cboumalh commented on code in PR #52883: URL: https://github.com/apache/spark/pull/52883#discussion_r2722579921
########## sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/aggregate/tupleSketchAgg.scala: ########## @@ -0,0 +1,530 @@ +/* + * 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.expressions.aggregate + +import org.apache.datasketches.tuple.{Sketch, SummaryFactory, SummarySetOperations, Union, UpdatableSketchBuilder, UpdatableSummary} +import org.apache.datasketches.tuple.adouble.{DoubleSummary, DoubleSummaryFactory, DoubleSummarySetOperations} +import org.apache.datasketches.tuple.aninteger.{IntegerSummary, IntegerSummaryFactory, IntegerSummarySetOperations} + +import org.apache.spark.SparkUnsupportedOperationException +import org.apache.spark.sql.catalyst.InternalRow +import org.apache.spark.sql.catalyst.analysis.{ExpressionBuilder, TypeCheckResult} +import org.apache.spark.sql.catalyst.expressions.{Expression, ExpressionDescription, ImplicitCastInputTypes, Literal} +import org.apache.spark.sql.catalyst.plans.logical.{FunctionSignature, InputParameter} +import org.apache.spark.sql.catalyst.trees.QuaternaryLike +import org.apache.spark.sql.catalyst.util.{ArrayData, CollationFactory, SketchSize, SummaryAggregateMode, ThetaSketchUtils, TupleSketchUtils, TupleSummaryMode} +import org.apache.spark.sql.errors.QueryExecutionErrors +import org.apache.spark.sql.internal.types.StringTypeWithCollation +import org.apache.spark.sql.types.{AbstractDataType, ArrayType, BinaryType, DataType, DoubleType, FloatType, IntegerType, LongType, StringType, TypeCollection} +import org.apache.spark.unsafe.types.UTF8String + +/** + * The TupleSketchAggDouble function utilizes a Datasketches TupleSketch instance to count a + * probabilistic approximation of the number of unique values in a given column with associated + * double type summary values that can be aggregated using different modes (sum, min, max, + * alwaysone), and outputs the binary representation of the TupleSketch. + * + * Keys are hashed internally based on their type and value - the same logical value in different + * types (e.g., String("123") and Int(123)) will be treated as distinct keys. However, summary + * value types must be consistent across all calls; mixing types can produce incorrect results or + * precision loss. The value type suffix in the function name (e.g., _double) ensures type safety. + * + * See [[https://datasketches.apache.org/docs/Tuple/TupleSketches.html]] for more information. + * + * @param key + * key expression against which unique counting will occur + * @param summary + * summary expression (double type) against which different mode aggregations will occur + * @param lgNomEntries + * the log-base-2 of nomEntries decides the number of buckets for the sketch + * @param mode + * the aggregation mode for numeric summaries (sum, min, max, alwaysone) + * @param mutableAggBufferOffset + * offset for mutable aggregation buffer + * @param inputAggBufferOffset + * offset for input aggregation buffer + */ +case class TupleSketchAggDouble( + key: Expression, + summary: Expression, + lgNomEntries: Expression, + mode: Expression, + override val mutableAggBufferOffset: Int, + override val inputAggBufferOffset: Int) + extends TupleSketchAggBase[java.lang.Double, DoubleSummary] + with QuaternaryLike[Expression] { + + // Constructors + def this(key: Expression, summary: Expression) = { + this( + key, + summary, + Literal(ThetaSketchUtils.DEFAULT_LG_NOM_LONGS), + Literal(TupleSummaryMode.Sum.toString), + 0, + 0) + } + + def this(key: Expression, summary: Expression, lgNomEntries: Expression) = { + this(key, summary, lgNomEntries, Literal(TupleSummaryMode.Sum.toString), 0, 0) + } + + def this(key: Expression, summary: Expression, lgNomEntries: Expression, mode: Expression) = { + this(key, summary, lgNomEntries, mode, 0, 0) + } + + // Copy constructors required by ImperativeAggregate + override def withNewMutableAggBufferOffset( + newMutableAggBufferOffset: Int): TupleSketchAggDouble = + copy(mutableAggBufferOffset = newMutableAggBufferOffset) + + override def withNewInputAggBufferOffset(newInputAggBufferOffset: Int): TupleSketchAggDouble = + copy(inputAggBufferOffset = newInputAggBufferOffset) + + override protected def withNewChildrenInternal( + newFirst: Expression, + newSecond: Expression, + newThird: Expression, + newFourth: Expression): TupleSketchAggDouble = + copy(key = newFirst, summary = newSecond, lgNomEntries = newThird, mode = newFourth) + + override def first: Expression = key + override def second: Expression = summary + override def third: Expression = lgNomEntries + override def fourth: Expression = mode + + // Override for TypedImperativeAggregate + override def prettyName: String = "tuple_sketch_agg_double" + + /** Specifies accepted summary input types (double). */ + override protected def summaryInputType: AbstractDataType = DoubleType + + /** + * Creates a DoubleSummaryFactory with the configured aggregation mode. + */ + override protected def createSummaryFactory(): SummaryFactory[DoubleSummary] = { + new DoubleSummaryFactory(modeEnum.toDoubleSummaryMode) + } + + /** + * Creates DoubleSummarySetOperations for merge operations with the configured mode. + */ + override protected def createSummarySetOperations(): SummarySetOperations[DoubleSummary] = { + new DoubleSummarySetOperations(modeEnum.toDoubleSummaryMode) + } + + /** + * Heapify a sketch from a byte array. + * + * @param buffer + * the serialized sketch byte array + * @return + * a Sketch[DoubleSummary] instance + */ + override protected def heapifySketch(buffer: Array[Byte]): Sketch[DoubleSummary] = { + TupleSketchUtils.heapifyDoubleSketch(buffer, prettyName) + } +} + +/** + * The TupleSketchAggInteger function utilizes a Datasketches TupleSketch instance to count a + * probabilistic approximation of the number of unique values in a given column with associated + * integer type summary values that can be aggregated using different modes (sum, min, max, + * alwaysone), and outputs the binary representation of the TupleSketch. + * + * Keys are hashed internally based on their type and value - the same logical value in different + * types (e.g., String("123") and Int(123)) will be treated as distinct keys. However, summary + * value types must be consistent across all calls; mixing types can produce incorrect results or + * precision loss. The value type suffix in the function name (e.g., _integer) ensures type safety. + * + * See [[https://datasketches.apache.org/docs/Tuple/TupleSketches.html]] for more information. + * + * @param key + * key expression against which unique counting will occur + * @param summary + * summary expression (integer type) against which different mode aggregations will occur + * @param lgNomEntries + * the log-base-2 of nomEntries decides the number of buckets for the sketch + * @param mode + * the aggregation mode for numeric summaries (sum, min, max, alwaysone) + * @param mutableAggBufferOffset + * offset for mutable aggregation buffer + * @param inputAggBufferOffset + * offset for input aggregation buffer + */ +case class TupleSketchAggInteger( + key: Expression, + summary: Expression, + lgNomEntries: Expression, + mode: Expression, + override val mutableAggBufferOffset: Int, + override val inputAggBufferOffset: Int) + extends TupleSketchAggBase[Integer, IntegerSummary] + with QuaternaryLike[Expression] { + + // Constructors + def this(key: Expression, summary: Expression) = { + this( + key, + summary, + Literal(ThetaSketchUtils.DEFAULT_LG_NOM_LONGS), + Literal(TupleSummaryMode.Sum.toString), + 0, + 0) + } + + def this(key: Expression, summary: Expression, lgNomEntries: Expression) = { + this(key, summary, lgNomEntries, Literal(TupleSummaryMode.Sum.toString), 0, 0) + } + + def this(key: Expression, summary: Expression, lgNomEntries: Expression, mode: Expression) = { + this(key, summary, lgNomEntries, mode, 0, 0) + } + + // Copy constructors required by ImperativeAggregate + override def withNewMutableAggBufferOffset( + newMutableAggBufferOffset: Int): TupleSketchAggInteger = + copy(mutableAggBufferOffset = newMutableAggBufferOffset) + + override def withNewInputAggBufferOffset(newInputAggBufferOffset: Int): TupleSketchAggInteger = + copy(inputAggBufferOffset = newInputAggBufferOffset) + + override protected def withNewChildrenInternal( + newFirst: Expression, + newSecond: Expression, + newThird: Expression, + newFourth: Expression): TupleSketchAggInteger = + copy(key = newFirst, summary = newSecond, lgNomEntries = newThird, mode = newFourth) + + override def first: Expression = key + override def second: Expression = summary + override def third: Expression = lgNomEntries + override def fourth: Expression = mode + + // Override for TypedImperativeAggregate + override def prettyName: String = "tuple_sketch_agg_integer" + + /** Specifies accepted summary input types (integer). */ + override protected def summaryInputType: AbstractDataType = + IntegerType + + /** + * Creates an IntegerSummaryFactory with the configured aggregation mode. + */ + override protected def createSummaryFactory(): SummaryFactory[IntegerSummary] = { + new IntegerSummaryFactory(modeEnum.toIntegerSummaryMode) + } + + /** + * Creates IntegerSummarySetOperations for merge operations with the configured mode. + */ + override protected def createSummarySetOperations(): SummarySetOperations[IntegerSummary] = { + val mode = modeEnum.toIntegerSummaryMode + new IntegerSummarySetOperations(mode, mode) + } + + /** + * Heapify a sketch from a byte array. + * + * @param buffer + * the serialized sketch byte array + * @return + * a Sketch[IntegerSummary] instance + */ + override protected def heapifySketch(buffer: Array[Byte]): Sketch[IntegerSummary] = { + TupleSketchUtils.heapifyIntegerSketch(buffer, prettyName) + } +} + +abstract class TupleSketchAggBase[U, S <: UpdatableSummary[U]] + extends TypedImperativeAggregate[TupleSketchState[S]] + with SketchSize + with SummaryAggregateMode + with ImplicitCastInputTypes { + + // Abstract methods that subclasses must implement + protected def summaryInputType: AbstractDataType + protected def createSummaryFactory(): SummaryFactory[S] + protected def createSummarySetOperations(): SummarySetOperations[S] + protected def heapifySketch(buffer: Array[Byte]): Sketch[S] + + // Abstract members that subclasses must implement + protected def key: Expression + protected def summary: Expression + + // Type and input validation overrides + protected final val keyInputTypes: AbstractDataType = + TypeCollection( + ArrayType(IntegerType), + ArrayType(LongType), + BinaryType, + DoubleType, + FloatType, + IntegerType, + LongType, + StringTypeWithCollation(supportsTrimCollation = true)) + + override def dataType: DataType = BinaryType + override def nullable: Boolean = false + + override def inputTypes: Seq[AbstractDataType] = + Seq( + keyInputTypes, + summaryInputType, + IntegerType, + StringTypeWithCollation(supportsTrimCollation = true)) + + override def checkInputDataTypes(): TypeCheckResult = { + val defaultCheck = super.checkInputDataTypes() + val lgCheck = checkLgNomEntriesParameter() + + if (defaultCheck.isFailure) { + defaultCheck + } else if (lgCheck.isFailure) { + lgCheck + } else { + checkModeParameter() + } + } + + /** + * Instantiate an UpdatableSketch instance using the lgNomEntries param and summary factory. + * + * @return + * an UpdatableSketch instance wrapped with UpdatableTupleSketchBuffer + */ + override def createAggregationBuffer(): TupleSketchState[S] = { + val factory = createSummaryFactory() + val builder = new UpdatableSketchBuilder[U, S](factory) + builder.setNominalEntries(1 << lgNomEntriesInput) + val sketch = builder.build() + UpdatableTupleSketchBuffer(sketch) + } + + /** + * Evaluate the input row and update the UpdatableSketch instance with the row's key and summary + * value. The update function only supports a subset of Spark SQL types, and an exception will + * be thrown for unsupported types. Notes: + * - Null values are ignored. + * - Empty byte arrays are ignored + * - Empty arrays of supported element types are ignored + * - Strings that are collation-equal to the empty string are ignored. + * + * @param updateBuffer + * A previously initialized UpdatableSketch instance + * @param input + * An input row + */ + override def update( + updateBuffer: TupleSketchState[S], + input: InternalRow): TupleSketchState[S] = { + val keyValue = key.eval(input) + val summaryValue = summary.eval(input) + + // Return early for null values. + if (keyValue == null || summaryValue == null) { + updateBuffer + } else { + // Type checking is already done by ImplicitCastInputTypes. + val normalizedSummary = summaryValue.asInstanceOf[U] + + // Initialized buffer should be UpdatableTupleSketchBuffer, else error out. + val sketch = updateBuffer match { + case UpdatableTupleSketchBuffer(s) => s + case _ => throw QueryExecutionErrors.tupleInvalidInputSketchBuffer(prettyName) + } + + (key.dataType, keyValue) match { + case (ArrayType(IntegerType, _), arr: ArrayData) => + sketch.update(arr.toIntArray(), normalizedSummary) + case (ArrayType(LongType, _), arr: ArrayData) => + sketch.update(arr.toLongArray(), normalizedSummary) + case (BinaryType, bytes: Array[Byte]) => + sketch.update(bytes, normalizedSummary) + case (DoubleType, v: Double) => + sketch.update(v, normalizedSummary) + case (FloatType, v: Float) => + sketch.update(v.toDouble, normalizedSummary) + case (IntegerType, v: Int) => + sketch.update(v.toLong, normalizedSummary) + case (LongType, v: Long) => + sketch.update(v, normalizedSummary) + case (st: StringType, str: UTF8String) => + val collation = CollationFactory.fetchCollation(st.collationId) + if (!collation.equalsFunction(str, UTF8String.EMPTY_UTF8)) { + sketch.update(collation.sortKeyFunction.apply(str), normalizedSummary) + } + case _ => + throw new SparkUnsupportedOperationException( + errorClass = "_LEGACY_ERROR_TEMP_3121", + messageParameters = Map("dataType" -> key.dataType.toString)) + } + + updateBuffer + } + } + + /** + * Merges an input CompactSketch into the UpdatableSketch which is acting as the aggregation + * buffer. + * + * @param updateBuffer + * The UpdatableSketch or Union instance used to store the aggregation result + * @param input + * An input UpdatableSketch, Union, or CompactSketch instance + */ + override def merge( + updateBuffer: TupleSketchState[S], + input: TupleSketchState[S]): TupleSketchState[S] = { + + def createUnionWith( + sketch1: Sketch[S], + sketch2: Sketch[S]): UnionTupleAggregationBuffer[S] = { + val summarySetOps = createSummarySetOperations() + val union = new Union[S](1 << lgNomEntriesInput, summarySetOps) + union.union(sketch1) + union.union(sketch2) + UnionTupleAggregationBuffer(union) + } + + (updateBuffer, input) match { + case (UnionTupleAggregationBuffer(union), buffer: UpdatableTupleSketchBuffer[_, S]) => + union.union(buffer.compactSketch) + UnionTupleAggregationBuffer(union) Review Comment: yeah I see what you're saying, I think changing the type of the aggregation buffer would be a bit too complex. The way the tuple and theta sketches are designed kind of forced this approach to abide by the constraints of `TypedImperativeAggregate`. I don't see a way around the unwrapping/pattern matching, but we can probably get away with the re-wrap since the changes to the aggregation buffer are mutable. I already did it for the update functions before, not sure why I forgot to extend it. This will save the unnecessary object creation. Thanks for pointing this out! -- 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]
