dtenedor commented on code in PR #41262: URL: https://github.com/apache/spark/pull/41262#discussion_r1205887470
########## sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/ResolveColumnDefaultInInsert.scala: ########## @@ -0,0 +1,166 @@ +/* + * 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.catalyst.SQLConfHelper +import org.apache.spark.sql.catalyst.expressions.{Alias, Literal} +import org.apache.spark.sql.catalyst.plans.logical._ +import org.apache.spark.sql.catalyst.trees.TreePattern.UNRESOLVED_ATTRIBUTE +import org.apache.spark.sql.catalyst.util.ResolveDefaultColumns.{containsExplicitDefaultColumn, getDefaultValueExpr, isExplicitDefaultColumn} +import org.apache.spark.sql.errors.QueryCompilationErrors +import org.apache.spark.sql.internal.SQLConf +import org.apache.spark.sql.types.StructField + +/** + * A virtual rule to resolve column "DEFAULT" in [[Project]] and [[UnresolvedInlineTable]] under + * [[InsertIntoStatement]]. It's only used by the real rule `ResolveReferences`. + * + * This virtual rule is triggered if: + * 1. The column "DEFAULT" can't be resolved normally by `ResolveReferences`. This is guaranteed as + * `ResolveReferences` resolves the query plan bottom up. This means that when we reach here to + * resolve [[InsertIntoStatement]], its child plans have already been resolved by + * `ResolveReferences`. + * 2. The plan nodes between [[Project]] and [[InsertIntoStatement]] are + * all unary nodes that inherit the output columns from its child. + * 3. The plan nodes between [[UnresolvedInlineTable]] and [[InsertIntoStatement]] are either + * [[Project]], or [[Aggregate]], or [[SubqueryAlias]]. + */ +case object ResolveColumnDefaultInInsert extends SQLConfHelper with ColumnResolutionHelper { + // TODO (SPARK-43752): support v2 write commands as well. + def apply(plan: LogicalPlan): LogicalPlan = plan match { + case i: InsertIntoStatement if conf.enableDefaultColumns && i.table.resolved && + i.query.containsPattern(UNRESOLVED_ATTRIBUTE) => + val staticPartCols = i.partitionSpec.filter(_._2.isDefined).keySet.map(normalizeFieldName) + // For INSERT with static partitions, such as `INSERT INTO t PARTITION(c=1) SELECT ...`, the + // input query schema should match the table schema excluding columns with static + // partition values. + val expectedQuerySchema = i.table.schema.filter { field => + !staticPartCols.contains(normalizeFieldName(field.name)) + } + // Normally, we should match the query schema with the table schema by position. If the n-th + // column of the query is the DEFAULT column, we should get the default value expression + // defined for the n-th column of the table. However, if the INSERT has a column list, such as + // `INSERT INTO t(b, c, a)`, the matching should be by name. For example, the first column of + // the query should match the column 'b' of the table. + // To simplify the implementation, `resolveColumnDefault` always does by-position match. If + // the INSERT has a column list, we reorder the table schema w.r.t. the column list and pass + // the reordered schema as the expected schema to `resolveColumnDefault`. + if (i.userSpecifiedCols.isEmpty) { + i.withNewChildren(Seq(resolveColumnDefault(i.query, expectedQuerySchema))) + } else { + val colNamesToFields: Map[String, StructField] = expectedQuerySchema.map { field => + normalizeFieldName(field.name) -> field + }.toMap + val reorder = i.userSpecifiedCols.map { col => + colNamesToFields.get(normalizeFieldName(col)) + } + if (reorder.forall(_.isDefined)) { + i.withNewChildren(Seq(resolveColumnDefault(i.query, reorder.flatten))) + } else { + i + } + } + + case _ => plan + } + + /** + * Resolves the column "DEFAULT" in [[Project]] and [[UnresolvedInlineTable]]. A column is a + * "DEFAULT" column if all the following conditions are met: + * 1. The expression inside project list or inline table expressions is a single + * [[UnresolvedAttribute]] with name "DEFAULT". This means `SELECT DEFAULT, ...` is valid but + * `SELECT DEFAULT + 1, ...` is not. + * 2. The project list or inline table expressions have less elements than the expected schema. + * To find the default value definition, we need to find the matching column for expressions + * inside project list or inline table expressions. This matching is by position and it + * doesn't make sense if we have more expressions than the columns of expected schema. + * 3. The plan nodes between [[Project]] and [[InsertIntoStatement]] are + * all unary nodes that inherit the output columns from its child. + * 4. The plan nodes between [[UnresolvedInlineTable]] and [[InsertIntoStatement]] are either + * [[Project]], or [[Aggregate]], or [[SubqueryAlias]]. + */ + private def resolveColumnDefault( + plan: LogicalPlan, + expectedQuerySchema: Seq[StructField], + acceptProject: Boolean = true, + acceptInlineTable: Boolean = true): LogicalPlan = { + plan match { + case _: SubqueryAlias => + plan.mapChildren( + resolveColumnDefault(_, expectedQuerySchema, acceptProject, acceptInlineTable)) + + case _: GlobalLimit | _: LocalLimit | _: Offset | _: Sort if acceptProject => + plan.mapChildren( + resolveColumnDefault(_, expectedQuerySchema, acceptInlineTable = false)) + + case p: Project if acceptProject && p.child.resolved && + p.containsPattern(UNRESOLVED_ATTRIBUTE) && + p.projectList.length <= expectedQuerySchema.length => + val newProjectList = p.projectList.zipWithIndex.map { + case (u: UnresolvedAttribute, i) if isExplicitDefaultColumn(u) => + val field = expectedQuerySchema(i) Review Comment: optional: when I wrote the original `ResolveDefaultColumns` rule, I named this variable `insertTableSchemaWithoutPartitionColumns` because I found myself confused frequently when reading the variable name. We could name this `insertTargetTableSchema` to clarify this, or `insertTargetTableSchemaWithoutPartitionColumns` if you don't think that's too verbose. ########## sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/ResolveColumnDefaultInInsert.scala: ########## @@ -0,0 +1,166 @@ +/* + * 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.catalyst.SQLConfHelper +import org.apache.spark.sql.catalyst.expressions.{Alias, Literal} +import org.apache.spark.sql.catalyst.plans.logical._ +import org.apache.spark.sql.catalyst.trees.TreePattern.UNRESOLVED_ATTRIBUTE +import org.apache.spark.sql.catalyst.util.ResolveDefaultColumns.{containsExplicitDefaultColumn, getDefaultValueExpr, isExplicitDefaultColumn} +import org.apache.spark.sql.errors.QueryCompilationErrors +import org.apache.spark.sql.internal.SQLConf +import org.apache.spark.sql.types.StructField + +/** + * A virtual rule to resolve column "DEFAULT" in [[Project]] and [[UnresolvedInlineTable]] under + * [[InsertIntoStatement]]. It's only used by the real rule `ResolveReferences`. + * + * This virtual rule is triggered if: + * 1. The column "DEFAULT" can't be resolved normally by `ResolveReferences`. This is guaranteed as + * `ResolveReferences` resolves the query plan bottom up. This means that when we reach here to + * resolve [[InsertIntoStatement]], its child plans have already been resolved by + * `ResolveReferences`. + * 2. The plan nodes between [[Project]] and [[InsertIntoStatement]] are + * all unary nodes that inherit the output columns from its child. + * 3. The plan nodes between [[UnresolvedInlineTable]] and [[InsertIntoStatement]] are either + * [[Project]], or [[Aggregate]], or [[SubqueryAlias]]. + */ +case object ResolveColumnDefaultInInsert extends SQLConfHelper with ColumnResolutionHelper { + // TODO (SPARK-43752): support v2 write commands as well. + def apply(plan: LogicalPlan): LogicalPlan = plan match { + case i: InsertIntoStatement if conf.enableDefaultColumns && i.table.resolved && + i.query.containsPattern(UNRESOLVED_ATTRIBUTE) => + val staticPartCols = i.partitionSpec.filter(_._2.isDefined).keySet.map(normalizeFieldName) + // For INSERT with static partitions, such as `INSERT INTO t PARTITION(c=1) SELECT ...`, the + // input query schema should match the table schema excluding columns with static + // partition values. + val expectedQuerySchema = i.table.schema.filter { field => + !staticPartCols.contains(normalizeFieldName(field.name)) + } + // Normally, we should match the query schema with the table schema by position. If the n-th + // column of the query is the DEFAULT column, we should get the default value expression + // defined for the n-th column of the table. However, if the INSERT has a column list, such as + // `INSERT INTO t(b, c, a)`, the matching should be by name. For example, the first column of + // the query should match the column 'b' of the table. + // To simplify the implementation, `resolveColumnDefault` always does by-position match. If + // the INSERT has a column list, we reorder the table schema w.r.t. the column list and pass + // the reordered schema as the expected schema to `resolveColumnDefault`. + if (i.userSpecifiedCols.isEmpty) { + i.withNewChildren(Seq(resolveColumnDefault(i.query, expectedQuerySchema))) + } else { + val colNamesToFields: Map[String, StructField] = expectedQuerySchema.map { field => + normalizeFieldName(field.name) -> field + }.toMap + val reorder = i.userSpecifiedCols.map { col => + colNamesToFields.get(normalizeFieldName(col)) + } + if (reorder.forall(_.isDefined)) { + i.withNewChildren(Seq(resolveColumnDefault(i.query, reorder.flatten))) + } else { + i + } + } + + case _ => plan + } + + /** + * Resolves the column "DEFAULT" in [[Project]] and [[UnresolvedInlineTable]]. A column is a + * "DEFAULT" column if all the following conditions are met: + * 1. The expression inside project list or inline table expressions is a single + * [[UnresolvedAttribute]] with name "DEFAULT". This means `SELECT DEFAULT, ...` is valid but + * `SELECT DEFAULT + 1, ...` is not. + * 2. The project list or inline table expressions have less elements than the expected schema. + * To find the default value definition, we need to find the matching column for expressions + * inside project list or inline table expressions. This matching is by position and it + * doesn't make sense if we have more expressions than the columns of expected schema. + * 3. The plan nodes between [[Project]] and [[InsertIntoStatement]] are + * all unary nodes that inherit the output columns from its child. + * 4. The plan nodes between [[UnresolvedInlineTable]] and [[InsertIntoStatement]] are either + * [[Project]], or [[Aggregate]], or [[SubqueryAlias]]. + */ + private def resolveColumnDefault( + plan: LogicalPlan, + expectedQuerySchema: Seq[StructField], + acceptProject: Boolean = true, + acceptInlineTable: Boolean = true): LogicalPlan = { + plan match { + case _: SubqueryAlias => + plan.mapChildren( + resolveColumnDefault(_, expectedQuerySchema, acceptProject, acceptInlineTable)) + + case _: GlobalLimit | _: LocalLimit | _: Offset | _: Sort if acceptProject => + plan.mapChildren( + resolveColumnDefault(_, expectedQuerySchema, acceptInlineTable = false)) Review Comment: optional: It looks like the only purpose of `acceptInlineTable` is setting it to false here in the event of a LIMIT and/or OFFSET and/or ORDER BY on top of a VALUES list. Do you think this check is strictly necessary? If not, we can simplify by removing `acceptInlineTable` as an argument to this function. ########## sql/core/src/test/scala/org/apache/spark/sql/execution/command/PlanResolutionSuite.scala: ########## @@ -1143,38 +1140,34 @@ class PlanResolutionSuite extends AnalysisTest { checkError( exception = intercept[AnalysisException] { - parseAndResolve(sql8) + parseAndResolve(sql8, checkAnalysis = true) Review Comment: This sounds reasonable, we can leave this test here, but we don't have to exhaustively cover all the cases. ########## sql/core/src/test/scala/org/apache/spark/sql/sources/InsertSuite.scala: ########## @@ -1236,37 +1206,37 @@ class InsertSuite extends DataSourceTest with SharedSparkSession { sql("create table t(i boolean, s bigint) using parquet") assert(intercept[AnalysisException] { sql("insert into t (i) values (true)") - }.getMessage.contains(addOneColButExpectedTwo)) + }.getMessage.contains("Cannot find data for output column 's'")) } withTable("t") { sql("create table t(i boolean default true, s bigint) using parquet") assert(intercept[AnalysisException] { sql("insert into t (i) values (default)") - }.getMessage.contains(addOneColButExpectedTwo)) + }.getMessage.contains("Cannot find data for output column 's'")) Review Comment: can we dedup this expected error message substring into one place, or even better, use `checkError` to assert on the error class? ########## sql/core/src/test/scala/org/apache/spark/sql/sources/InsertSuite.scala: ########## @@ -142,17 +142,6 @@ class InsertSuite extends DataSourceTest with SharedSparkSession { ) } - test("SELECT clause generating a different number of columns is not allowed.") { Review Comment: should we keep the test, but change its result to assert that it succeeds? or is this behavior exercised elsewhere in this test file? ########## sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/ResolveColumnDefaultInInsert.scala: ########## @@ -0,0 +1,142 @@ +/* + * 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.catalyst.SQLConfHelper +import org.apache.spark.sql.catalyst.expressions.{Alias, Literal} +import org.apache.spark.sql.catalyst.plans.logical._ +import org.apache.spark.sql.catalyst.trees.TreePattern.UNRESOLVED_ATTRIBUTE +import org.apache.spark.sql.catalyst.util.ResolveDefaultColumns.{containsExplicitDefaultColumn, getDefaultValueExpr, isExplicitDefaultColumn} +import org.apache.spark.sql.errors.QueryCompilationErrors +import org.apache.spark.sql.internal.SQLConf +import org.apache.spark.sql.types.StructField + +/** + * A virtual rule to resolve column "DEFAULT" in [[Project]] and [[UnresolvedInlineTable]] under + * [[InsertIntoStatement]]. It's only used by the real rule `ResolveReferences`. + * + * This virtual rule is triggered if: + * 1. The column "DEFAULT" can't be resolved normally by `ResolveReferences`. This is guaranteed as + * `ResolveReferences` resolves the query plan bottom up. This means that when we reach here to + * resolve [[InsertIntoStatement]], its child plans have already been resolved by + * `ResolveReferences`. + * 2. The plan nodes between [[Project]] and [[InsertIntoStatement]] are + * all unary nodes that inherit the output columns from its child. + * 3. The plan nodes between [[UnresolvedInlineTable]] and [[InsertIntoStatement]] are either + * [[Project]], or [[Aggregate]], or [[SubqueryAlias]]. + */ +case object ResolveColumnDefaultInInsert extends SQLConfHelper with ColumnResolutionHelper { + // TODO (SPARK-43752): support v2 write commands as well. + def apply(plan: LogicalPlan): LogicalPlan = plan match { + case i: InsertIntoStatement if conf.enableDefaultColumns && i.table.resolved && + i.query.containsPattern(UNRESOLVED_ATTRIBUTE) => + val staticPartCols = i.partitionSpec.filter(_._2.isDefined).keys.map(normalizeFieldName).toSet + val expectedQuerySchema = i.table.schema.filter { field => + !staticPartCols.contains(normalizeFieldName(field.name)) + } + if (i.userSpecifiedCols.isEmpty) { + i.withNewChildren(Seq(resolveColumnDefault(i.query, expectedQuerySchema))) + } else { + // Reorder the fields in `expectedQuerySchema` according to the user-specified column list + // of the INSERT command. + val colNamesToFields: Map[String, StructField] = expectedQuerySchema.map { field => + normalizeFieldName(field.name) -> field + }.toMap + val reorder = i.userSpecifiedCols.map { col => + colNamesToFields.get(normalizeFieldName(col)) + } + if (reorder.forall(_.isDefined)) { + i.withNewChildren(Seq(resolveColumnDefault(i.query, reorder.flatten))) + } else { + i + } + } + + case _ => plan + } + + private def resolveColumnDefault( + plan: LogicalPlan, + expectedQuerySchema: Seq[StructField], + acceptProject: Boolean = true, + acceptInlineTable: Boolean = true): LogicalPlan = { + plan match { + case _: SubqueryAlias => + plan.mapChildren( + resolveColumnDefault(_, expectedQuerySchema, acceptProject, acceptInlineTable)) + + case _: GlobalLimit | _: LocalLimit | _: Offset | _: Sort if acceptProject => + plan.mapChildren( + resolveColumnDefault(_, expectedQuerySchema, acceptInlineTable = false)) + + case p: Project if acceptProject && p.child.resolved && + p.containsPattern(UNRESOLVED_ATTRIBUTE) && + p.projectList.length <= expectedQuerySchema.length => + val newProjectList = p.projectList.zipWithIndex.map { + case (u: UnresolvedAttribute, i) if isExplicitDefaultColumn(u) => + val field = expectedQuerySchema(i) + Alias(getDefaultValueExpr(field).getOrElse(Literal(null, field.dataType)), u.name)() + case (other, _) if containsExplicitDefaultColumn(other) => + throw QueryCompilationErrors + .defaultReferencesNotAllowedInComplexExpressionsInInsertValuesList() + case (other, _) => other + } + val newChild = resolveColumnDefault(p.child, expectedQuerySchema, acceptProject = false) + val newProj = p.copy(projectList = newProjectList, child = newChild) + newProj.copyTagsFrom(p) + newProj + + case _: Project | _: Aggregate if acceptInlineTable => + plan.mapChildren(resolveColumnDefault(_, expectedQuerySchema, acceptProject = false)) + + case inlineTable: UnresolvedInlineTable if acceptInlineTable && + inlineTable.containsPattern(UNRESOLVED_ATTRIBUTE) && + inlineTable.rows.forall(exprs => exprs.length <= expectedQuerySchema.length) => + val newRows = inlineTable.rows.map { exprs => + exprs.zipWithIndex.map { + case (u: UnresolvedAttribute, i) if isExplicitDefaultColumn(u) => + val field = expectedQuerySchema(i) + getDefaultValueExpr(field).getOrElse(Literal(null, field.dataType)) Review Comment: optional: should we add a boolean argument to `getDefaultValueExprOrNullLiteral` to switch the behavior between the two modes? -- 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]
