cloud-fan commented on a change in pull request #35395: URL: https://github.com/apache/spark/pull/35395#discussion_r822656024
########## File path: sql/catalyst/src/test/scala/org/apache/spark/sql/connector/catalog/InMemoryRowLevelOperationTable.scala ########## @@ -0,0 +1,96 @@ +/* + * 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.catalog + +import java.util + +import org.apache.spark.sql.connector.distributions.{Distribution, Distributions} +import org.apache.spark.sql.connector.expressions.{FieldReference, LogicalExpressions, NamedReference, SortDirection, SortOrder, Transform} +import org.apache.spark.sql.connector.read.{Scan, ScanBuilder} +import org.apache.spark.sql.connector.write.{BatchWrite, LogicalWriteInfo, RequiresDistributionAndOrdering, RowLevelOperation, RowLevelOperationBuilder, RowLevelOperationInfo, Write, WriteBuilder, WriterCommitMessage} +import org.apache.spark.sql.connector.write.RowLevelOperation.Command +import org.apache.spark.sql.types.StructType +import org.apache.spark.sql.util.CaseInsensitiveStringMap + +class InMemoryRowLevelOperationTable( + name: String, + schema: StructType, + partitioning: Array[Transform], + properties: util.Map[String, String]) + extends InMemoryTable(name, schema, partitioning, properties) with SupportsRowLevelOperations { + + override def newRowLevelOperationBuilder( + info: RowLevelOperationInfo): RowLevelOperationBuilder = { + () => PartitionBasedOperation(info.command) + } + + case class PartitionBasedOperation(command: Command) extends RowLevelOperation { + private final val PARTITION_COLUMN_REF = FieldReference(PartitionKeyColumn.name) + + var configuredScan: InMemoryBatchScan = _ + + override def requiredMetadataAttributes(): Array[NamedReference] = { + Array(PARTITION_COLUMN_REF) + } + + override def newScanBuilder(options: CaseInsensitiveStringMap): ScanBuilder = { + new InMemoryScanBuilder(schema) { + override def build: Scan = { + val scan = super.build() + configuredScan = scan.asInstanceOf[InMemoryBatchScan] + scan + } + } + } + + override def newWriteBuilder(info: LogicalWriteInfo): WriteBuilder = new WriteBuilder { + + override def build(): Write = new Write with RequiresDistributionAndOrdering { + override def requiredDistribution(): Distribution = { + Distributions.clustered(Array(PARTITION_COLUMN_REF)) + } + + override def requiredOrdering(): Array[SortOrder] = { + Array[SortOrder]( + LogicalExpressions.sort( + PARTITION_COLUMN_REF, + SortDirection.ASCENDING, + SortDirection.ASCENDING.defaultNullOrdering()) + ) + } + + override def toBatch: BatchWrite = PartitionBasedReplaceData(configuredScan) + + override def description(): String = "InMemoryWrite" + } + } + + override def description(): String = "InMemoryPartitionReplaceOperation" + } + + private case class PartitionBasedReplaceData(scan: InMemoryBatchScan) extends TestBatchWrite { + + override def commit(messages: Array[WriterCommitMessage]): Unit = dataMap.synchronized { + val newData = messages.map(_.asInstanceOf[BufferedRows]) + val readRows = scan.data.flatMap(_.asInstanceOf[BufferedRows].rows) Review comment: Is this the place where the writer gets the affected "groups"? It seems to me that the assumption is, we can always push down the delete condition to the data source scan to get the affected groups, which requires two things: 1. we can translate all the catalyst predicates to data source Filters. 2. subqueries can also work as runtime filters. I think the first thing will be true eventually, but the second thing may not always be possible. For example, `DELETE FROM t1 WHERE t1.c IN (SELECT c FROM t2)`. We won't get a runtime filter if `t2` is very big. In my proposal, Spark is responsible to run a query and collect affected "groups". In Delta, the query for the above example is like: `SELECT DISTINCT _file_name FROM t1 WHERE t1.c IN (SELECT c FROM t2)`. This query will be planned as a left semi join, and works with big `t2` as well (use shuffle join instead of broadcast join). The general idea is, Spark should drive the calculating of affected "groups", and data source should just report the "group id". I think this is the only way to support all kinds of UPDATE/DELETE/MERGE conditions: let Spark evaluate the condition. -- 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]
