cloud-fan commented on code in PR #57264: URL: https://github.com/apache/spark/pull/57264#discussion_r3589525981
########## sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/ResolveAsOfJoin.scala: ########## @@ -0,0 +1,202 @@ +/* + * 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.{ + Expression, + RowOrdering, + SubqueryExpression, + WindowExpression +} +import org.apache.spark.sql.catalyst.expressions.AttributeSet +import org.apache.spark.sql.catalyst.expressions.aggregate.AggregateExpression +import org.apache.spark.sql.catalyst.plans.logical.{AsOfJoin, LogicalPlan, Project} +import org.apache.spark.sql.catalyst.rules.Rule +import org.apache.spark.sql.catalyst.trees.TreePattern.{AS_OF_JOIN, GENERATOR} +import org.apache.spark.sql.catalyst.util._ +import org.apache.spark.sql.errors.QueryErrorsBase +import org.apache.spark.sql.types.{ArrayType, DataType, DatetimeType, StringType, StructType} + +/** + * Resolves SQL [[AsOfJoin]] operators: materializes `MATCH_CONDITION` into `asOfCondition` and + * `orderExpression`, and expands `USING` column lists into equi-join predicates. + */ +object ResolveAsOfJoin extends Rule[LogicalPlan] with SQLConfHelper { + + override def apply(plan: LogicalPlan): LogicalPlan = plan.resolveOperatorsUpWithPruning( + _.containsPattern(AS_OF_JOIN), ruleId) { + case j @ AsOfJoin( + left, + right, + _, + condition, + _, + _, + _, + usingColumns, + matchLeft, + matchOp, + matchRight, + _, + _, + _) + if left.resolved && right.resolved && condition.forall(_.resolved) => + val (joinBase, usingProjection) = usingColumns match { + case Some(cols) if condition.isEmpty => + val (projectList, hiddenList, newCondition) = + NaturalAndUsingJoinResolution.computeJoinOutputsAndNewCondition( + left, + left.output, + right, + right.output, + j.joinType, + cols, + None, + (l, r) => conf.resolver(l, r)) + (j.copy(condition = newCondition, usingColumns = None), Some((projectList, hiddenList))) + case _ => (j, None) + } + val resolvedJoin = (matchLeft, matchOp, matchRight) match { + case (Some(leftExpr), Some(operator), Some(rightExpr)) => + AsOfJoinValidation.validateMatchConditionTableReferences( + joinBase, left, right, leftExpr, rightExpr) + if (leftExpr.resolved && rightExpr.resolved) { + AsOfJoinValidation.validateMatchConditionOperands(joinBase, leftExpr, rightExpr) + val (asOfCondition, orderExpression, leftSortExprs, rightSortExprs) = + AsOfJoin.resolveMatchComparison(left, right, leftExpr, operator, rightExpr) + joinBase.copy( + asOfCondition = asOfCondition, + orderExpression = orderExpression, + leftSortExprs = leftSortExprs, + rightSortExprs = rightSortExprs, + matchLeftOperand = None, + matchOperator = None, + matchRightOperand = None) + } else { + joinBase + } + case (None, None, None) => joinBase + case _ => joinBase + } + usingProjection match { + case Some((projectList, hiddenList)) => + val project = Project(projectList, resolvedJoin) + project.setTagValue( + Project.hiddenOutputTag, + hiddenList.map(_.markAsQualifiedAccessOnly())) + project + case None => resolvedJoin + } + } +} + +private[analysis] object AsOfJoinValidation extends QueryErrorsBase { + + def validateMatchConditionTableReferences( + join: AsOfJoin, + left: LogicalPlan, + right: LogicalPlan, + leftExpr: Expression, + rightExpr: Expression): Unit = { + val leftSet = left.outputSet + val rightSet = right.outputSet + + def referencesBothJoinSides(refs: AttributeSet): Boolean = { + refs.nonEmpty && + refs.intersect(leftSet).nonEmpty && + refs.intersect(rightSet).nonEmpty + } + + val leftRefs = leftExpr.references + val rightRefs = rightExpr.references + if (referencesBothJoinSides(leftRefs) || referencesBothJoinSides(rightRefs)) { + join.failAnalysis( + errorClass = "ASOF_JOIN_MATCH_CONDITION_TABLE_REFERENCE", + messageParameters = Map( + "refs1" -> toSQLExpr(leftExpr), + "refs2" -> toSQLExpr(rightExpr))) + } + } + + def validateMatchConditionOperands( + join: AsOfJoin, + leftExpr: Expression, + rightExpr: Expression): Unit = { + Seq(leftExpr, rightExpr).foreach { expr => + findInvalidMatchConditionExpression(expr).foreach { invalidExpr => + join.failAnalysis( + errorClass = "ASOF_JOIN_MATCH_CONDITION_INVALID_EXPRESSION", + messageParameters = Map("expr" -> toSQLExpr(invalidExpr))) + } + } + if (!RowOrdering.isOrderable(leftExpr.dataType) || + !RowOrdering.isOrderable(rightExpr.dataType) || Review Comment: This type check accepts any orderable, type-compatible operands, but the only planner consumer can't handle all of them. When `sortMergeAsOfJoin.enabled` is on, a SQL AsOfJoin isn't rewritten and reaches `AsOfJoinSelection.findFromOrder` (SparkStrategies.scala:514-528), which only recognizes `Subtract` / `If(_, Subtract, _)`. But a plain STRING operand produces `If(EqualTo, 0, If(GreaterThan, 1, -1))` (via `buildSignedComparisonDistance`), and struct/array operands produce `CreateStruct`/`ZipWith` — none of which `findFromOrder` matches. So e.g. `MATCH_CONDITION (t.symbol >= q.symbol)` (both confs on) passes analysis, skips the rewrite, then throws `IllegalStateException("Cannot extract as-of keys from order expression: ...")` at planning. I traced this from source but didn't run a build — can you confirm the planning-time throw? If it reproduces, gating the operand types this analysis supports down to what the physical path can consume (with a clear analysis error for the rest) until the composite/multi-column sort-merge operator lands would keep supported-syntax queries from failing with an internal error. ########## sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/logical/basicLogicalOperators.scala: ########## @@ -2563,6 +2566,8 @@ case class AsOfJoin( matchOperator.isEmpty && matchRightOperand.isEmpty && expressions.forall(_.resolved) && + leftSortExprs.forall(_.resolved) && Review Comment: These two clauses are redundant: `expressions.forall(_.resolved)` on the line above already covers `leftSortExprs`/`rightSortExprs`, since `QueryPlan._expressions` recurses into `Seq[Expression]` case-class fields. Harmless, but can be dropped. ########## sql/core/src/test/scala/org/apache/spark/sql/AsOfJoinSQLSuite.scala: ########## @@ -38,6 +38,19 @@ class AsOfJoinSQLSuite extends QueryTest with SharedSparkSession { super.afterAll() } + private def setupTradeQuoteViews(): Unit = { Review Comment: All the added tests are negative; the one test with a valid `MATCH_CONDITION` ("requires sort-merge conf") asserts failure with the conf off. A positive analyze-only test with `sortMergeAsOfJoin.enabled=true` — asserting a valid condition resolves to the expected plan, no execution backend needed — would cover the resolution path directly, and a case with a STRING (or struct) operand would have surfaced the planner-capability gap flagged in `ResolveAsOfJoin`. -- 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]
