vinothchandar commented on a change in pull request #1938: URL: https://github.com/apache/hudi/pull/1938#discussion_r480292973
########## File path: hudi-spark/src/main/scala/org/apache/hudi/MergeOnReadIncrementalRelation.scala ########## @@ -0,0 +1,209 @@ +/* + * 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.hudi + +import org.apache.hudi.common.model.{HoodieCommitMetadata, HoodieRecord, HoodieTableType} +import org.apache.hudi.common.table.{HoodieTableMetaClient, TableSchemaResolver} +import org.apache.hudi.exception.HoodieException +import org.apache.hadoop.fs.{FileStatus, FileSystem, GlobPattern, Path} +import org.apache.hadoop.mapred.JobConf +import org.apache.hudi.common.fs.FSUtils +import org.apache.hudi.common.table.view.HoodieTableFileSystemView +import org.apache.hudi.hadoop.utils.HoodieRealtimeRecordReaderUtils.getMaxCompactionMemoryInBytes +import org.apache.log4j.LogManager +import org.apache.spark.deploy.SparkHadoopUtil +import org.apache.spark.rdd.RDD +import org.apache.spark.sql.catalyst.InternalRow +import org.apache.spark.sql.execution.datasources.PartitionedFile +import org.apache.spark.sql.execution.datasources.parquet.ParquetFileFormat +import org.apache.spark.sql.sources.{BaseRelation, Filter, GreaterThanOrEqual, IsNotNull, LessThanOrEqual, PrunedFilteredScan, TableScan} +import org.apache.spark.sql.types.StructType +import org.apache.spark.sql.{Row, SQLContext} + +import scala.collection.JavaConversions._ +import scala.collection.mutable.ListBuffer + +/** + * Experimental. + * Relation, that implements the Hoodie incremental view for Merge On Read table. + * + */ +class MergeOnReadIncrementalRelation(val sqlContext: SQLContext, + val optParams: Map[String, String], + val userSchema: StructType, + val metaClient: HoodieTableMetaClient) + extends BaseRelation with PrunedFilteredScan { + + private val log = LogManager.getLogger(classOf[MergeOnReadIncrementalRelation]) + private val conf = sqlContext.sparkContext.hadoopConfiguration + private val jobConf = new JobConf(conf) + private val fs = FSUtils.getFs(metaClient.getBasePath, conf) + private val commitTimeline = metaClient.getCommitsAndCompactionTimeline.filterCompletedInstants() + if (commitTimeline.empty()) { + throw new HoodieException("No instants to incrementally pull") + } + if (!optParams.contains(DataSourceReadOptions.BEGIN_INSTANTTIME_OPT_KEY)) { + throw new HoodieException(s"Specify the begin instant time to pull from using " + + s"option ${DataSourceReadOptions.BEGIN_INSTANTTIME_OPT_KEY}") + } + + private val lastInstant = commitTimeline.lastInstant().get() + private val mergeType = optParams.getOrElse( + DataSourceReadOptions.REALTIME_MERGE_OPT_KEY, + DataSourceReadOptions.DEFAULT_REALTIME_MERGE_OPT_VAL) + + private val commitsTimelineToReturn = commitTimeline.findInstantsInRange( + optParams(DataSourceReadOptions.BEGIN_INSTANTTIME_OPT_KEY), + optParams.getOrElse(DataSourceReadOptions.END_INSTANTTIME_OPT_KEY, lastInstant.getTimestamp)) + log.debug(s"${commitsTimelineToReturn.getInstants.iterator().toList.map(f => f.toString).mkString(",")}") + private val commitsToReturn = commitsTimelineToReturn.getInstants.iterator().toList + private val schemaUtil = new TableSchemaResolver(metaClient) + private val tableAvroSchema = schemaUtil.getTableAvroSchema + private val tableStructSchema = AvroConversionUtils.convertAvroSchemaToStructType(tableAvroSchema) + private val maxCompactionMemoryInBytes = getMaxCompactionMemoryInBytes(jobConf) + private val fileIndex = buildFileIndex() + + override def schema: StructType = tableStructSchema + + override def needConversion: Boolean = false + + override def unhandledFilters(filters: Array[Filter]): Array[Filter] = { + val isNotNullFilter = IsNotNull(HoodieRecord.COMMIT_TIME_METADATA_FIELD) + val largerThanFilter = GreaterThanOrEqual(HoodieRecord.COMMIT_TIME_METADATA_FIELD, commitsToReturn.head.getTimestamp) + val lessThanFilter = LessThanOrEqual(HoodieRecord.COMMIT_TIME_METADATA_FIELD, commitsToReturn.last.getTimestamp) + filters :+isNotNullFilter :+ largerThanFilter :+ lessThanFilter + } + + override def buildScan(requiredColumns: Array[String], filters: Array[Filter]): RDD[Row] = { + log.debug(s"buildScan requiredColumns = ${requiredColumns.mkString(",")}") + log.debug(s"buildScan filters = ${filters.mkString(",")}") + // config to ensure the push down filter for parquet will be applied. + sqlContext.sparkSession.sessionState.conf.setConfString("spark.sql.parquet.recordLevelFilter.enabled", "true") + sqlContext.sparkSession.sessionState.conf.setConfString("spark.sql.parquet.enableVectorizedReader", "false") + val pushDownFilter = { + val isNotNullFilter = IsNotNull(HoodieRecord.COMMIT_TIME_METADATA_FIELD) + val largerThanFilter = GreaterThanOrEqual(HoodieRecord.COMMIT_TIME_METADATA_FIELD, commitsToReturn.head.getTimestamp) + val lessThanFilter = LessThanOrEqual(HoodieRecord.COMMIT_TIME_METADATA_FIELD, commitsToReturn.last.getTimestamp) + filters :+isNotNullFilter :+ largerThanFilter :+ lessThanFilter + } + var requiredStructSchema = StructType(Seq()) + requiredColumns.foreach(col => { + val field = tableStructSchema.find(_.name == col) + if (field.isDefined) { + requiredStructSchema = requiredStructSchema.add(field.get) + } + }) + val requiredAvroSchema = AvroConversionUtils + .convertStructTypeToAvroSchema(requiredStructSchema, tableAvroSchema.getName, tableAvroSchema.getNamespace) + val hoodieTableState = HoodieMergeOnReadTableState( + tableStructSchema, + requiredStructSchema, + tableAvroSchema.toString, + requiredAvroSchema.toString, + fileIndex + ) + val fullSchemaParquetReader = new ParquetFileFormat().buildReaderWithPartitionValues( + sparkSession = sqlContext.sparkSession, + dataSchema = tableStructSchema, + partitionSchema = StructType(Nil), + requiredSchema = tableStructSchema, + filters = pushDownFilter, + options = optParams, + hadoopConf = sqlContext.sparkSession.sessionState.newHadoopConf() + ) + val requiredSchemaParquetReader = new ParquetFileFormat().buildReaderWithPartitionValues( + sparkSession = sqlContext.sparkSession, + dataSchema = tableStructSchema, + partitionSchema = StructType(Nil), + requiredSchema = requiredStructSchema, + filters = pushDownFilter, + options = optParams, + hadoopConf = sqlContext.sparkSession.sessionState.newHadoopConf() + ) + + // Follow the implementation of Spark internal HadoopRDD to handle the broadcast configuration. + FileSystem.getLocal(jobConf) + SparkHadoopUtil.get.addCredentials(jobConf) + val rdd = new HoodieMergeOnReadRDD( + sqlContext.sparkContext, + jobConf, + fullSchemaParquetReader, + requiredSchemaParquetReader, + hoodieTableState + ) + rdd.asInstanceOf[RDD[Row]] + } + + def buildFileIndex(): List[HoodieMergeOnReadFileSplit] = { + val affectedFileStatus = new ListBuffer[FileStatus] + for (commit <- commitsToReturn) { + val metadata: HoodieCommitMetadata = HoodieCommitMetadata.fromBytes(commitsTimelineToReturn.getInstantDetails(commit) + .get, classOf[HoodieCommitMetadata]) + val idWithPath = metadata.getFileIdAndFullPaths(metaClient.getBasePath).toMap + idWithPath.foreach(p => { + val file = fs.getFileStatus(new Path(p._2)) + affectedFileStatus += file + }) + } + val fsView = new HoodieTableFileSystemView(metaClient, + commitsTimelineToReturn, affectedFileStatus.toArray) + val fileGroup = fsView.fetchAllStoredFileGroups().iterator().toList + val latestCommit = fsView.getLastInstant.get().getTimestamp + if (log.isDebugEnabled) { + fileGroup.foreach(f => log.debug(s"current file group id: " + + s"${f.getFileGroupId} and file slices ${f.getLatestFileSlice.get().toString}")) + } + val pathGlobPattern = optParams.getOrElse( + DataSourceReadOptions.INCR_PATH_GLOB_OPT_KEY, + DataSourceReadOptions.DEFAULT_INCR_PATH_GLOB_OPT_VAL) + val filteredFileGroup = if(!pathGlobPattern + .equals(DataSourceReadOptions.DEFAULT_INCR_PATH_GLOB_OPT_VAL)) { + val globMatcher = new GlobPattern("*" + pathGlobPattern) + fileGroup.filter(f => { + if (f.getLatestFileSlice.get().getBaseFile.isPresent) { + globMatcher.matches(f.getLatestFileSlice.get().getBaseFile.get.getPath) + } else { + globMatcher.matches(f.getLatestFileSlice.get().getLatestLogFile.get().getPath.toString) + } + }) + } else { + fileGroup + } + + filteredFileGroup.map(f => { + val baseFile = f.getLatestDataFile Review comment: IIUC, your approach here is to find all the file groups impacted in commit range, then pull affected records from the latest file slice? the mergeOnReadRDD will handle the merging of such an file slice, with the commit filters applied. let me know @garyli1019 if I am understanding this correctly. it will help me review the code. in the meantime, can we add a separate flag/option to turn on `new.incremental.relation=true` or sth, to control this and push the changes ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
