c21 commented on a change in pull request #29804: URL: https://github.com/apache/spark/pull/29804#discussion_r493222439
########## File path: sql/core/src/main/scala/org/apache/spark/sql/execution/bucketing/DisableUnnecessaryBucketedScan.scala ########## @@ -0,0 +1,153 @@ +/* + * 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.spark.sql.execution.bucketing + +import org.apache.spark.sql.catalyst.expressions.aggregate.{Partial, PartialMerge} +import org.apache.spark.sql.catalyst.plans.physical.{ClusteredDistribution, HashClusteredDistribution} +import org.apache.spark.sql.catalyst.rules.Rule +import org.apache.spark.sql.execution.{FileSourceScanExec, FilterExec, ProjectExec, SortExec, SparkPlan} +import org.apache.spark.sql.execution.aggregate.BaseAggregateExec +import org.apache.spark.sql.execution.exchange.Exchange +import org.apache.spark.sql.internal.SQLConf + +/** + * Disable unnecessary bucketed table scan based on actual physical query plan. + * NOTE: this rule is designed to be applied right after [[EnsureRequirements]], + * where all [[ShuffleExchangeExec]] and [[SortExec]] have been added to plan properly. + * + * When BUCKETING_ENABLED and AUTO_BUCKETED_SCAN_ENABLED are set to true, go through + * query plan to check where bucketed table scan is unnecessary, and disable bucketed table + * scan if needed. + * + * For all operators which [[hasInterestingPartition]] (i.e., require [[ClusteredDistribution]] + * or [[HashClusteredDistribution]]), check if the sub-plan for operator has [[Exchange]] and + * bucketed table scan. If yes, disable the bucketed table scan in the sub-plan. + * Only allow certain operators in sub-plan, which guarantees each sub-plan is single lineage + * (i.e., each operator has only one child). See details in + * [[disableBucketWithInterestingPartition]]). + * + * Examples: + * (1).join: + * SortMergeJoin(t1.i = t2.j) + * / \ + * Sort(i) Sort(j) + * / \ + * Shuffle(i) Scan(t2: i, j) + * / (bucketed on column j, enable bucketed scan) + * Scan(t1: i, j) + * (bucketed on column j, DISABLE bucketed scan) + * + * (2).aggregate: + * HashAggregate(i, ..., Final) + * | + * Shuffle(i) + * | + * HashAggregate(i, ..., Partial) + * | + * Filter + * | + * Scan(t1: i, j) + * (bucketed on column j, DISABLE bucketed scan) + * + * The idea of [[hasInterestingPartition]] is inspired from "interesting order" in + * the paper "Access Path Selection in a Relational Database Management System" + * (http://www.inf.ed.ac.uk/teaching/courses/adbs/AccessPath.pdf). Review comment: @maropu - sure, updated. Was just following the reference in [CBO join reorder](https://github.com/apache/spark/blob/master/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/CostBasedJoinReorder.scala#L117). Updated the link there as well. ########## File path: sql/core/src/main/scala/org/apache/spark/sql/execution/DataSourceScanExec.scala ########## @@ -339,7 +343,7 @@ case class FileSourceScanExec( location.getClass.getSimpleName + Utils.buildLocationMetadata(location.rootPaths, maxMetadataValueLength) val metadata = - Map( + HashMap( Review comment: @maropu - updated. Was just following other code in the same file. ########## File path: sql/core/src/main/scala/org/apache/spark/sql/execution/bucketing/DisableUnnecessaryBucketedScan.scala ########## @@ -0,0 +1,153 @@ +/* + * 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.spark.sql.execution.bucketing + +import org.apache.spark.sql.catalyst.expressions.aggregate.{Partial, PartialMerge} +import org.apache.spark.sql.catalyst.plans.physical.{ClusteredDistribution, HashClusteredDistribution} +import org.apache.spark.sql.catalyst.rules.Rule +import org.apache.spark.sql.execution.{FileSourceScanExec, FilterExec, ProjectExec, SortExec, SparkPlan} +import org.apache.spark.sql.execution.aggregate.BaseAggregateExec +import org.apache.spark.sql.execution.exchange.Exchange +import org.apache.spark.sql.internal.SQLConf + +/** + * Disable unnecessary bucketed table scan based on actual physical query plan. + * NOTE: this rule is designed to be applied right after [[EnsureRequirements]], + * where all [[ShuffleExchangeExec]] and [[SortExec]] have been added to plan properly. + * + * When BUCKETING_ENABLED and AUTO_BUCKETED_SCAN_ENABLED are set to true, go through + * query plan to check where bucketed table scan is unnecessary, and disable bucketed table + * scan if needed. + * + * For all operators which [[hasInterestingPartition]] (i.e., require [[ClusteredDistribution]] + * or [[HashClusteredDistribution]]), check if the sub-plan for operator has [[Exchange]] and + * bucketed table scan. If yes, disable the bucketed table scan in the sub-plan. + * Only allow certain operators in sub-plan, which guarantees each sub-plan is single lineage + * (i.e., each operator has only one child). See details in + * [[disableBucketWithInterestingPartition]]). + * + * Examples: + * (1).join: + * SortMergeJoin(t1.i = t2.j) + * / \ + * Sort(i) Sort(j) + * / \ + * Shuffle(i) Scan(t2: i, j) + * / (bucketed on column j, enable bucketed scan) + * Scan(t1: i, j) + * (bucketed on column j, DISABLE bucketed scan) + * + * (2).aggregate: + * HashAggregate(i, ..., Final) + * | + * Shuffle(i) + * | + * HashAggregate(i, ..., Partial) + * | + * Filter + * | + * Scan(t1: i, j) + * (bucketed on column j, DISABLE bucketed scan) + * + * The idea of [[hasInterestingPartition]] is inspired from "interesting order" in + * the paper "Access Path Selection in a Relational Database Management System" + * (http://www.inf.ed.ac.uk/teaching/courses/adbs/AccessPath.pdf). + */ +case class DisableUnnecessaryBucketedScan(conf: SQLConf) extends Rule[SparkPlan] { + + /** + * Disable bucketed table scan with pre-order traversal of plan. + * + * @param withInterestingPartition The traversed plan has operator with interesting partition. + * @param withExchange The traversed plan has [[Exchange]] operator. + */ + private def disableBucketWithInterestingPartition( + plan: SparkPlan, + withInterestingPartition: Boolean, + withExchange: Boolean): SparkPlan = { + plan match { + case p if hasInterestingPartition(p) => + // Operators with interesting partition, propagates `withInterestingPartition` as true + // to its children. + p.mapChildren(disableBucketWithInterestingPartition(_, true, false)) + case exchange: Exchange if withInterestingPartition => + // Exchange operator propagates `withExchange` as true to its child + // if the plan has interesting partition. + exchange.mapChildren(disableBucketWithInterestingPartition( + _, withInterestingPartition, true)) + case scan: FileSourceScanExec + if withInterestingPartition && withExchange && isBucketedScanWithoutFilter(scan) => + // Disable bucketed table scan if the plan has interesting partition, + // and [[Exchange]] in the plan. + scan.copy(disableBucketedScan = true) + case o => + if (isAllowedUnaryExecNode(o)) { + // Propagates `withInterestingPartition` and `withExchange` from parent + // for only allowed single-child nodes. + o.mapChildren(disableBucketWithInterestingPartition( + _, withInterestingPartition, withExchange)) + } else { + o.mapChildren(disableBucketWithInterestingPartition(_, false, false)) + } + } + } + + private def hasInterestingPartition(plan: SparkPlan): Boolean = { + plan.requiredChildDistribution.exists { + case _: ClusteredDistribution | _: HashClusteredDistribution => true + case _ => false + } + } + + private def isAllowedUnaryExecNode(plan: SparkPlan): Boolean = { + plan match { + case _: SortExec | _: Exchange | _: ProjectExec | _: FilterExec | + _: FileSourceScanExec => true + case partialAgg: BaseAggregateExec => + val modes = partialAgg.aggregateExpressions.map(_.mode) + modes.nonEmpty && modes.forall(mode => mode == Partial || mode == PartialMerge) + case _ => false + } + } + + private def isBucketedScanWithoutFilter(scan: FileSourceScanExec): Boolean = { + // Do not disable bucketed table scan if it has filter pruning, + // because bucketed table scan is still useful here to save CPU/IO cost with + // only reading selected bucket files. + scan.bucketedScan && scan.optionalBucketSet.isEmpty + } + + private def disableAllBucketedScan(plan: SparkPlan): SparkPlan = { + plan.transformUp { + case scan: FileSourceScanExec if isBucketedScanWithoutFilter(scan) => + scan.copy(disableBucketedScan = true) + } + } + + def apply(plan: SparkPlan): SparkPlan = { + if (!conf.bucketingEnabled || !conf.autoBucketedScanEnabled) { Review comment: @maropu - sure, updated. ########## File path: sql/core/src/main/scala/org/apache/spark/sql/execution/bucketing/DisableUnnecessaryBucketedScan.scala ########## @@ -0,0 +1,153 @@ +/* + * 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.spark.sql.execution.bucketing + +import org.apache.spark.sql.catalyst.expressions.aggregate.{Partial, PartialMerge} +import org.apache.spark.sql.catalyst.plans.physical.{ClusteredDistribution, HashClusteredDistribution} +import org.apache.spark.sql.catalyst.rules.Rule +import org.apache.spark.sql.execution.{FileSourceScanExec, FilterExec, ProjectExec, SortExec, SparkPlan} +import org.apache.spark.sql.execution.aggregate.BaseAggregateExec +import org.apache.spark.sql.execution.exchange.Exchange +import org.apache.spark.sql.internal.SQLConf + +/** + * Disable unnecessary bucketed table scan based on actual physical query plan. + * NOTE: this rule is designed to be applied right after [[EnsureRequirements]], + * where all [[ShuffleExchangeExec]] and [[SortExec]] have been added to plan properly. + * + * When BUCKETING_ENABLED and AUTO_BUCKETED_SCAN_ENABLED are set to true, go through + * query plan to check where bucketed table scan is unnecessary, and disable bucketed table + * scan if needed. + * + * For all operators which [[hasInterestingPartition]] (i.e., require [[ClusteredDistribution]] + * or [[HashClusteredDistribution]]), check if the sub-plan for operator has [[Exchange]] and + * bucketed table scan. If yes, disable the bucketed table scan in the sub-plan. + * Only allow certain operators in sub-plan, which guarantees each sub-plan is single lineage + * (i.e., each operator has only one child). See details in + * [[disableBucketWithInterestingPartition]]). + * + * Examples: + * (1).join: + * SortMergeJoin(t1.i = t2.j) + * / \ + * Sort(i) Sort(j) + * / \ + * Shuffle(i) Scan(t2: i, j) + * / (bucketed on column j, enable bucketed scan) + * Scan(t1: i, j) + * (bucketed on column j, DISABLE bucketed scan) + * + * (2).aggregate: + * HashAggregate(i, ..., Final) + * | + * Shuffle(i) + * | + * HashAggregate(i, ..., Partial) + * | + * Filter + * | + * Scan(t1: i, j) + * (bucketed on column j, DISABLE bucketed scan) + * + * The idea of [[hasInterestingPartition]] is inspired from "interesting order" in + * the paper "Access Path Selection in a Relational Database Management System" + * (http://www.inf.ed.ac.uk/teaching/courses/adbs/AccessPath.pdf). + */ +case class DisableUnnecessaryBucketedScan(conf: SQLConf) extends Rule[SparkPlan] { + + /** + * Disable bucketed table scan with pre-order traversal of plan. + * + * @param withInterestingPartition The traversed plan has operator with interesting partition. + * @param withExchange The traversed plan has [[Exchange]] operator. + */ + private def disableBucketWithInterestingPartition( + plan: SparkPlan, + withInterestingPartition: Boolean, + withExchange: Boolean): SparkPlan = { + plan match { + case p if hasInterestingPartition(p) => + // Operators with interesting partition, propagates `withInterestingPartition` as true + // to its children. + p.mapChildren(disableBucketWithInterestingPartition(_, true, false)) + case exchange: Exchange if withInterestingPartition => + // Exchange operator propagates `withExchange` as true to its child + // if the plan has interesting partition. + exchange.mapChildren(disableBucketWithInterestingPartition( + _, withInterestingPartition, true)) + case scan: FileSourceScanExec + if withInterestingPartition && withExchange && isBucketedScanWithoutFilter(scan) => + // Disable bucketed table scan if the plan has interesting partition, + // and [[Exchange]] in the plan. + scan.copy(disableBucketedScan = true) + case o => + if (isAllowedUnaryExecNode(o)) { + // Propagates `withInterestingPartition` and `withExchange` from parent + // for only allowed single-child nodes. + o.mapChildren(disableBucketWithInterestingPartition( + _, withInterestingPartition, withExchange)) + } else { + o.mapChildren(disableBucketWithInterestingPartition(_, false, false)) + } + } + } + + private def hasInterestingPartition(plan: SparkPlan): Boolean = { + plan.requiredChildDistribution.exists { + case _: ClusteredDistribution | _: HashClusteredDistribution => true + case _ => false + } + } + + private def isAllowedUnaryExecNode(plan: SparkPlan): Boolean = { + plan match { + case _: SortExec | _: Exchange | _: ProjectExec | _: FilterExec | + _: FileSourceScanExec => true Review comment: @maropu - `FileSourceScanExec` should be pattern matched earlier in `disableBucketWithInterestingPartition` before `isAllowedUnaryExecNode` check. But thanks for catch, we don't need to add `Exchange` and `FileSourceScanExec` here anyway, removed both. ########## File path: sql/core/src/main/scala/org/apache/spark/sql/execution/bucketing/DisableUnnecessaryBucketedScan.scala ########## @@ -0,0 +1,153 @@ +/* + * 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.spark.sql.execution.bucketing + +import org.apache.spark.sql.catalyst.expressions.aggregate.{Partial, PartialMerge} +import org.apache.spark.sql.catalyst.plans.physical.{ClusteredDistribution, HashClusteredDistribution} +import org.apache.spark.sql.catalyst.rules.Rule +import org.apache.spark.sql.execution.{FileSourceScanExec, FilterExec, ProjectExec, SortExec, SparkPlan} +import org.apache.spark.sql.execution.aggregate.BaseAggregateExec +import org.apache.spark.sql.execution.exchange.Exchange +import org.apache.spark.sql.internal.SQLConf + +/** + * Disable unnecessary bucketed table scan based on actual physical query plan. + * NOTE: this rule is designed to be applied right after [[EnsureRequirements]], + * where all [[ShuffleExchangeExec]] and [[SortExec]] have been added to plan properly. + * + * When BUCKETING_ENABLED and AUTO_BUCKETED_SCAN_ENABLED are set to true, go through + * query plan to check where bucketed table scan is unnecessary, and disable bucketed table + * scan if needed. + * + * For all operators which [[hasInterestingPartition]] (i.e., require [[ClusteredDistribution]] + * or [[HashClusteredDistribution]]), check if the sub-plan for operator has [[Exchange]] and + * bucketed table scan. If yes, disable the bucketed table scan in the sub-plan. + * Only allow certain operators in sub-plan, which guarantees each sub-plan is single lineage + * (i.e., each operator has only one child). See details in + * [[disableBucketWithInterestingPartition]]). + * + * Examples: + * (1).join: + * SortMergeJoin(t1.i = t2.j) + * / \ + * Sort(i) Sort(j) + * / \ + * Shuffle(i) Scan(t2: i, j) + * / (bucketed on column j, enable bucketed scan) + * Scan(t1: i, j) + * (bucketed on column j, DISABLE bucketed scan) + * + * (2).aggregate: + * HashAggregate(i, ..., Final) + * | + * Shuffle(i) + * | + * HashAggregate(i, ..., Partial) + * | + * Filter + * | + * Scan(t1: i, j) + * (bucketed on column j, DISABLE bucketed scan) + * + * The idea of [[hasInterestingPartition]] is inspired from "interesting order" in + * the paper "Access Path Selection in a Relational Database Management System" + * (http://www.inf.ed.ac.uk/teaching/courses/adbs/AccessPath.pdf). + */ +case class DisableUnnecessaryBucketedScan(conf: SQLConf) extends Rule[SparkPlan] { + + /** + * Disable bucketed table scan with pre-order traversal of plan. + * + * @param withInterestingPartition The traversed plan has operator with interesting partition. + * @param withExchange The traversed plan has [[Exchange]] operator. + */ + private def disableBucketWithInterestingPartition( + plan: SparkPlan, + withInterestingPartition: Boolean, + withExchange: Boolean): SparkPlan = { + plan match { + case p if hasInterestingPartition(p) => + // Operators with interesting partition, propagates `withInterestingPartition` as true + // to its children. + p.mapChildren(disableBucketWithInterestingPartition(_, true, false)) Review comment: @maropu - this is good question. I think currently only matching `Exchange` will also do the same job. Either way is fine with me. `InterestingPartition` and later on `InterestingOrder` (for bucketed sorted scan) looks like more general and we can extend them later. One interesting extension I can think of - if bucketed scan parallelism is too low (to few # of buckets), we may decide to not do a bucketed scan for join to trade-off for query run-time vs extra shuffle cost (in this case, there's no `Exchange` before join). ########## File path: sql/core/src/main/scala/org/apache/spark/sql/execution/bucketing/DisableUnnecessaryBucketedScan.scala ########## @@ -0,0 +1,153 @@ +/* + * 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.spark.sql.execution.bucketing + +import org.apache.spark.sql.catalyst.expressions.aggregate.{Partial, PartialMerge} +import org.apache.spark.sql.catalyst.plans.physical.{ClusteredDistribution, HashClusteredDistribution} +import org.apache.spark.sql.catalyst.rules.Rule +import org.apache.spark.sql.execution.{FileSourceScanExec, FilterExec, ProjectExec, SortExec, SparkPlan} +import org.apache.spark.sql.execution.aggregate.BaseAggregateExec +import org.apache.spark.sql.execution.exchange.Exchange +import org.apache.spark.sql.internal.SQLConf + +/** + * Disable unnecessary bucketed table scan based on actual physical query plan. + * NOTE: this rule is designed to be applied right after [[EnsureRequirements]], + * where all [[ShuffleExchangeExec]] and [[SortExec]] have been added to plan properly. + * + * When BUCKETING_ENABLED and AUTO_BUCKETED_SCAN_ENABLED are set to true, go through + * query plan to check where bucketed table scan is unnecessary, and disable bucketed table + * scan if needed. + * + * For all operators which [[hasInterestingPartition]] (i.e., require [[ClusteredDistribution]] + * or [[HashClusteredDistribution]]), check if the sub-plan for operator has [[Exchange]] and + * bucketed table scan. If yes, disable the bucketed table scan in the sub-plan. + * Only allow certain operators in sub-plan, which guarantees each sub-plan is single lineage + * (i.e., each operator has only one child). See details in + * [[disableBucketWithInterestingPartition]]). + * + * Examples: + * (1).join: + * SortMergeJoin(t1.i = t2.j) + * / \ + * Sort(i) Sort(j) + * / \ + * Shuffle(i) Scan(t2: i, j) + * / (bucketed on column j, enable bucketed scan) + * Scan(t1: i, j) + * (bucketed on column j, DISABLE bucketed scan) + * + * (2).aggregate: + * HashAggregate(i, ..., Final) + * | + * Shuffle(i) + * | + * HashAggregate(i, ..., Partial) + * | + * Filter + * | + * Scan(t1: i, j) + * (bucketed on column j, DISABLE bucketed scan) + * + * The idea of [[hasInterestingPartition]] is inspired from "interesting order" in + * the paper "Access Path Selection in a Relational Database Management System" + * (http://www.inf.ed.ac.uk/teaching/courses/adbs/AccessPath.pdf). + */ +case class DisableUnnecessaryBucketedScan(conf: SQLConf) extends Rule[SparkPlan] { + + /** + * Disable bucketed table scan with pre-order traversal of plan. + * + * @param withInterestingPartition The traversed plan has operator with interesting partition. + * @param withExchange The traversed plan has [[Exchange]] operator. + */ + private def disableBucketWithInterestingPartition( + plan: SparkPlan, + withInterestingPartition: Boolean, + withExchange: Boolean): SparkPlan = { + plan match { + case p if hasInterestingPartition(p) => + // Operators with interesting partition, propagates `withInterestingPartition` as true + // to its children. + p.mapChildren(disableBucketWithInterestingPartition(_, true, false)) + case exchange: Exchange if withInterestingPartition => + // Exchange operator propagates `withExchange` as true to its child + // if the plan has interesting partition. + exchange.mapChildren(disableBucketWithInterestingPartition( + _, withInterestingPartition, true)) + case scan: FileSourceScanExec + if withInterestingPartition && withExchange && isBucketedScanWithoutFilter(scan) => + // Disable bucketed table scan if the plan has interesting partition, + // and [[Exchange]] in the plan. + scan.copy(disableBucketedScan = true) + case o => + if (isAllowedUnaryExecNode(o)) { + // Propagates `withInterestingPartition` and `withExchange` from parent + // for only allowed single-child nodes. + o.mapChildren(disableBucketWithInterestingPartition( + _, withInterestingPartition, withExchange)) + } else { + o.mapChildren(disableBucketWithInterestingPartition(_, false, false)) + } + } + } + + private def hasInterestingPartition(plan: SparkPlan): Boolean = { + plan.requiredChildDistribution.exists { + case _: ClusteredDistribution | _: HashClusteredDistribution => true + case _ => false + } + } + + private def isAllowedUnaryExecNode(plan: SparkPlan): Boolean = { + plan match { + case _: SortExec | _: Exchange | _: ProjectExec | _: FilterExec | + _: FileSourceScanExec => true + case partialAgg: BaseAggregateExec => + val modes = partialAgg.aggregateExpressions.map(_.mode) + modes.nonEmpty && modes.forall(mode => mode == Partial || mode == PartialMerge) + case _ => false + } + } + + private def isBucketedScanWithoutFilter(scan: FileSourceScanExec): Boolean = { + // Do not disable bucketed table scan if it has filter pruning, + // because bucketed table scan is still useful here to save CPU/IO cost with + // only reading selected bucket files. + scan.bucketedScan && scan.optionalBucketSet.isEmpty Review comment: @maropu - this is a good question, and I think it is kind of out of scope for this PR and needs more thoughts later. We don't have a cost model to decide whether to do (bucketed filter + bucketed scan) vs (normal filter + non-bucketed scan). It can depend on number of buckets, size of filtered buckets, CPU cost for filter, etc. ########## File path: sql/core/src/main/scala/org/apache/spark/sql/execution/DataSourceScanExec.scala ########## @@ -348,20 +352,22 @@ case class FileSourceScanExec( "DataFilters" -> seqToString(dataFilters), "Location" -> locationDesc) - val withSelectedBucketsCount = relation.bucketSpec.map { spec => - val numSelectedBuckets = optionalBucketSet.map { b => - b.cardinality() - } getOrElse { - spec.numBuckets + if (bucketedScan) { + relation.bucketSpec.map { spec => Review comment: @maropu - just for my own education, why does it matter? Updated anyway. ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
