majian1998 commented on code in PR #10226: URL: https://github.com/apache/hudi/pull/10226#discussion_r1413424885
########## hudi-spark-datasource/hudi-spark/src/main/scala/org/apache/spark/sql/hudi/command/procedures/ShowColumnStatsOverlapProcedure.scala: ########## @@ -0,0 +1,355 @@ +/* + * 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.hudi.command.procedures + +import org.apache.avro.generic.IndexedRecord +import org.apache.hadoop.fs.{FileStatus, Path} +import org.apache.hudi.avro.model._ +import org.apache.hudi.client.common.HoodieSparkEngineContext +import org.apache.hudi.common.config.HoodieMetadataConfig +import org.apache.hudi.common.data.HoodieData +import org.apache.hudi.common.fs.FSUtils +import org.apache.hudi.common.model.{FileSlice, HoodieRecord} +import org.apache.hudi.common.table.timeline.{HoodieDefaultTimeline, HoodieInstant} +import org.apache.hudi.common.table.view.HoodieTableFileSystemView +import org.apache.hudi.common.table.{HoodieTableMetaClient, TableSchemaResolver} +import org.apache.hudi.common.util.{Option => HOption} +import org.apache.hudi.exception.HoodieException +import org.apache.hudi.metadata.HoodieTableMetadata +import org.apache.hudi.{AvroConversionUtils, ColumnStatsIndexSupport} +import org.apache.spark.internal.Logging +import org.apache.spark.sql.Row +import org.apache.spark.sql.types.{DataTypes, Metadata, StructField, StructType} + +import java.util +import java.util.function.{Function, Supplier} +import scala.collection.JavaConversions.asScalaBuffer +import scala.collection.{JavaConversions, mutable} +import scala.jdk.CollectionConverters.{asScalaBufferConverter, asScalaIteratorConverter, seqAsJavaListConverter} + +/** + * Calculate the degree of overlap between column stats. + * The overlap represents the extent to which the min-max ranges cover each other. + * By referring to the overlap, we can visually demonstrate the degree of data skipping + * for different columns under the current table's data layout. + * The calculation is performed at the partition level (assuming that data skipping is based on partition pruning). + * + * For example, consider three files: a.parquet, b.parquet, and c.parquet. + * Taking an integer-type column 'id' as an example, the range (min-max) for 'a' is 1–5, + * for 'b' is 3–7, and for 'c' is 7–8. This results in their values overlapping on the coordinate axis as follows: + * Value Range: 1 2 3 4 5 6 7 8 + * a.parquet: [-------] + * b.parquet: [--------] + * c.parquet: [-] + * Thus, there will be overlap within the ranges 3–5 and 7. + * If the filter conditions for 'id' during data skipping include these values, + * multiple files will be filtered out. For a simpler case, if it's an equality query, + * 2 files will be filtered within these ranges, and no more than one file will be filtered in other cases (possibly outside of the range). + * + * Additionally, calculating the degree of overlap based solely on the maximum values + * may not provide sufficient information. Therefore, we sample and calculate the overlap degree + * for all values involved in the min-max range. We also compute the degree of overlap + * at different percentiles and tally the count of these values.An example of a result is as follows: + * |Partition path |Field name |Average overlap |Maximum file overlap |Total file number |50% overlap |75% overlap |95% overlap |99% overlap |Total value number | + * ---------------------------------------------------------------------- + * |path |c8 |1.33 |2 |2 |1 |1 |1 |1 |3 | + + */ +class ShowColumnStatsOverlapProcedure extends BaseProcedure with ProcedureBuilder with Logging { + private val PARAMETERS = Array[ProcedureParameter]( + ProcedureParameter.required(0, "table", DataTypes.StringType), + ProcedureParameter.optional(1, "partition", DataTypes.StringType), + ProcedureParameter.optional(2, "targetColumns", DataTypes.StringType) + ) + + private val OUTPUT_TYPE = new StructType(Array[StructField]( + StructField("Partition path", DataTypes.StringType, nullable = true, Metadata.empty), + StructField("Field name", DataTypes.StringType, nullable = true, Metadata.empty), + StructField("Average overlap", DataTypes.DoubleType, nullable = true, Metadata.empty), + StructField("Maximum file overlap", DataTypes.IntegerType, nullable = true, Metadata.empty), + StructField("Total file number", DataTypes.IntegerType, nullable = true, Metadata.empty), + StructField("50% overlap", DataTypes.IntegerType, nullable = true, Metadata.empty), + StructField("75% overlap", DataTypes.IntegerType, nullable = true, Metadata.empty), + StructField("95% overlap", DataTypes.IntegerType, nullable = true, Metadata.empty), + StructField("99% overlap", DataTypes.IntegerType, nullable = true, Metadata.empty), + StructField("Total value number", DataTypes.IntegerType, nullable = true, Metadata.empty) + )) + + def parameters: Array[ProcedureParameter] = PARAMETERS + + def outputType: StructType = OUTPUT_TYPE + + override def call(args: ProcedureArgs): Seq[Row] = { + super.checkArgs(PARAMETERS, args) + + val table = getArgValueOrDefault(args, PARAMETERS(0)) + val partitions = getArgValueOrDefault(args, PARAMETERS(1)).getOrElse("").toString + val partitionsSeq = partitions.split(",").filter(_.nonEmpty).toSeq + + val targetColumnsSeq = getTargetColumnsSeq(args) + val basePath = getBasePath(table) + val metadataConfig = HoodieMetadataConfig.newBuilder().enable(true).build + val metaClient = HoodieTableMetaClient.builder.setConf(jsc.hadoopConfiguration()).setBasePath(basePath).build + val schema = getSchema(metaClient) + val columnStatsIndex = new ColumnStatsIndexSupport(spark, schema, metadataConfig, metaClient) + val fsView = buildFileSystemView(table) + val engineCtx = new HoodieSparkEngineContext(jsc) + val metaTable = HoodieTableMetadata.create(engineCtx, metadataConfig, basePath) + val allFileSlices = getAllFileSlices(partitionsSeq, metaTable, fsView) + val fileSlicesSizeByPartition = allFileSlices.groupBy(_.getPartitionPath).mapValues(_.size) + + val allFileNamesMap = getAllFileNamesMap(allFileSlices) + val colStatsRecords = getColStatsRecords(targetColumnsSeq, columnStatsIndex, schema) + + val pointList = getPointList(colStatsRecords, allFileNamesMap, schema) + + // Group points by column name + val groupedPoints = pointList.groupBy(p => (p.partitionPath, p.columnName)) + + val rows = new util.ArrayList[Row] + addStatisticsToRows(groupedPoints, fileSlicesSizeByPartition, rows) + + // The returned results are sorted by column name and average value + rows.toList.sortBy(row => (row.getString(1), row.getDouble(2))) + } + + def getTargetColumnsSeq(args: ProcedureArgs): Seq[String] = { + val targetColumns = getArgValueOrDefault(args, PARAMETERS(2)).getOrElse("").toString + if (targetColumns != "") { + targetColumns.split(",").toSeq + } else { + Seq.empty[String] + } + } + + def getSchema(metaClient: HoodieTableMetaClient): StructType = { + val schemaUtil = new TableSchemaResolver(metaClient) + AvroConversionUtils.convertAvroSchemaToStructType(schemaUtil.getTableAvroSchema) + } + + + def getAllFileSlices(partitionsSeq: Seq[String], metaTable: HoodieTableMetadata, fsView: HoodieTableFileSystemView): Set[FileSlice] = { + if (partitionsSeq.isEmpty) { + getFileSlices(metaTable.getAllPartitionPaths, fsView) + } else { + val filteredPartitions = metaTable.getAllPartitionPaths.asScala + .filter(partition => partitionsSeq.exists(prefix => partition.startsWith(prefix))) + .toList + .asJava + getFileSlices(filteredPartitions, fsView) + } + } + + def getFileSlices(partitionPaths: util.List[String], fsView: HoodieTableFileSystemView): Set[FileSlice] = { + partitionPaths + .asScala + .flatMap(path => fsView.getLatestFileSlices(path).iterator().asScala) + .toSet + } + + def getAllFileNamesMap(allFileSlices: Set[FileSlice]): Map[String, String] = { + allFileSlices.map { fileSlice => + val fileName = fileSlice.getBaseFile.get().getFileName + val partitionPath = fileSlice.getPartitionPath + fileName -> partitionPath + }.toMap + } + + def getColStatsRecords(targetColumnsSeq: Seq[String], columnStatsIndex: ColumnStatsIndexSupport, schema: StructType): HoodieData[HoodieMetadataColumnStats] = { + if (!targetColumnsSeq.isEmpty) { + columnStatsIndex.loadColumnStatsIndexRecords(targetColumnsSeq, false) + } else { + columnStatsIndex.loadColumnStatsIndexRecords( + schema.fields.filter(field => !HoodieRecord.HOODIE_META_COLUMNS.contains(field.name)).map(_.name).toSeq, + false + ) + } + } + + def getPointList(colStatsRecords: HoodieData[HoodieMetadataColumnStats], allFileNamesMap: Map[String, String], schema: StructType): List[ColumnStatsPoint] = { + colStatsRecords.collectAsList().asScala + .filter(c => allFileNamesMap.keySet.contains(c.getFileName)) + .flatMap(c => + (getColumnStatsValue(c.getMinValue), getColumnStatsValue(c.getMaxValue)) match { + case (Some(minValue), Some(maxValue)) => + Seq( + new ColumnStatsPoint(allFileNamesMap.get(c.getFileName).getOrElse(c.getColumnName), c.getColumnName, minValue, "min", schema(c.getColumnName).dataType.typeName), + new ColumnStatsPoint(allFileNamesMap.get(c.getFileName).getOrElse(c.getColumnName), c.getColumnName, maxValue, "max", schema(c.getColumnName).dataType.typeName) + ) + case _ => Seq.empty + } + ) + .toList + } + + /** + * Adds statistical information to the result rows. + * + * @param groupedPoints Data points grouped by partition path and column name + * @param fileSlicesSizeByPartition Calculated number of file slices by partition path + * @param rows List of rows storing the results + */ + def addStatisticsToRows(groupedPoints: Map[(String, String), List[ColumnStatsPoint]], + fileSlicesSizeByPartition: Map[String, Int], + rows: util.ArrayList[Row]): Unit = { + groupedPoints.map { case ((partitionPath, columnName), points) => + val sortedPoints = points.sorted + var maxCount, currentCount = 0 + val valueToCountMap: mutable.ListMap[String, Int] = mutable.ListMap.empty[String, Int] + + sortedPoints.foreach { point => + if (point.pType == "min") { + currentCount += 1 + maxCount = Math.max(maxCount, currentCount) + valueToCountMap(point.value) = currentCount + } else { + if (!valueToCountMap.contains(point.value)) { + valueToCountMap(point.value) = currentCount + } + currentCount -= 1 + } + } + + val averageCount = + if (valueToCountMap.nonEmpty) valueToCountMap.values.sum.toDouble / valueToCountMap.size + else 0 + val sortedCounts = valueToCountMap.values.toList.sorted + + rows.add(Row( + partitionPath, + columnName, + averageCount, + maxCount, + fileSlicesSizeByPartition.get(partitionPath), + calculatePercentile(sortedCounts, 50), + calculatePercentile(sortedCounts, 75), + calculatePercentile(sortedCounts, 95), + calculatePercentile(sortedCounts, 99), + sortedCounts.size + )) + } + } + + def calculatePercentile(values: List[Int], percentile: Double): Int = { + val index = (percentile / 100.0 * (values.size - 1)).toInt + values(index) + } + + def buildFileSystemView(table: Option[Any]): HoodieTableFileSystemView = { + val basePath = getBasePath(table) + val metaClient = HoodieTableMetaClient.builder.setConf(jsc.hadoopConfiguration()).setBasePath(basePath).build + val fs = metaClient.getFs + val globPath = s"$basePath/*/*/*" + val statuses = FSUtils.getGlobStatusExcludingMetaFolder(fs, new Path(globPath)) + + val timeline = metaClient.getActiveTimeline.getCommitsTimeline.filterCompletedInstants() + + val maxInstant = metaClient.createNewInstantTime() + val instants = timeline.getInstants.iterator().asScala.filter(_.getTimestamp < maxInstant) + + val details = new Function[HoodieInstant, org.apache.hudi.common.util.Option[Array[Byte]]] + with java.io.Serializable { + override def apply(instant: HoodieInstant): HOption[Array[Byte]] = { + metaClient.getActiveTimeline.getInstantDetails(instant) + } + } + + val filteredTimeline = new HoodieDefaultTimeline( + new java.util.ArrayList[HoodieInstant](JavaConversions.asJavaCollection(instants.toList)).stream(), details) + + new HoodieTableFileSystemView(metaClient, filteredTimeline, statuses.toArray(new Array[FileStatus](statuses.size))) + } + + def getColumnStatsValue(stats_value: Any): Option[String] = { + if (stats_value == null) { + logInfo("invalid value " + stats_value) + None Review Comment: Done, this has been abstracted into HoodieTableMetadataUtil. -- 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: commits-unsubscr...@hudi.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
