rdblue commented on a change in pull request #1852: URL: https://github.com/apache/iceberg/pull/1852#discussion_r533553144
########## File path: spark3-extensions/src/main/scala/org/apache/spark/sql/catalyst/optimizer/RewriteDelete.scala ########## @@ -0,0 +1,152 @@ +/* + * 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.optimizer + +import java.util.UUID +import org.apache.spark.internal.Logging +import org.apache.spark.sql.{sources, AnalysisException} +import org.apache.spark.sql.catalyst.expressions.{Alias, Ascending, Attribute, AttributeReference, EqualNullSafe, Expression, InputFileName, Literal, Not, PredicateHelper, SortOrder, SubqueryExpression} +import org.apache.spark.sql.catalyst.plans.logical.{Aggregate, DeleteFromTable, DynamicFileFilter, Filter, LogicalPlan, Project, RepartitionByExpression, ReplaceData, Sort} +import org.apache.spark.sql.catalyst.rules.Rule +import org.apache.spark.sql.connector.catalog.Table +import org.apache.spark.sql.connector.iceberg.catalog.ExtendedSupportsDelete +import org.apache.spark.sql.connector.iceberg.read.SupportsFileFilter +import org.apache.spark.sql.connector.iceberg.write.MergeBuilder +import org.apache.spark.sql.connector.write.{LogicalWriteInfo, LogicalWriteInfoImpl} +import org.apache.spark.sql.execution.datasources.DataSourceStrategy +import org.apache.spark.sql.execution.datasources.v2.{DataSourceV2Relation, DataSourceV2ScanRelation, PushDownUtils} +import org.apache.spark.sql.internal.SQLConf +import org.apache.spark.sql.types.{BooleanType, StructType} +import org.apache.spark.sql.util.CaseInsensitiveStringMap + +// TODO: should be part of early scan push down after the delete condition is optimized +object RewriteDelete extends Rule[LogicalPlan] with PredicateHelper with Logging { + + import org.apache.spark.sql.execution.datasources.v2.ExtendedDataSourceV2Implicits._ + + private val FILE_NAME_COL = "_file" + + override def apply(plan: LogicalPlan): LogicalPlan = plan resolveOperators { + // don't rewrite deletes that can be answered by passing filters to deleteWhere in SupportsDelete + case d @ DeleteFromTable(r: DataSourceV2Relation, Some(cond)) if isDeleteWhereCase(r, cond) => + d + + // rewrite all operations that require reading the table to delete records + case DeleteFromTable(r: DataSourceV2Relation, Some(cond)) => + // TODO: do a switch based on whether we get BatchWrite or DeltaBatchWrite + val writeInfo = newWriteInfo(r.schema) + val mergeBuilder = r.table.asMergeable.newMergeBuilder(writeInfo) + + val scanPlan = buildScanPlan(r.table, r.output, mergeBuilder, cond) + + val remainingRowFilter = Not(EqualNullSafe(cond, Literal(true, BooleanType))) + val remainingRowsPlan = Filter(remainingRowFilter, scanPlan) + + val batchWrite = mergeBuilder.asWriteBuilder.buildForBatch() + val writePlan = buildWritePlan(remainingRowsPlan, r.output) + ReplaceData(r, batchWrite, writePlan) + } + + private def buildScanPlan( + table: Table, + output: Seq[AttributeReference], + mergeBuilder: MergeBuilder, + cond: Expression): LogicalPlan = { + + val scanBuilder = mergeBuilder.asScanBuilder + + val predicates = splitConjunctivePredicates(cond) + val normalizedPredicates = DataSourceStrategy.normalizeExprs(predicates, output) + PushDownUtils.pushFilters(scanBuilder, normalizedPredicates) + + val scan = scanBuilder.build() + val scanRelation = DataSourceV2ScanRelation(table, scan, output) + + val scanPlan = scan match { + case _: SupportsFileFilter => + val matchingFilePlan = buildFileFilterPlan(cond, scanRelation) + val dynamicFileFilter = DynamicFileFilter(scanRelation, matchingFilePlan) + dynamicFileFilter + case _ => + scanRelation + } + + // include file name so that we can group data back + val fileNameExpr = Alias(InputFileName(), FILE_NAME_COL)() + Project(scanPlan.output :+ fileNameExpr, scanPlan) + } + + private def buildWritePlan( + remainingRowsPlan: LogicalPlan, + output: Seq[AttributeReference]): LogicalPlan = { + + // TODO: sort by _pos to keep the original ordering of rows + // TODO: consider setting a file size limit Review comment: The problem with a soft limit is that you have a hard limit somewhere. You could have 1.2 GB of data and the soft limit would make you cut at 1.1 GB. I think the issue you can't avoid is not knowing how much more data there is coming from the writer. I would not spend time worrying about more complicated logic here. We can still tune the files after the fact with rewrites. This problem also has less of an impact if we get the _pos changes done, so I would focus effort there. Keeping the original order will help keep files the same size, and we can always group just the rows from one input file into an output file. ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
