.doris.nereids.CascadesContext; +import org.apache.doris.nereids.cost.Cost; +import org.apache.doris.nereids.cost.CostCalculator; +import org.apache.doris.nereids.memo.Group; +import org.apache.doris.nereids.memo.GroupExpression; +import org.apache.doris.nereids.properties.PhysicalProperties; +import org.apache.doris.nereids.trees.expressions.CTEId; +import org.apache.doris.nereids.trees.plans.Plan; +import org.apache.doris.nereids.trees.plans.algebra.Join; +import org.apache.doris.nereids.trees.plans.logical.LogicalCTEConsumer; +import org.apache.doris.nereids.trees.plans.logical.LogicalCTEProducer; +import org.apache.doris.nereids.trees.plans.logical.LogicalProject; +import org.apache.doris.nereids.trees.plans.physical.PhysicalCTEConsumer; +import org.apache.doris.nereids.trees.plans.physical.PhysicalCTEProducer; +import org.apache.doris.nereids.trees.plans.physical.PhysicalProject; +import org.apache.doris.qe.ConnectContext; +import org.apache.doris.statistics.Statistics; + +import com.google.common.collect.Lists; + +import java.util.ArrayList; +import java.util.Collections; +import java.util.HashMap; +import java.util.HashSet; +import java.util.LinkedHashMap; +import java.util.List; +import java.util.Map; +import java.util.Optional; +import java.util.Set; + +/** + * Re-estimate memo logical row counts and rebuild physical costs. + * and rebuild physical cost state. + */ +public final class MemoStatsAndCostRecomputer { + private static final double CHOSEN_PROJECT_STATS_DIVERGENCE_RATIO_THRESHOLD = 1_000D; + private final CascadesContext cascadesContext; + private final Map<CTEId, Statistics> cteIdToStats = new HashMap<>(); + private final LogicalExpressionRowCountSyncPolicy logicalExpressionRowCountSyncPolicy; + + private MemoStatsAndCostRecomputer(CascadesContext cascadesContext, + LogicalExpressionRowCountSyncPolicy logicalExpressionRowCountSyncPolicy) { + this.cascadesContext = cascadesContext; + this.logicalExpressionRowCountSyncPolicy = logicalExpressionRowCountSyncPolicy; + } + + /** + * recompute + */ + public static void recompute(Group rootGroup, PhysicalProperties physicalProperties, + CascadesContext cascadesContext) { + recompute(rootGroup, physicalProperties, cascadesContext, + LogicalExpressionRowCountSyncPolicy.KEEP_INDIVIDUAL_EXPRESSION_ROW_COUNT); + } + + /** + * recompute with configurable logical expression row count sync behavior. + */ + public static void recompute(Group rootGroup, PhysicalProperties physicalProperties, + CascadesContext cascadesContext, + LogicalExpressionRowCountSyncPolicy logicalExpressionRowCountSyncPolicy) { + MemoStatsAndCostRecomputer recomputer = new MemoStatsAndCostRecomputer(cascadesContext, + logicalExpressionRowCountSyncPolicy); + recomputer.seedProducerStats(rootGroup, new HashSet<>()); + recomputer.reestimateLogicalStatsBottomUp(rootGroup, new HashSet<>()); + // Run a second pass so CTE consumers and their ancestors can settle on producer stats refreshed above. + recomputer.reestimateLogicalStatsBottomUp(rootGroup, new HashSet<>()); + recomputer.recomputePhysicalCostsBottomUp(rootGroup, new HashSet<>()); + } + + private void seedProducerStats(Group group, Set<Group> visited) { + if (!visited.add(group)) { + return; + } + Statistics statistics = group.getStatistics(); + if (statistics != null) { + recordProducerStats(group, statistics); + } + for (Group child : getTraversalChildren(group)) { + seedProducerStats(child, visited); + } + } + + private void reestimateLogicalStatsBottomUp(Group group, Set<Group> visited) { + if (!visited.add(group)) { + return; + } + for (Group child : getTraversalChildren(group)) { + reestimateLogicalStatsBottomUp(child, visited); + } + reestimateCurrentGroup(group); + refreshEnforcerRowCount(group); + } + + private void reestimateCurrentGroup(Group group) { + List<GroupExpression> estimableExpressions = getEstimableLogicalExpressions(group); + if (estimableExpressions.isEmpty()) { + if (group.getLogicalExpressions().isEmpty()) { + reestimatePhysicalOnlyGroup(group); + } + return; + } + Statistics originalStatistics = group.getStatistics(); + boolean originalStatsReliable = group.isStatsReliable(); + Map<GroupExpression, Statistics> candidateStatisticsByExpression = new LinkedHashMap<>(); + Map<GroupExpression, Boolean> candidateStatsReliableByExpression = new LinkedHashMap<>(); + for (GroupExpression logicalExpression : estimableExpressions) { + List<Statistics> originalChildStatistics = replaceChildStatisticsForLogicalEstimation(logicalExpression); + group.setStatistics(null); + try { + estimateStats(logicalExpression); + } finally { + restoreChildStatistics(logicalExpression, originalChildStatistics); + } + Statistics estimatedStatistics = group.getStatistics(); + if (estimatedStatistics == null || !isValidCandidateStatistics(estimatedStatistics)) { + continue; + } + logicalExpression.setEstOutputRowCount(estimatedStatistics.getRowCount()); + candidateStatisticsByExpression.put(logicalExpression, new Statistics(estimatedStatistics)); + candidateStatsReliableByExpression.put(logicalExpression, group.isStatsReliable()); + } + if (candidateStatisticsByExpression.isEmpty()) { + group.setStatistics(originalStatistics); + group.setStatsReliable(originalStatsReliable); + return; + } + LogicalRowCountAggregationPolicy aggregationPolicy = getLogicalRowCountAggregationPolicy(); + Map<GroupExpression, Statistics> selectedCandidateStatisticsByExpression = filterCandidateStatisticsByPolicy( + aggregationPolicy, candidateStatisticsByExpression); + List<Statistics> candidateStatistics = new ArrayList<>(selectedCandidateStatisticsByExpression.values()); + double aggregatedRowCount = aggregationPolicy.aggregate(candidateStatistics); + Statistics updatedStatistics = resolveUpdatedGroupStatistics(group, selectedCandidateStatisticsByExpression, + candidateStatistics, aggregatedRowCount, originalStatistics); + boolean resolvedStatsReliable = resolveUpdatedGroupStatsReliability(group, + selectedCandidateStatisticsByExpression, candidateStatsReliableByExpression, + aggregatedRowCount); + group.setStatsReliable(resolvedStatsReliable); + group.setStatistics(updatedStatistics); + repairInvalidLogicalExpressionRowCounts(group, aggregatedRowCount); + refreshPhysicalExpressionRowCount(group, updatedStatistics.getRowCount()); + recordProducerStats(group, updatedStatistics); + if (shouldSyncLogicalExpressionRowCount()) { + syncLogicalExpressionRowCount(group, updatedStatistics.getRowCount()); + } + } + + private void reestimatePhysicalOnlyGroup(Group group) { + List<GroupExpression> estimableExpressions = getEstimablePhysicalExpressions(group); + if (estimableExpressions.isEmpty()) { + return; + } + Statistics originalStatistics = group.getStatistics(); + boolean originalStatsReliable = group.isStatsReliable(); + Map<GroupExpression, Statistics> candidateStatisticsByExpression = new LinkedHashMap<>(); + Map<GroupExpression, Boolean> candidateStatsReliableByExpression = new LinkedHashMap<>(); + for (GroupExpression physicalExpression : estimableExpressions) { + group.setStatistics(null); + estimateStats(physicalExpression); + Statistics estimatedStatistics = group.getStatistics(); + if (estimatedStatistics == null || !isValidCandidateStatistics(estimatedStatistics)) { + continue; + } + physicalExpression.setEstOutputRowCount(estimatedStatistics.getRowCount()); + candidateStatisticsByExpression.put(physicalExpression, new Statistics(estimatedStatistics)); + candidateStatsReliableByExpression.put(physicalExpression, group.isStatsReliable()); + } + if (candidateStatisticsByExpression.isEmpty()) { + group.setStatistics(originalStatistics); + group.setStatsReliable(originalStatsReliable); + return; + } + Statistics updatedStatistics = choosePhysicalOnlyGroupStatistics(group, candidateStatisticsByExpression, + originalStatistics); + boolean resolvedStatsReliable = resolvePhysicalOnlyGroupStatsReliability(group, + candidateStatisticsByExpression, candidateStatsReliableByExpression, + originalStatistics); + group.setStatsReliable(resolvedStatsReliable); + group.setStatistics(updatedStatistics); + refreshPhysicalExpressionRowCount(group, updatedStatistics.getRowCount()); + recordProducerStats(group, updatedStatistics); + } + + private boolean isValidCandidateStatistics(Statistics statistics) { + return Double.isFinite(statistics.getRowCount()) && statistics.getRowCount() >= 0; + } + + private void estimateStats(GroupExpression groupExpression) { + ConnectContext connectContext = cascadesContext.getConnectContext(); + StatsCalculator statsCalculator = new StatsCalculator( + groupExpression, + connectContext.getSessionVariable().getForbidUnknownColStats(), + connectContext.getTotalColumnStatisticMap(), + connectContext.getSessionVariable().isPlayNereidsDump(), + cteIdToStats, + cascadesContext); + statsCalculator.estimate(); + } + + private List<Statistics> replaceChildStatisticsForLogicalEstimation(GroupExpression logicalExpression) { + return Collections.emptyList(); + } + + private void restoreChildStatistics(GroupExpression logicalExpression, List<Statistics> originalChildStatistics) { + if (originalChildStatistics.size() != logicalExpression.arity()) { + return; + } + for (int i = 0; i < logicalExpression.arity(); i++) { + logicalExpression.child(i).setStatistics(originalChildStatistics.get(i)); + } + } + + private void recomputePhysicalCostsBottomUp(Group group, Set<Group> visited) { + if (!visited.add(group)) { + return; + } + for (Group child : getTraversalChildren(group)) { + recomputePhysicalCostsBottomUp(child, visited); + } + if (group.getStatistics() == null + || (group.getPhysicalExpressions().isEmpty() && group.getEnforcers().isEmpty())) { + refreshEnforcerRowCount(group); + return; + } + Map<PhysicalProperties, Pair<Cost, GroupExpression>> originalLowestCostPlans = + snapshotLowestCostPlans(group); + group.clearLowestCostPlans(); + for (GroupExpression physicalExpression : group.getPhysicalExpressions()) { + recomputeGroupExpressionCost(group, physicalExpression); + } + refreshEnforcerRowCount(group); + for (GroupExpression enforcer : group.getEnforcers().values()) { + recomputeGroupExpressionCost(group, enforcer); + } + restoreMissingLowestCostPlans(group, originalLowestCostPlans); + } + + private void recomputeGroupExpressionCost(Group ownerGroup, GroupExpression groupExpression) { + if (ownerGroup.getStatistics() == null || !hasCompleteChildStatistics(groupExpression)) { + return; + } + Cost originalCost = groupExpression.getCost(); + Map<PhysicalProperties, Pair<Cost, List<PhysicalProperties>>> originalLowestCostTable + = new LinkedHashMap<>(groupExpression.getLowestCostTable()); + Map<PhysicalProperties, PhysicalProperties> originalRequestPropertiesMap + = new LinkedHashMap<>(groupExpression.getRequestPropertiesMap()); + groupExpression.clearCostState(); + + Cost bestNodeCost = null; + for (Map.Entry<PhysicalProperties, Pair<Cost, List<PhysicalProperties>>> entry + : originalLowestCostTable.entrySet()) { + PhysicalProperties outputProperties = entry.getKey(); + List<PhysicalProperties> childInputProperties = entry.getValue().second; + if (!hasAvailableChildBestPlan(groupExpression, childInputProperties)) { + continue; + } + Cost nodeCost = CostCalculator.calculateCost(cascadesContext.getConnectContext(), + groupExpression, childInputProperties); + Cost totalCost = nodeCost; + for (int i = 0; i < childInputProperties.size(); i++) { + Optional<Pair<Cost, GroupExpression>> childBestPlan = groupExpression.child(i) + .getLowestCostPlan(childInputProperties.get(i)); + if (!childBestPlan.isPresent()) { + totalCost = null; + break; + } + totalCost = CostCalculator.addChildCost(cascadesContext.getConnectContext(), + groupExpression.getPlan(), totalCost, childBestPlan.get().first, i); + } + if (totalCost == null) { + continue; + } + groupExpression.updateLowestCostTable( + outputProperties, childInputProperties, totalCost); + ownerGroup.setBestPlan(groupExpression, totalCost, outputProperties); + if (bestNodeCost == null || nodeCost.getValue() < bestNodeCost.getValue()) { + bestNodeCost = nodeCost; + } + } + restoreMissingExpressionCostState(groupExpression, originalLowestCostTable, originalRequestPropertiesMap); + if (bestNodeCost != null) { + groupExpression.setCost(bestNodeCost); + } else { + groupExpression.setCost(originalCost); + } + for (Map.Entry<PhysicalProperties, PhysicalProperties> entry + : originalRequestPropertiesMap.entrySet()) { + if (groupExpression.getLowestCostTable().containsKey(entry.getKey())) { + groupExpression.putOutputPropertiesMap(entry.getValue(), entry.getKey()); + } + } + } + + private void restoreMissingExpressionCostState(GroupExpression groupExpression, + Map<PhysicalProperties, Pair<Cost, List<PhysicalProperties>>> originalLowestCostTable, + Map<PhysicalProperties, PhysicalProperties> originalRequestPropertiesMap) { + for (Map.Entry<PhysicalProperties, Pair<Cost, List<PhysicalProperties>>> entry + : originalLowestCostTable.entrySet()) { + if (!groupExpression.getLowestCostTable().containsKey(entry.getKey())) { + groupExpression.updateLowestCostTable(entry.getKey(), entry.getValue().second, entry.getValue().first); + } + } + for (Map.Entry<PhysicalProperties, PhysicalProperties> entry : originalRequestPropertiesMap.entrySet()) { + if (groupExpression.getLowestCostTable().containsKey(entry.getKey())) { + groupExpression.putOutputPropertiesMap(entry.getValue(), entry.getKey()); + } + } + } + + private boolean hasAvailableChildBestPlan(GroupExpression groupExpression, + List<PhysicalProperties> childInputProperties) { + if (childInputProperties.size() != groupExpression.arity()) { + return false; + } + for (int i = 0; i < childInputProperties.size(); i++) { + if (!groupExpression.child(i) + .getLowestCostPlan(childInputProperties.get(i)).isPresent()) { + return false; + } + } + return true; + } + + private void syncLogicalExpressionRowCount(Group group, double rowCount) { + for (GroupExpression logicalExpression : group.getLogicalExpressions()) { + if (logicalExpression.getEstOutputRowCount() > 0 + || !Double.isFinite(logicalExpression.getEstOutputRowCount())) { + logicalExpression.setEstOutputRowCount(rowCount); + } + } + } + + private void refreshPhysicalExpressionRowCount(Group group, double rowCount) { + for (GroupExpression physicalExpression : group.getPhysicalExpressions()) { + physicalExpression.setEstOutputRowCount(getPhysicalExpressionRowCount(physicalExpression, rowCount)); + } + } + + private double getPhysicalExpressionRowCount(GroupExpression physicalExpression, double rowCount) { + if (physicalExpression.getPlan() instanceof PhysicalProject && physicalExpression.arity() == 1) { + Statistics childStatistics = physicalExpression.child(0).getStatistics(); + if (childStatistics != null && Double.isFinite(childStatistics.getRowCount()) + && childStatistics.getRowCount() >= 0) { + return childStatistics.getRowCount(); + } + } + return rowCount; + } + + private boolean shouldSyncLogicalExpressionRowCount() { + return logicalExpressionRowCountSyncPolicy + == LogicalExpressionRowCountSyncPolicy.SYNC_WITH_GROUP_ROW_COUNT; + } + + private Map<GroupExpression, Statistics> filterCandidateStatisticsByPolicy( + LogicalRowCountAggregationPolicy aggregationPolicy, + Map<GroupExpression, Statistics> candidateStatisticsByExpression) { + if (aggregationPolicy != LogicalRowCountAggregationPolicy.TRUST_JOIN_COUNT + || candidateStatisticsByExpression.size() < 2) { + return candidateStatisticsByExpression; + } + int maxTrustJoinCount = Integer.MIN_VALUE; + Map<GroupExpression, Statistics> selectedCandidateStatisticsByExpression = new LinkedHashMap<>(); + for (Map.Entry<GroupExpression, Statistics> entry : candidateStatisticsByExpression.entrySet()) { + int trustJoinCount = countTrustJoins(entry.getKey(), new HashSet<>()); + if (trustJoinCount > maxTrustJoinCount) { + selectedCandidateStatisticsByExpression.clear(); + maxTrustJoinCount = trustJoinCount; + } + if (trustJoinCount == maxTrustJoinCount) { + selectedCandidateStatisticsByExpression.put(entry.getKey(), entry.getValue()); + } + } + return selectedCandidateStatisticsByExpression; + } + + private int countTrustJoins(GroupExpression groupExpression, Set<GroupExpression> visiting) { + if (!visiting.add(groupExpression)) { + return 0; + } + int trustJoinCount = isTrustJoin(groupExpression) ? 1 : 0; + for (Group child : groupExpression.children()) { + if (!child.getLogicalExpressions().isEmpty()) { + trustJoinCount += countTrustJoins(child.getFirstLogicalExpression(), visiting); Review Comment: The `trust_join_count` policy is still order-dependent here. While scoring a parent candidate, `countTrustJoins()` ignores the child expression chosen by that candidate and always recurses into `child.getFirstLogicalExpression()`. DPHyp stores all alternatives for the same bitmap in one memo group, so whichever child alternative was inserted first supplies the score for every parent candidate that references that group. That can make `filterCandidateStatisticsByPolicy(TRUST_JOIN_COUNT, ...)` keep or discard candidates based on memo insertion order rather than the join order being evaluated, even if per-expression row counts are later fixed. Please score the actual child alternative used by the candidate, or use a deterministic child-group score such as max/selected trust count instead of `getFirstLogicalExpression()`. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
