AnishMahto commented on code in PR #56016: URL: https://github.com/apache/spark/pull/56016#discussion_r3291876465
########## sql/pipelines/src/main/scala/org/apache/spark/sql/pipelines/autocdc/Scd1ForeachBatchExec.scala: ########## @@ -0,0 +1,72 @@ +/* + * 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.pipelines.autocdc + +import org.apache.spark.sql.catalyst.TableIdentifier +import org.apache.spark.sql.classic.DataFrame + +/** + * Exposes an API to execute one SCD Type 1 AutoCDC microbatch reconciliation on a + * foreachBatch streaming query. + */ +case class Scd1ForeachBatchExec( + batchProcessor: Scd1BatchProcessor, + auxiliaryTableIdentifier: TableIdentifier, + targetTableIdentifier: TableIdentifier) { + + /** + * Process a single CDC microbatch and merge it into the auxiliary and target tables. + */ + def execute(batchDf: DataFrame, batchId: Long): Unit = { + ScdBatchValidator( + destinationIdentifier = targetTableIdentifier, + changeArgs = batchProcessor.changeArgs, + batchDf = batchDf, + batchId = batchId + ).validateMicrobatch() + + val deduplicatedMicrobatch = batchProcessor.deduplicateMicrobatch( + validatedMicrobatch = batchDf + ) + + val microbatchWithCdcMetadata = batchProcessor.extendMicrobatchRowsWithCdcMetadata( + validatedMicrobatch = deduplicatedMicrobatch + ) + + val projectedMicrobatch = batchProcessor.projectTargetColumnsOntoMicrobatch( + microbatchWithCdcMetadataDf = microbatchWithCdcMetadata + ) + + val reconciledMicrobatch = batchProcessor.applyTombstonesToMicrobatch( + microbatchDf = projectedMicrobatch, + auxiliaryTableDf = batchDf.sparkSession.read.table( Review Comment: Yep this is a good question. We do indeed expect the auxiliary table to be small, _especially_ for SCD1 - there can be at most one tombstone per key in the universe of possible keys, and if a row is upserted to post deletion, the tombstone is GC'd. It _is_ worth mentioning if a row is deleted in the upstream source and never touched again (not totally uncommon, i.e row is permanently deleted), its tombstone will continue to live on indefinitely - necessary for correctness, but there are future paths where we can consider a TTL for tombstone rows to eventually clean them up. This would be a correctness vs time/space efficiency tradeoff. Anyhow given that we expect the auxiliary table to generally be small, we should expect this join to typically use a broadcast join - should be relatively fast. That being said I agree there's room for spark engine based optimization such as pruning/clustering for rarer cases where the auxiliary table does grow larger in size. I'll leave a follow up comment. -- 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]
