cboumalh commented on code in PR #52883: URL: https://github.com/apache/spark/pull/52883#discussion_r2677079413
########## sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/aggregate/tuplesketchesAggregates.scala: ########## @@ -0,0 +1,1487 @@ +/* + * 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.{Intersection, Sketch, Summary, SummaryFactory, SummarySetOperations, Union, UpdatableSketch, 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.analysis.TypeCheckResult.DataTypeMismatch +import org.apache.spark.sql.catalyst.expressions.{Expression, ExpressionDescription, ImplicitCastInputTypes, Literal} +import org.apache.spark.sql.catalyst.expressions.aggregate.TypedImperativeAggregate +import org.apache.spark.sql.catalyst.plans.logical.{FunctionSignature, InputParameter} +import org.apache.spark.sql.catalyst.trees.{BinaryLike, QuaternaryLike, TernaryLike} +import org.apache.spark.sql.catalyst.util.{ArrayData, CollationFactory, ThetaSketchUtils} +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 + +sealed trait TupleSketchState[S <: Summary] { + def serialize(): Array[Byte] + def eval(): Array[Byte] +} +case class UpdatableTupleSketchBuffer[U, S <: UpdatableSummary[U]](sketch: UpdatableSketch[U, S]) + extends TupleSketchState[S] { + override def serialize(): Array[Byte] = sketch.compact.toByteArray + override def eval(): Array[Byte] = sketch.compact.toByteArray + + /** Returns compact form of the sketch, needed for merge operations that require Sketch type. */ + def compactSketch: Sketch[S] = sketch.compact +} +case class UnionTupleAggregationBuffer[S <: Summary](union: Union[S]) + extends TupleSketchState[S] { + override def serialize(): Array[Byte] = union.getResult.toByteArray + override def eval(): Array[Byte] = union.getResult.toByteArray +} +case class IntersectionTupleAggregationBuffer[S <: Summary](intersection: Intersection[S]) + extends TupleSketchState[S] { + override def serialize(): Array[Byte] = intersection.getResult.toByteArray + override def eval(): Array[Byte] = intersection.getResult.toByteArray +} +case class FinalizedTupleSketch[S <: Summary](sketch: Sketch[S]) extends TupleSketchState[S] { + override def serialize(): Array[Byte] = sketch.toByteArray + override def eval(): Array[Byte] = sketch.toByteArray +} + +/** + * 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] + with SummaryAggregateMode { + + // Constructors + def this(key: Expression, summary: Expression) = { + this( + key, + summary, + Literal(ThetaSketchUtils.DEFAULT_LG_NOM_LONGS), + Literal(ThetaSketchUtils.MODE_SUM), + 0, + 0) + } + + def this(key: Expression, summary: Expression, lgNomEntries: Expression) = { + this(key, summary, lgNomEntries, Literal(ThetaSketchUtils.MODE_SUM), 0, 0) + } + + def this(key: Expression, summary: Expression, lgNomEntries: Expression, mode: Expression) = { + this(key, summary, lgNomEntries, mode, 0, 0) + } + + /** + * Override inputTypes to specify key, summary (double/float), lgNomEntries (int), and mode + * (string) parameters. + */ + override def inputTypes: Seq[AbstractDataType] = + Seq( + keyInputTypes, + summaryInputType, + IntegerType, + StringTypeWithCollation(supportsTrimCollation = true)) + + /** + * Override checkInputDataTypes to validate base inputs (key, summary, lgNomEntries) and mode + * parameter. + */ + override def checkInputDataTypes(): TypeCheckResult = { + val defaultCheck = checkBaseInputDataTypes() + if (defaultCheck.isFailure) { + defaultCheck + } else { + checkModeParameter() + } + } + + // 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 + + // Overrides 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] = { + val mode = ThetaSketchUtils.getDoubleSummaryMode(modeInput) + new DoubleSummaryFactory(mode) + } + + /** + * Creates DoubleSummarySetOperations for merge operations with the configured mode. + */ + override protected def createSummarySetOperations(): SummarySetOperations[DoubleSummary] = { + val mode = ThetaSketchUtils.getDoubleSummaryMode(modeInput) + new DoubleSummarySetOperations(mode) + } + + + /** + * Heapify a CompactSketch from the sketch byte array. + * + * @param buffer + * A serialized sketch byte array + * @return + * A CompactSketch instance wrapped with FinalizedTupleSketch + */ + override def deserialize(buffer: Array[Byte]): TupleSketchState[DoubleSummary] = { + if (buffer.nonEmpty) { + FinalizedTupleSketch( + ThetaSketchUtils.heapifyDoubleTupleSketch(buffer, prettyName)) + } else { + createAggregationBuffer() + } + } +} + +/** + * 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] + with SummaryAggregateMode { + + // Constructors + def this(key: Expression, summary: Expression) = { + this( + key, + summary, + Literal(ThetaSketchUtils.DEFAULT_LG_NOM_LONGS), + Literal(ThetaSketchUtils.MODE_SUM), + 0, + 0) + } + + def this(key: Expression, summary: Expression, lgNomEntries: Expression) = { + this(key, summary, lgNomEntries, Literal(ThetaSketchUtils.MODE_SUM), 0, 0) + } + + def this(key: Expression, summary: Expression, lgNomEntries: Expression, mode: Expression) = { + this(key, summary, lgNomEntries, mode, 0, 0) + } + + /** + * Override inputTypes to specify key, summary (integer), lgNomEntries (int), and mode (string) + * parameters. + */ + override def inputTypes: Seq[AbstractDataType] = + Seq( + keyInputTypes, + summaryInputType, + IntegerType, + StringTypeWithCollation(supportsTrimCollation = true)) + + /** + * Override checkInputDataTypes to validate base inputs (key, summary, lgNomEntries) and mode + * parameter. + */ + override def checkInputDataTypes(): TypeCheckResult = { + val defaultCheck = checkBaseInputDataTypes() + if (defaultCheck.isFailure) { + defaultCheck + } else { + checkModeParameter() + } + } + + // 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 + + // Overrides 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] = { + val mode = ThetaSketchUtils.getIntegerSummaryMode(modeInput) + new IntegerSummaryFactory(mode) + } + + /** + * Creates IntegerSummarySetOperations for merge operations with the configured mode. + */ + override protected def createSummarySetOperations(): SummarySetOperations[IntegerSummary] = { + val mode = ThetaSketchUtils.getIntegerSummaryMode(modeInput) + new IntegerSummarySetOperations(mode, mode) + } + + + /** + * Heapify a CompactSketch from the sketch byte array. + * + * @param buffer + * A serialized sketch byte array + * @return + * A CompactSketch instance wrapped with FinalizedTupleSketch + */ + override def deserialize(buffer: Array[Byte]): TupleSketchState[IntegerSummary] = { + if (buffer.nonEmpty) { + FinalizedTupleSketch( + ThetaSketchUtils.heapifyIntegerTupleSketch(buffer, prettyName)) + } else { + createAggregationBuffer() + } + } +} + +abstract class TupleSketchAggBase[U, S <: UpdatableSummary[U]] + extends TypedImperativeAggregate[TupleSketchState[S]] + with SketchSize + with ImplicitCastInputTypes { + + // Abstract methods that subclasses must implement + protected def summaryInputType: AbstractDataType + protected def createSummaryFactory(): SummaryFactory[S] + protected def createSummarySetOperations(): SummarySetOperations[S] + + // Abstract members that subclasses must implement + protected def key: Expression + protected def summary: Expression + + 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 + + protected def checkBaseInputDataTypes(): TypeCheckResult = { + val defaultCheck = super.checkInputDataTypes() + if (defaultCheck.isFailure) { + defaultCheck + } else { + checkLgNomEntriesParameter() + } + } + + /** + * 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 match { Review Comment: Done! I added the object TupleSummaryMode in the ThetaSketchUtils 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. 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]
