cloud-fan commented on a change in pull request #35657: URL: https://github.com/apache/spark/pull/35657#discussion_r840698389
########## File path: sql/core/src/test/scala/org/apache/spark/sql/connector/KeyGroupedPartitioningSuite.scala ########## @@ -0,0 +1,474 @@ +/* + * 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.connector + +import java.util.Collections + +import org.apache.spark.sql.{DataFrame, Row} +import org.apache.spark.sql.catalyst.InternalRow +import org.apache.spark.sql.catalyst.expressions.{Ascending, SortOrder => V1SortOrder, TransformExpression} +import org.apache.spark.sql.catalyst.plans.{physical => v1} +import org.apache.spark.sql.catalyst.plans.physical.{ClusteredDistribution, KeyGroupedPartitioning} +import org.apache.spark.sql.connector.catalog.Identifier +import org.apache.spark.sql.connector.catalog.InMemoryTableCatalog +import org.apache.spark.sql.connector.catalog.functions._ +import org.apache.spark.sql.connector.distributions.Distribution +import org.apache.spark.sql.connector.distributions.Distributions +import org.apache.spark.sql.connector.expressions._ +import org.apache.spark.sql.connector.expressions.Expressions._ +import org.apache.spark.sql.execution.SparkPlan +import org.apache.spark.sql.execution.datasources.v2.BatchScanExec +import org.apache.spark.sql.execution.datasources.v2.DataSourceV2ScanRelation +import org.apache.spark.sql.execution.exchange.ShuffleExchangeExec +import org.apache.spark.sql.execution.joins.SortMergeJoinExec +import org.apache.spark.sql.internal.SQLConf +import org.apache.spark.sql.internal.SQLConf._ +import org.apache.spark.sql.types._ + +class KeyGroupedPartitioningSuite extends DistributionAndOrderingSuiteBase { + private var originalV2BucketingEnabled: Boolean = false + private var originalAutoBroadcastJoinThreshold: Long = -1 + + override def beforeAll(): Unit = { + super.beforeAll() + originalV2BucketingEnabled = conf.getConf(V2_BUCKETING_ENABLED) + conf.setConf(V2_BUCKETING_ENABLED, true) + originalAutoBroadcastJoinThreshold = conf.getConf(AUTO_BROADCASTJOIN_THRESHOLD) + conf.setConf(AUTO_BROADCASTJOIN_THRESHOLD, -1L) + } + + override def afterAll(): Unit = { + try { + super.afterAll() + } finally { + conf.setConf(V2_BUCKETING_ENABLED, originalV2BucketingEnabled) + conf.setConf(AUTO_BROADCASTJOIN_THRESHOLD, originalAutoBroadcastJoinThreshold) + } + } + + before { + Seq(UnboundYearsFunction, UnboundDaysFunction, UnboundBucketFunction).foreach { f => + catalog.createFunction(Identifier.of(Array.empty, f.name()), f) + } + } + + after { + catalog.clearTables() + catalog.clearFunctions() + } + + private val emptyProps: java.util.Map[String, String] = { + Collections.emptyMap[String, String] + } + private val table: String = "tbl" + private val schema = new StructType() + .add("id", IntegerType) + .add("data", StringType) + .add("ts", TimestampType) + + test("clustered distribution: output partitioning should be KeyGroupedPartitioning") { + val partitions: Array[Transform] = Array(Expressions.years("ts")) + + // create a table with 3 partitions, partitioned by `years` transform + createTable(table, schema, partitions, + Distributions.clustered(partitions.map(_.asInstanceOf[Expression]))) + sql(s"INSERT INTO testcat.ns.$table VALUES " + + s"(0, 'aaa', CAST('2022-01-01' AS timestamp)), " + + s"(1, 'bbb', CAST('2021-01-01' AS timestamp)), " + + s"(2, 'ccc', CAST('2020-01-01' AS timestamp))") + + var df = sql(s"SELECT count(*) FROM testcat.ns.$table GROUP BY ts") + val v1Distribution = v1.ClusteredDistribution( + Seq(TransformExpression(YearsFunction, Seq(attr("ts"))))) + val partitionValues = Seq(50, 51, 52).map(v => InternalRow.fromSeq(Seq(v))) + + checkQueryPlan(df, v1Distribution, + KeyGroupedPartitioning(v1Distribution.clustering, partitionValues)) + + // multiple group keys should work too as long as partition keys are subset of them + df = sql(s"SELECT count(*) FROM testcat.ns.$table GROUP BY id, ts") + checkQueryPlan(df, v1Distribution, + KeyGroupedPartitioning(v1Distribution.clustering, partitionValues)) + } + + test("non-clustered distribution: fallback to super.partitioning") { + val partitions: Array[Transform] = Array(years("ts")) + val ordering: Array[SortOrder] = Array(sort(FieldReference("ts"), + SortDirection.ASCENDING, NullOrdering.NULLS_FIRST)) + + createTable(table, schema, partitions, Distributions.ordered(ordering), ordering) + sql(s"INSERT INTO testcat.ns.$table VALUES " + + s"(0, 'aaa', CAST('2022-01-01' AS timestamp)), " + + s"(1, 'bbb', CAST('2021-01-01' AS timestamp)), " + + s"(2, 'ccc', CAST('2020-01-01' AS timestamp))") + + val df = sql(s"SELECT * FROM testcat.ns.$table") + val v1Ordering = Seq(V1SortOrder(attr("ts"), Ascending)) + val v1Distribution = v1.OrderedDistribution(v1Ordering) + + checkQueryPlan(df, v1Distribution, v1.UnknownPartitioning(0)) + } + + test("non-clustered distribution: no partition") { + val partitions: Array[Transform] = Array(bucket(32, "ts")) + createTable(table, schema, partitions, + Distributions.clustered(partitions.map(_.asInstanceOf[Expression]))) + + val df = sql(s"SELECT * FROM testcat.ns.$table") + val distribution = v1.ClusteredDistribution( + Seq(TransformExpression(BucketFunction, Seq(attr("ts")), Some(32)))) + + checkQueryPlan(df, distribution, v1.UnknownPartitioning(0)) + } + + test("non-clustered distribution: single partition") { + val partitions: Array[Transform] = Array(bucket(32, "ts")) + createTable(table, schema, partitions, + Distributions.clustered(partitions.map(_.asInstanceOf[Expression]))) + sql(s"INSERT INTO testcat.ns.$table VALUES (0, 'aaa', CAST('2020-01-01' AS timestamp))") + + val df = sql(s"SELECT * FROM testcat.ns.$table") + val distribution = v1.ClusteredDistribution( + Seq(TransformExpression(BucketFunction, Seq(attr("ts")), Some(32)))) + + checkQueryPlan(df, distribution, v1.SinglePartition) + } + + test("non-clustered distribution: no V2 catalog") { + spark.conf.set("spark.sql.catalog.testcat2", classOf[InMemoryTableCatalog].getName) + val nonFunctionCatalog = spark.sessionState.catalogManager.catalog("testcat2") + .asInstanceOf[InMemoryTableCatalog] + val partitions: Array[Transform] = Array(bucket(32, "ts")) + createTable(table, schema, partitions, + Distributions.clustered(partitions.map(_.asInstanceOf[Expression])), + catalog = nonFunctionCatalog) + sql(s"INSERT INTO testcat2.ns.$table VALUES " + + s"(0, 'aaa', CAST('2022-01-01' AS timestamp)), " + + s"(1, 'bbb', CAST('2021-01-01' AS timestamp)), " + + s"(2, 'ccc', CAST('2020-01-01' AS timestamp))") + + val df = sql(s"SELECT * FROM testcat2.ns.$table") + val distribution = v1.UnspecifiedDistribution + + try { + checkQueryPlan(df, distribution, v1.UnknownPartitioning(0)) + } finally { + spark.conf.unset("spark.sql.catalog.testcat2") + } + } + + test("non-clustered distribution: no V2 function provided") { + catalog.clearFunctions() + + val partitions: Array[Transform] = Array(bucket(32, "ts")) + createTable(table, schema, partitions, + Distributions.clustered(partitions.map(_.asInstanceOf[Expression]))) + sql(s"INSERT INTO testcat.ns.$table VALUES " + + s"(0, 'aaa', CAST('2022-01-01' AS timestamp)), " + + s"(1, 'bbb', CAST('2021-01-01' AS timestamp)), " + + s"(2, 'ccc', CAST('2020-01-01' AS timestamp))") + + val df = sql(s"SELECT * FROM testcat.ns.$table") + val distribution = v1.UnspecifiedDistribution + + checkQueryPlan(df, distribution, v1.UnknownPartitioning(0)) + } + + test("non-clustered distribution: V2 bucketing disabled") { + withSQLConf(SQLConf.V2_BUCKETING_ENABLED.key -> "false") { + val partitions: Array[Transform] = Array(bucket(32, "ts")) + createTable(table, schema, partitions, + Distributions.clustered(partitions.map(_.asInstanceOf[Expression]))) + sql(s"INSERT INTO testcat.ns.$table VALUES " + + s"(0, 'aaa', CAST('2022-01-01' AS timestamp)), " + + s"(1, 'bbb', CAST('2021-01-01' AS timestamp)), " + + s"(2, 'ccc', CAST('2020-01-01' AS timestamp))") + + val df = sql(s"SELECT * FROM testcat.ns.$table") + val distribution = v1.ClusteredDistribution( + Seq(TransformExpression(BucketFunction, Seq(attr("ts")), Some(32)))) + + checkQueryPlan(df, distribution, v1.UnknownPartitioning(0)) + } + } + + /** + * Check whether the query plan from `df` has the expected `distribution`, `ordering` and + * `partitioning`. + */ + private def checkQueryPlan( + df: DataFrame, + distribution: v1.Distribution, + partitioning: v1.Partitioning): Unit = { + // check distribution & ordering are correctly populated in logical plan + val relation = df.queryExecution.optimizedPlan.collect { + case r: DataSourceV2ScanRelation => r + }.head + + resolveDistribution(distribution, relation) match { + case ClusteredDistribution(clustering, _, _) => + assert(relation.keyGroupedPartitioning.isDefined && relation.keyGroupedPartitioning.get == clustering) + case _ => + assert(relation.keyGroupedPartitioning.isEmpty) + } + + // check distribution, ordering and output partitioning are correctly populated in physical plan + val scan = collect(df.queryExecution.executedPlan) { + case s: BatchScanExec => s + }.head + + val expectedPartitioning = resolvePartitioning(partitioning, scan) + assert(expectedPartitioning == scan.outputPartitioning) + } + + private def createTable( + table: String, + schema: StructType, + partitions: Array[Transform], + distribution: Distribution = Distributions.unspecified(), + ordering: Array[expressions.SortOrder] = Array.empty, + catalog: InMemoryTableCatalog = catalog): Unit = { + catalog.createTable(Identifier.of(Array("ns"), table), + schema, partitions, emptyProps, distribution, ordering, None) + } + + private val customers: String = "customers" + private val customers_schema = new StructType() + .add("customer_name", StringType) + .add("customer_age", IntegerType) + .add("customer_id", LongType) + + private val orders: String = "orders" + private val orders_schema = new StructType() + .add("order_amount", DoubleType) + .add("customer_id", LongType) + + private def testWithCustomersAndOrders( + customers_partitions: Array[Transform], + customers_distribution: Distribution, + orders_partitions: Array[Transform], + orders_distribution: Distribution, + expectedNumOfShuffleExecs: Int): Unit = { + createTable(customers, customers_schema, customers_partitions, customers_distribution) + sql(s"INSERT INTO testcat.ns.$customers VALUES " + + s"('aaa', 10, 1), ('bbb', 20, 2), ('ccc', 30, 3)") + + createTable(orders, orders_schema, orders_partitions, orders_distribution) + sql(s"INSERT INTO testcat.ns.$orders VALUES " + + s"(100.0, 1), (200.0, 1), (150.0, 2), (250.0, 2), (350.0, 2), (400.50, 3)") + + val df = sql("SELECT customer_name, customer_age, order_amount " + + s"FROM testcat.ns.$customers c JOIN testcat.ns.$orders o " + + "ON c.customer_id = o.customer_id ORDER BY c.customer_id, order_amount") + + val shuffles = collectShuffles(df.queryExecution.executedPlan) + assert(shuffles.length == expectedNumOfShuffleExecs) + + checkAnswer(df, + Seq(Row("aaa", 10, 100.0), Row("aaa", 10, 200.0), Row("bbb", 20, 150.0), + Row("bbb", 20, 250.0), Row("bbb", 20, 350.0), Row("ccc", 30, 400.50))) + } + + private def collectShuffles(plan: SparkPlan): Seq[ShuffleExchangeExec] = { + collect(plan) { + case s: SortMergeJoinExec => s Review comment: why don't we collect `ShuffleExchangeExec` directly here? -- 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]
