garyli1019 commented on a change in pull request #1722: URL: https://github.com/apache/hudi/pull/1722#discussion_r439769643
########## File path: hudi-spark/src/main/scala/org/apache/hudi/SnapshotRelation.scala ########## @@ -0,0 +1,132 @@ +/* + * 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.avro.HoodieAvroUtils +import org.apache.hudi.common.table.{HoodieTableMetaClient, TableSchemaResolver} +import org.apache.hudi.config.HoodieWriteConfig +import org.apache.hudi.hadoop.{HoodieParquetInputFormat, HoodieROTablePathFilter} +import org.apache.hudi.hadoop.utils.HoodieRealtimeInputFormatUtils +import org.apache.hudi.exception.HoodieException +import org.apache.hudi.table.HoodieTable + +import org.apache.hadoop.mapred.JobConf +import org.apache.log4j.LogManager +import org.apache.spark.rdd.RDD +import org.apache.spark.sql.{Row, SQLContext} +import org.apache.spark.sql.sources.{BaseRelation, TableScan} +import org.apache.spark.sql.types.StructType + +import java.util +import scala.collection.JavaConverters._ + +/** + * This is the Spark DataSourceV1 relation to read Hudi MOR table. + * @param sqlContext + * @param basePath + * @param optParams + * @param userSchema + */ +class SnapshotRelation(val sqlContext: SQLContext, + val basePath: String, + val optParams: Map[String, String], + val userSchema: StructType) extends BaseRelation with TableScan { + + private val log = LogManager.getLogger(classOf[SnapshotRelation]) + private val conf = sqlContext.sparkContext.hadoopConfiguration + + // Set config for listStatus() in HoodieParquetInputFormat + conf.setClass( + "mapreduce.input.pathFilter.class", + classOf[HoodieROTablePathFilter], + classOf[org.apache.hadoop.fs.PathFilter]) + conf.setStrings("mapreduce.input.fileinputformat.inputdir", basePath) + conf.setStrings("mapreduce.input.fileinputformat.input.dir.recursive", "true") + + private val HoodieInputFormat = new HoodieParquetInputFormat + HoodieInputFormat.setConf(conf) + private val fileStatus = HoodieInputFormat.listStatus(new JobConf(conf)) + private val fileGroup = HoodieRealtimeInputFormatUtils.groupLogsByBaseFile(conf, util.Arrays.stream(fileStatus)).asScala + + // Split the file group to: parquet file without a matching log file, parquet file need to merge with log files + private val parquetWithoutLogPaths: List[String] = fileGroup.filter(p => p._2.size() == 0).keys.toList + private val fileWithLogMap: Map[String, String] = fileGroup.filter(p => p._2.size() > 0).map{ case(k, v) => (k, v.asScala.toList.mkString(","))}.toMap + + if (log.isDebugEnabled) { + log.debug("All parquet files" + fileStatus.map(s => s.getPath.toString).mkString(",")) + log.debug("ParquetWithoutLogPaths" + parquetWithoutLogPaths.mkString(",")) + log.debug("ParquetWithLogPaths" + fileWithLogMap.map(m => s"${m._1}:${m._2}").mkString(",")) + } + + // Add log file map to options + private val finalOps = optParams ++ fileWithLogMap Review comment: I agree here. Other possible implementations could be: - Add it to Hadoop config, but it could cause the same issue as using the option hashmap. ~- Concatenate the logPath to the parquetPath, then split path in the executor before doing the file split.~(EDIT: Doesn't work, in the `FileFormat`, the file is already a `PartitionedFile` type) - Search log based on parquetPath on each executor. This could put pressure on the name node as well. (EDIT: This probably doesn't work as well. Getting the meta client and file system view is very expensive.) ~I personally prefer the second way if we can efficiently handle all the file listing on the driver side. I will see if it's possible to implement it.~ ---------------------------------------------------------------- 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]
