pengzhiwei2018 commented on a change in pull request #2651: URL: https://github.com/apache/hudi/pull/2651#discussion_r599296787
########## File path: hudi-spark-datasource/hudi-spark/src/main/scala/org/apache/hudi/HoodieFileIndex.scala ########## @@ -0,0 +1,317 @@ +/* + * 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 java.util.Properties + +import scala.collection.JavaConverters._ + +import org.apache.hadoop.fs.{FileStatus, Path} +import org.apache.hudi.client.common.HoodieSparkEngineContext +import org.apache.hudi.common.config.{HoodieMetadataConfig, SerializableConfiguration} +import org.apache.hudi.common.engine.HoodieLocalEngineContext +import org.apache.hudi.common.fs.FSUtils +import org.apache.hudi.common.model.HoodieBaseFile +import org.apache.hudi.common.table.{HoodieTableMetaClient, TableSchemaResolver} +import org.apache.hudi.common.table.view.HoodieTableFileSystemView +import org.apache.hudi.config.HoodieWriteConfig +import org.apache.spark.api.java.JavaSparkContext +import org.apache.spark.internal.Logging +import org.apache.spark.sql.catalyst.{InternalRow, expressions} +import org.apache.spark.sql.SparkSession +import org.apache.spark.sql.avro.SchemaConverters +import org.apache.spark.sql.catalyst.expressions.{AttributeReference, BoundReference, Expression, InterpretedPredicate} +import org.apache.spark.sql.catalyst.util.{CaseInsensitiveMap, DateTimeUtils} +import org.apache.spark.sql.execution.datasources.{FileIndex, PartitionDirectory, PartitionUtils} +import org.apache.spark.sql.internal.SQLConf +import org.apache.spark.sql.types.StructType +import org.apache.spark.unsafe.types.UTF8String + +/** + * A File Index which support partition prune for hoodie snapshot and read-optimized + * query. + * Main steps to get the file list for query: + * 1、Load all files and partition values from the table path. + * 2、Do the partition prune by the partition filter condition. + * + * There are 3 cases for this: + * 1、If the partition columns size is equal to the actually partition path level, we + * read it as partitioned table.(e.g partition column is "dt", the partition path is "2021-03-10") + * + * 2、If the partition columns size is not equal to the partition path level, but the partition + * column size is "1" (e.g. partition column is "dt", but the partition path is "2021/03/10" + * who'es directory level is 3).We can still read it as a partitioned table. We will mapping the + * partition path (e.g. 2021/03/10) to the only partition column (e.g. "dt"). + * + * 3、Else the the partition columns size is not equal to the partition directory level and the + * size is great than "1" (e.g. partition column is "dt,hh", the partition path is "2021/03/10/12") + * , we read it as a None Partitioned table because we cannot know how to mapping the partition + * path with the partition columns in this case. + */ +case class HoodieFileIndex( + spark: SparkSession, + basePath: String, + schemaSpec: Option[StructType], + options: Map[String, String]) + extends FileIndex with Logging { + + @transient private val hadoopConf = spark.sessionState.newHadoopConf() + private lazy val metaClient = HoodieTableMetaClient + .builder().setConf(hadoopConf).setBasePath(basePath).build() + + @transient private val queryPath = new Path(options.getOrElse("path", "'path' option required")) + /** + * Get the schema of the table. + */ + lazy val schema: StructType = schemaSpec.getOrElse({ + val schemaUtil = new TableSchemaResolver(metaClient) + SchemaConverters.toSqlType(schemaUtil.getTableAvroSchema) + .dataType.asInstanceOf[StructType] + }) + + /** + * Get the partition schema from the hoodie.properties. + */ + private lazy val _partitionSchemaFromProperties: StructType = { + val tableConfig = metaClient.getTableConfig + val partitionColumns = tableConfig.getPartitionColumns + val nameFieldMap = schema.fields.map(filed => filed.name -> filed).toMap + + if (partitionColumns.isPresent) { + val partitionFields = partitionColumns.get().map(column => + nameFieldMap.getOrElse(column, throw new IllegalArgumentException(s"Cannot find column " + + s"$column in the schema[${schema.fields.mkString(",")}]"))) + new StructType(partitionFields) + } else { // If the partition columns have not stored in hoodie.properites(the table that was + // created earlier), we trait it as a none-partitioned table. + new StructType() + } + } + + @transient @volatile private var fileSystemView: HoodieTableFileSystemView = _ + @transient @volatile private var cachedAllInputFiles: Array[HoodieBaseFile] = _ + @transient @volatile private var cachedFileSize: Long = 0L + @transient @volatile private var cachedAllPartitionPaths: Seq[PartitionPath] = _ + + @volatile private var queryAsNonePartitionedTable: Boolean = _ + + refresh() + + override def rootPaths: Seq[Path] = queryPath :: Nil + + override def listFiles(partitionFilters: Seq[Expression], + dataFilters: Seq[Expression]): Seq[PartitionDirectory] = { + if (queryAsNonePartitionedTable) { // Read as None Partitioned table. + Seq(PartitionDirectory(InternalRow.empty, allFiles)) + } else { + // Prune the partition path by the partition filters + val prunedPartitions = prunePartition(cachedAllPartitionPaths, partitionFilters) + prunedPartitions.map { partition => + val fileStatues = fileSystemView.getLatestBaseFiles(partition.partitionPath).iterator() + .asScala.toSeq + .map(_.getFileStatus) + PartitionDirectory(partition.values, fileStatues) + } + } + } + + override def inputFiles: Array[String] = { + cachedAllInputFiles.map(_.getFileStatus.getPath.toString) + } + + override def refresh(): Unit = { Review comment: Hi @umehrot2 , AFAIK, The `refresh()` will be called the follow case: 1、The Refresh Table Command. 2、After write finish, e.g `InsertIntoHadoopFsRelationCommand` will call `fileIndex.foreach(_.refresh())`. > The InMemoryFileIndex mantains refresh() and refresh0(). The refresh0() is called from within the index class at the time of creation. While refresh() is called from other places too. Do you know why we don't need to follow similar model ? The `InMemoryFileIndex` has some logical about the `fileStatusCache` in the `refresh`. So it has to distinguish between `refresh` and `refresh0`. After we introduce the `fileStatusCache` for the across queries, we should do this too. -- 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: [email protected]
