anchovYu commented on code in PR #39508: URL: https://github.com/apache/spark/pull/39508#discussion_r1092357999
########## sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/ResolveReferencesInAggregate.scala: ########## @@ -0,0 +1,201 @@ +/* + * 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.catalyst.analysis + +import org.apache.spark.sql.AnalysisException +import org.apache.spark.sql.catalyst.SQLConfHelper +import org.apache.spark.sql.catalyst.expressions.{AliasHelper, Attribute, Expression, NamedExpression} +import org.apache.spark.sql.catalyst.expressions.aggregate.AggregateExpression +import org.apache.spark.sql.catalyst.plans.logical.{Aggregate, AppendColumns, LogicalPlan} +import org.apache.spark.sql.catalyst.trees.TreePattern.{LATERAL_COLUMN_ALIAS_REFERENCE, UNRESOLVED_ATTRIBUTE} + +/** + * A virtual rule to resolve [[UnresolvedAttribute]] in [[Aggregate]]. It's only used by the real + * rule `ResolveReferences`. The column resolution order for [[Aggregate]] is: + * 1. Resolves the columns to [[AttributeReference]] with the output of the child plan. This + * includes metadata columns as well. + * 2. Resolves the columns to a literal function which is allowed to be invoked without braces, e.g. + * `SELECT col, current_date FROM t`. + * 3. If aggregate expressions are all resolved, resolve GROUP BY alias and GROUP BY ALL. + * 3.1. If the grouping expressions contain an unresolved column whose name matches an alias in the + * SELECT list, resolves that unresolved column to the alias. This is to support SQL pattern + * like `SELECT a + b AS c, max(col) FROM t GROUP BY c`. + * 3.2. If the grouping expressions only have one single unresolved column named 'ALL', expanded it + * to include all non-aggregate columns in the SELECT list. This is to support SQL pattern like + * `SELECT col1, col2, agg_expr(...) FROM t GROUP BY ALL`. + * 4. Resolves the columns in aggregate expressions to [[LateralColumnAliasReference]] if + * it references the alias defined previously in the SELECT list. The rule + * `ResolveLateralColumnAliasReference` will further resolve [[LateralColumnAliasReference]] and + * rewrite the plan. This is to support SQL pattern like + * `SELECT col1 + 1 AS x, x + 1 AS y, y + 1 AS z FROM t`. + * 5. Resolves the columns to outer references with the outer plan if we are resolving subquery + * expressions. + */ +object ResolveReferencesInAggregate extends SQLConfHelper + with ColumnResolutionHelper with AliasHelper { + def apply(a: Aggregate): Aggregate = { + val planForResolve = a.child match { + // SPARK-25942: Resolves aggregate expressions with `AppendColumns`'s children, instead of + // `AppendColumns`, because `AppendColumns`'s serializer might produce conflict attribute + // names leading to ambiguous references exception. + case appendColumns: AppendColumns => appendColumns + case _ => a + } + + val resolvedGroupExprsNoOuter = a.groupingExpressions + .map(resolveExpressionByPlanChildren(_, planForResolve, allowOuter = false)) + val resolvedAggExprsNoOuter = a.aggregateExpressions.map( + resolveExpressionByPlanChildren(_, planForResolve, allowOuter = false)) + val resolvedAggExprsWithLCA = resolveLateralColumnAlias(resolvedAggExprsNoOuter) + val resolvedAggExprsWithOuter = resolvedAggExprsWithLCA.map(resolveOuterRef) + .map(_.asInstanceOf[NamedExpression]) + // `groupingExpressions` may rely on `aggregateExpressions`, due to features like GROUP BY alias + // and GROUP BY ALL. We only do basic resolution for `groupingExpressions`, and will further + // resolve it after `aggregateExpressions` are all resolved. Note: the basic resolution is + // needed as `aggregateExpressions` may rely on `groupingExpressions` as well, for the session + // window feature. See the rule `SessionWindowing` for more details. + val resolvedGroupExprs = if (resolvedAggExprsWithOuter.forall(_.resolved)) { + val resolved = resolveGroupByAll( + resolvedAggExprsWithOuter, + resolveGroupByAlias(resolvedAggExprsWithOuter, resolvedGroupExprsNoOuter) + ).map(resolveOuterRef) + // TODO: currently we don't support LCA in `groupingExpressions` yet. + if (resolved.exists(_.containsPattern(LATERAL_COLUMN_ALIAS_REFERENCE))) { + throw new AnalysisException( + errorClass = "UNSUPPORTED_FEATURE.LATERAL_COLUMN_ALIAS_IN_GROUP_BY", + messageParameters = Map.empty) + } + resolved + } else { + // Do not resolve columns in grouping expressions to outer references here, as the aggregate + // expressions are not fully resolved yet and we still have chances to resolve GROUP BY + // alias/ALL in the next iteration. If aggregate expressions end up as unresolved, we don't + // need to resolve grouping expressions at all, as `CheckAnalysis` will report error for + // aggregate expressions first. + resolvedGroupExprsNoOuter + } + a.copy( + // The aliases in grouping expressions are useless and will be removed at the end of analysis + // by the rule `CleanupAliases`. However, some rules need to find the grouping expressions + // from aggregate expressions during analysis. If we don't remove alias here, then these rules + // can't find the grouping expressions via `semanticEquals` and the analysis will fail. + // Example rules: ResolveGroupingAnalytics (See SPARK-31670 for more details) and + // ResolveLateralColumnAliasReference. + groupingExpressions = resolvedGroupExprs.map(trimAliases), Review Comment: For details in how it affects LCA: Without the trim, ``` === Applying Rule org.apache.spark.sql.catalyst.analysis.ResolveLateralColumnAliasReference === !Aggregate [salary#19 AS lca#15], [salary#19 AS lca#15, (lateralAliasReference(lca#15, lca, lca#15) + 1) AS col#16] Project [lca#15, (lateralAliasReference(lca#15, lca, lca#15) + 1) AS col#16] !+- SubqueryAlias spark_catalog.default.employee +- Aggregate [salary#19 AS lca#15], [salary#19 AS lca#15] ! +- Relation spark_catalog.default.employee[dept#17,name#18,salary#19,bonus#20,properties#21] orc +- SubqueryAlias spark_catalog.default.employee ! +- Relation spark_catalog.default.employee[dept#17,name#18,salary#19,bonus#20,properties#21] orc ``` `ResolveLateralColumnAliasReference` matches the whole `salary#19 AS lca#15` same as the grouping expression and pushes it down to the Aggregate (this rule finds matches in grouping expression and aggregate functions to push down). It causes problem for later application of this rule on Project: there is no alias in the Project any more so the `lateralAliasReference` can't find an alias to match. -- 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]
