szehon-ho commented on code in PR #55969: URL: https://github.com/apache/spark/pull/55969#discussion_r3276377359
########## sql/pipelines/src/main/scala/org/apache/spark/sql/pipelines/autocdc/Scd1BatchProcessor.scala: ########## @@ -0,0 +1,67 @@ +/* + * 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.{functions => F} +import org.apache.spark.sql.catalyst.util.QuotingUtils +import org.apache.spark.sql.classic.DataFrame +import org.apache.spark.util.ArrayImplicits._ + +/** + * Per-microbatch processor for SCD Type 1 AutoCDC flows, complying to the specified [[changeArgs]] + * configuration. + */ +case class Scd1BatchProcessor(changeArgs: ChangeArgs) { + /** + * Deduplicate the incoming CDC microbatch by key, keeping the most recent event per key + * as ordered by [[ChangeArgs.sequencing]]. + * + * For SCD1 we only care about the most recent (by sequence value) event per key. When + * multiple events share the same key and the same sequence value, the row selected is + * non-deterministic and undefined. + * + * @param validatedMicrobatch A microbatch that has already been validated such that the + * sequencing column should not contain null values, and its data type + * should support ordering. + * + * The schema of the returned dataframe matches the schema of the microbatch exactly. + */ + def deduplicateMicrobatch(validatedMicrobatch: DataFrame): DataFrame = { + // The `max_by` API can only return a single column, so pack/unpack the entire row into a + // temporary column before and after the `max_by` operation. + val winningRowCol = Scd1BatchProcessor.winningRowColName + + val allMicrobatchColumns = + validatedMicrobatch.columns + .map(colName => F.col(QuotingUtils.quoteIdentifier(colName))) + .toImmutableArraySeq + + validatedMicrobatch + .groupBy(changeArgs.keys.map(k => F.col(k.quoted)): _*) + .agg( + F.max_by(F.struct(allMicrobatchColumns: _*), changeArgs.sequencing) + .as(winningRowCol) + ) + .select(F.col(s"$winningRowCol.*")) + } +} + +object Scd1BatchProcessor { + // Columns prefixed with `__spark_autocdc_` are reserved for internal SDP AutoCDC processing. + private val winningRowColName = "__spark_autocdc_winning_row" Review Comment: We document `__spark_autocdc_*` as reserved but do not fail fast if the microbatch schema already uses this name (e.g. `__spark_autocdc_winning_row`). That can make the `struct` + `.as(winningRowCol)` plan ambiguous. Suggest validating at microbatch registration or at the start of `deduplicateMicrobatch` (once foreachBatch validation exists): reject any input column whose name starts with `__spark_autocdc_` via `AnalysisException` + a new `AUTOCDC_*` error in `error-conditions.json` (same pattern as other AutoCDC errors on #55836). ########## sql/pipelines/src/test/scala/org/apache/spark/sql/pipelines/autocdc/Scd1BatchProcessorSuite.scala: ########## @@ -0,0 +1,366 @@ +/* + * 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.QueryTest +import org.apache.spark.sql.{functions => F, Row} +import org.apache.spark.sql.classic.DataFrame +import org.apache.spark.sql.test.SharedSparkSession +import org.apache.spark.sql.types._ + +class Scd1BatchProcessorSuite extends QueryTest with SharedSparkSession { + + /** Build a microbatch [[DataFrame]] from explicit rows and an explicit schema. */ + private def microbatchOf(schema: StructType)(rows: Row*): DataFrame = + spark.createDataFrame(spark.sparkContext.parallelize(rows), schema) + + /** + * Returns the `(name, dataType)` pairs of `schema`'s fields. Used to compare two schemas for + * structural equivalence while deliberately ignoring nullability and metadata, which can shift + * benignly when columns are unpacked from a struct. + */ + private def columnNamesAndDataTypes(schema: StructType): Seq[(String, DataType)] = + schema.fields.map(f => (f.name, f.dataType)).toSeq + + test("deduplicateMicrobatch keeps only the row with the largest sequence value per key") { + val schema = new StructType() + .add("id", IntegerType) + .add("seq", LongType) + .add("value", StringType) + + val batch = microbatchOf(schema)( + Row(1, 10L, "first"), + Row(1, 30L, "winner"), + Row(1, 20L, "middle") + ) + + val processor = Scd1BatchProcessor( + changeArgs = ChangeArgs( + keys = Seq(UnqualifiedColumnName("id")), + sequencing = F.col("seq"), + storedAsScdType = ScdType.Type1 + ) + ) + + checkAnswer( + df = processor.deduplicateMicrobatch(batch), + expectedAnswer = Row(1, 30L, "winner") + ) + } Review Comment: Tests always use `sequencing = F.col("seq")`. Since `ChangeArgs.sequencing` is a `Column` (not just a column name), please add a case with an arbitrary sequencing expression if registration allows it — e.g. `F.col("seq") + F.lit(0L)` or a second column `F.greatest(F.col("seq"), F.col("alt_seq"))` — so we lock in that `max_by(..., changeArgs.sequencing)` works when sequencing is not a bare identifier. ########## sql/pipelines/src/test/scala/org/apache/spark/sql/pipelines/autocdc/Scd1BatchProcessorSuite.scala: ########## @@ -0,0 +1,366 @@ +/* + * 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.QueryTest +import org.apache.spark.sql.{functions => F, Row} +import org.apache.spark.sql.classic.DataFrame +import org.apache.spark.sql.test.SharedSparkSession +import org.apache.spark.sql.types._ + +class Scd1BatchProcessorSuite extends QueryTest with SharedSparkSession { + + /** Build a microbatch [[DataFrame]] from explicit rows and an explicit schema. */ + private def microbatchOf(schema: StructType)(rows: Row*): DataFrame = + spark.createDataFrame(spark.sparkContext.parallelize(rows), schema) + + /** + * Returns the `(name, dataType)` pairs of `schema`'s fields. Used to compare two schemas for + * structural equivalence while deliberately ignoring nullability and metadata, which can shift + * benignly when columns are unpacked from a struct. + */ + private def columnNamesAndDataTypes(schema: StructType): Seq[(String, DataType)] = + schema.fields.map(f => (f.name, f.dataType)).toSeq + + test("deduplicateMicrobatch keeps only the row with the largest sequence value per key") { + val schema = new StructType() + .add("id", IntegerType) + .add("seq", LongType) + .add("value", StringType) + + val batch = microbatchOf(schema)( + Row(1, 10L, "first"), + Row(1, 30L, "winner"), + Row(1, 20L, "middle") + ) + + val processor = Scd1BatchProcessor( + changeArgs = ChangeArgs( + keys = Seq(UnqualifiedColumnName("id")), + sequencing = F.col("seq"), + storedAsScdType = ScdType.Type1 + ) + ) + + checkAnswer( + df = processor.deduplicateMicrobatch(batch), + expectedAnswer = Row(1, 30L, "winner") + ) + } + + test("deduplicateMicrobatch is no-op if there's a single event for a key") { + val schema = new StructType() + .add("id", IntegerType) + .add("seq", LongType) + .add("value", StringType) + + val batch = microbatchOf(schema)( + Row(1, 10L, "only-row") + ) + + val processor = Scd1BatchProcessor( + changeArgs = ChangeArgs( + keys = Seq(UnqualifiedColumnName("id")), + sequencing = F.col("seq"), + storedAsScdType = ScdType.Type1 + ) + ) + + checkAnswer( + df = processor.deduplicateMicrobatch(batch), + expectedAnswer = Row(1, 10L, "only-row") + ) + } + + test("deduplicateMicrobatch handles equal sequencing values for the same key") { Review Comment: Good that this documents the tie contract (one survivor, value non-deterministic). To make that contract harder to regress, consider strengthening the test — e.g. run `deduplicateMicrobatch` twice on the same input and assert `length == 1` both times (and optionally that the survivor is still one of the tied values). We do not need to assert the same row twice, but a second invocation makes "exactly one row per key" explicit beyond a single `collect()`. -- 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]
