JoshRosen commented on code in PR #48661: URL: https://github.com/apache/spark/pull/48661#discussion_r1821458683
########## sql/core/src/test/scala/org/apache/spark/sql/execution/InsertSortForLimitAndOffsetSuite.scala: ########## @@ -0,0 +1,120 @@ +/* + * 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 + +import org.apache.spark.sql.QueryTest +import org.apache.spark.sql.execution.adaptive.AdaptiveSparkPlanHelper +import org.apache.spark.sql.internal.SQLConf +import org.apache.spark.sql.test.SharedSparkSession + +class InsertSortForLimitAndOffsetSuite extends QueryTest + with SharedSparkSession + with AdaptiveSparkPlanHelper { + import testImplicits._ + + private def assertHasTopKSort(plan: SparkPlan): Unit = { + assert(find(plan) { + case _: TakeOrderedAndProjectExec => true + case _ => false + }.isDefined) + } + + private def assertHasCollectLimitExec(plan: SparkPlan): Unit = { + assert(find(plan) { + case _: CollectLimitExec => true + case _ => false + }.isDefined) + } + + private def assertHasGlobalLimitExec(plan: SparkPlan): Unit = { + assert(find(plan) { + case _: GlobalLimitExec => true + case _ => false + }.isDefined) + } + + private def hasLocalSort(plan: SparkPlan): Boolean = { + find(plan) { + case GlobalLimitExec(_, s: SortExec, _) => !s.global + case _ => false + }.isDefined + } + + test("root LIMIT preserves data ordering with top-K sort") { + val df = spark.range(10).orderBy($"id" % 8).limit(2) + df.collect() + val physicalPlan = df.queryExecution.executedPlan + assertHasTopKSort(physicalPlan) + // Extra local sort is not needed for LIMIT with top-K sort optimization. + assert(!hasLocalSort(physicalPlan)) + } + + test("middle LIMIT preserves data ordering with top-K sort") { + val df = spark.range(10).orderBy($"id" % 8).limit(2).distinct() + df.collect() + val physicalPlan = df.queryExecution.executedPlan + assertHasTopKSort(physicalPlan) + // Extra local sort is not needed for LIMIT with top-K sort optimization. + assert(!hasLocalSort(physicalPlan)) + } + + test("root LIMIT preserves data ordering with CollectLimitExec") { + withSQLConf(SQLConf.TOP_K_SORT_FALLBACK_THRESHOLD.key -> "1") { + val df = spark.range(10).orderBy($"id" % 8).limit(2) + df.collect() + val physicalPlan = df.queryExecution.executedPlan + assertHasCollectLimitExec(physicalPlan) + // Extra local sort is not needed for root LIMIT + assert(!hasLocalSort(physicalPlan)) + } + } + + test("middle LIMIT preserves data ordering with the extra sort") { + withSQLConf( + SQLConf.TOP_K_SORT_FALLBACK_THRESHOLD.key -> "1", + // To trigger the bug, we have to disable the coalescing optimization. Otherwise we use only + // one partition to read the range-partition shuffle and there is only one shuffle block for + // the final single-partition shuffle, random fetch order is no longer an issue. + SQLConf.COALESCE_PARTITIONS_ENABLED.key -> "false") { Review Comment: As an aside: Given that AQE is enabled by default in Spark 3.x, I wonder whether many of our unit tests are de-facto running non-distributed execution by quickly reducing complex query plans into single partition plans. If that's the case then we might be missing out on adequate distributed execution coverage in certain tests. As a potential future followup, I wonder whether we should consider a test-only behavior where in most suites AQE coalescing is either disabled or is enabled but with a _floor_ on the number of post-coalesce partitions to always guarantee at least 2 so that we still exercise distributed behaviors. -- 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]
