gengliangwang commented on code in PR #55508: URL: https://github.com/apache/spark/pull/55508#discussion_r3140713642
########## sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/ResolveChangelogTable.scala: ########## @@ -0,0 +1,311 @@ +/* + * 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.expressions._ +import org.apache.spark.sql.catalyst.expressions.aggregate.{Count, Max, Min} +import org.apache.spark.sql.catalyst.plans.logical._ +import org.apache.spark.sql.catalyst.rules.Rule +import org.apache.spark.sql.catalyst.streaming.StreamingRelationV2 +import org.apache.spark.sql.connector.catalog.{Changelog, ChangelogInfo} +import org.apache.spark.sql.errors.QueryCompilationErrors +import org.apache.spark.sql.execution.datasources.v2.{ChangelogTable, DataSourceV2Relation} +import org.apache.spark.sql.types.{IntegerType, StringType} + +/** + * Post-processes a resolved [[ChangelogTable]] read to apply CDC option semantics + * (carry-over removal, update detection) and to enforce supported option combinations. + * + * Fires after [[ResolveRelations]] has wrapped the connector's [[Changelog]] in a + * [[ChangelogTable]]. Both batch ([[DataSourceV2Relation]]) and streaming + * ([[StreamingRelationV2]]) reads are handled: + * - Batch: the requested post-processing passes are injected as logical operators on top + * of the relation. Carry-over removal and update detection are fused into a single + * pass over a (rowId, _commit_version)-partitioned Window: the Filter drops CoW + * carry-over pairs (same rowVersion on both sides) and the subsequent Project relabels + * real delete+insert pairs as update_preimage / update_postimage. + * - Streaming: post-processing is not yet supported. If the requested options would + * require any post-processing, the rule throws an explicit [[AnalysisException]] to + * prevent silent wrong results. Streams that don't require post-processing pass + * through unchanged. + * + * Net change computation (`deduplicationMode = netChanges`) is not yet implemented and + * is rejected up-front for both batch and streaming. + */ +object ResolveChangelogTable extends Rule[LogicalPlan] { + + /** + * Reserved (`__spark_cdc_*`) column names used internally by post-processing; + * connectors must not emit columns with these names. + */ + object HelperColumn { + final val DelCnt = "__spark_cdc_del_cnt" + final val InsCnt = "__spark_cdc_ins_cnt" + final val MinRv = "__spark_cdc_min_rv" + final val MaxRv = "__spark_cdc_max_rv" + final val RvCnt = "__spark_cdc_rv_cnt" + + val all: Set[String] = Set(DelCnt, InsCnt, MinRv, MaxRv, RvCnt) + } + + override def apply(plan: LogicalPlan): LogicalPlan = plan.resolveOperatorsUp { + case rel @ DataSourceV2Relation(table: ChangelogTable, _, _, _, _, _) if !table.resolved => + val changelog = table.changelog + val req = evaluateRequirements(changelog, table.changelogInfo) + + val resolvedRel = rel.copy(table = table.copy(resolved = true)) + var updatedRel: LogicalPlan = resolvedRel + if (req.requiresCarryOverRemoval || req.requiresUpdateDetection) { + updatedRel = addRowLevelPostProcessing( + resolvedRel, changelog, req.requiresCarryOverRemoval, req.requiresUpdateDetection) + } + if (req.requiresNetChanges) { + updatedRel = injectNetChangeComputation(updatedRel, changelog) + } + updatedRel + + case rel @ StreamingRelationV2(_, _, table: ChangelogTable, _, _, _, _, _, _) + if !table.resolved => + // Streaming CDC reads do not yet apply post-processing. Run the same option / + // capability validation as the batch path so silent wrong results are impossible: + // either no post-processing would be required (fall through, return raw stream), + // or we throw an explicit AnalysisException. + val changelog = table.changelog + val req = evaluateRequirements(changelog, table.changelogInfo) + if (req.needsAny) { + throw QueryCompilationErrors.cdcStreamingPostProcessingNotSupported(changelog.name()) + } + rel.copy(table = table.copy(resolved = true)) + } + + // --------------------------------------------------------------------------- + // Option validation & Requirement Computation + // --------------------------------------------------------------------------- + + /** + * Captures which post-processing passes a CDC query requires, derived from the + * user-provided [[ChangelogInfo]] options and the connector-declared [[Changelog]] + * capability flags. + */ + private case class PostProcessingRequirements( + requiresCarryOverRemoval: Boolean, + requiresUpdateDetection: Boolean, + requiresNetChanges: Boolean) { + def needsAny: Boolean = + requiresCarryOverRemoval || requiresUpdateDetection || requiresNetChanges + } + + /** + * Validates CDC option/capability combinations and computes which post-processing + * passes are required. Throws an [[org.apache.spark.sql.AnalysisException]] for + * unsupported or contradictory combinations (currently: `netChanges` deduplication, + * and `computeUpdates` with surfaced carry-overs but no carry-over removal). + */ + private def evaluateRequirements( + changelog: Changelog, + options: ChangelogInfo): PostProcessingRequirements = { + // Net change computation is not yet implemented. + if (options.deduplicationMode() == ChangelogInfo.DeduplicationMode.NET_CHANGES) { + throw QueryCompilationErrors.cdcNetChangesNotYetSupported(changelog.name()) + } + + val requiresCarryOverRemoval = + options.deduplicationMode() != ChangelogInfo.DeduplicationMode.NONE && + changelog.containsCarryoverRows() + val requiresUpdateDetection = + options.computeUpdates() && changelog.representsUpdateAsDeleteAndInsert() + val requiresNetChanges = + options.deduplicationMode() == ChangelogInfo.DeduplicationMode.NET_CHANGES && + changelog.containsIntermediateChanges() + + // If carry-overs are surfaced and update detection is enabled without carry-over + // removal, carry-overs would be falsely classified as updates, leading to wrong + // results. Hence we throw. + if (requiresUpdateDetection && + changelog.containsCarryoverRows() && + options.deduplicationMode() == ChangelogInfo.DeduplicationMode.NONE) { + throw QueryCompilationErrors.cdcUpdateDetectionRequiresCarryOverRemoval( + changelog.name()) + } + + PostProcessingRequirements( + requiresCarryOverRemoval, requiresUpdateDetection, requiresNetChanges) + } + + // --------------------------------------------------------------------------- + // Row Level Post Processing (Update Detection & Carry-over Removal) + // --------------------------------------------------------------------------- + + /** + * Adds row-level post-processing (carry-over removal and/or update detection) on top of + * the given plan. `counts` = per-partition delete and insert change_type row counts over + * `(rowId, _commit_version)`. `rv bounds` = per-partition min/max of `rowVersion`. + * Equal bounds signal a copy-on-write carry-over. + * - both active -> Window(counts + rv bounds) -> Filter -> Project(relabel) -> Drop helpers Review Comment: Nit: this summary hasn't been updated to reflect the `_rv_cnt = 2` NULL-safety clause. `_min_rv = _max_rv` alone no longer signals a carry-over — both sides must also be non-null. ```suggestion * Adds row-level post-processing (carry-over removal and/or update detection) on top of * the given plan. `counts` = per-partition delete and insert change_type row counts over * `(rowId, _commit_version)`. `rv bounds + rv count` = per-partition min/max of `rowVersion` * plus the non-null count; equal bounds with both sides non-null signal a copy-on-write * carry-over. ``` -- 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]
