Spark 之 HiveStrategies

HiveTableRelation 相关代码

HiveStrategies.scala

当 relation.tableMeta.stats.isEmpty 是, 即调用 hiveTableWithStats

复制代码
class DetermineTableStats(session: SparkSession) extends Rule[LogicalPlan] {
  private def hiveTableWithStats(relation: HiveTableRelation): HiveTableRelation = {
    val table = relation.tableMeta
    val partitionCols = relation.partitionCols
    // For partitioned tables, the partition directory may be outside of the table directory.
    // Which is expensive to get table size. Please see how we implemented it in the AnalyzeTable.
    val sizeInBytes = if (conf.fallBackToHdfsForStatsEnabled && partitionCols.isEmpty) {
      try {
        val hadoopConf = session.sessionState.newHadoopConf()
        val tablePath = new Path(table.location)
        val fs: FileSystem = tablePath.getFileSystem(hadoopConf)
        fs.getContentSummary(tablePath).getLength
      } catch {
        case e: IOException =>
          logWarning("Failed to get table size from HDFS.", e)
          conf.defaultSizeInBytes
      }
    } else {
      conf.defaultSizeInBytes
    }

    val stats = Some(Statistics(sizeInBytes = BigInt(sizeInBytes)))
    relation.copy(tableStats = stats)
  }

  override def apply(plan: LogicalPlan): LogicalPlan = plan resolveOperators {
    case relation: HiveTableRelation
      if DDLUtils.isHiveTable(relation.tableMeta) && relation.tableMeta.stats.isEmpty =>
      hiveTableWithStats(relation)

    // handles InsertIntoStatement specially as the table in InsertIntoStatement is not added in its
    // children, hence not matched directly by previous HiveTableRelation case.
    case i @ InsertIntoStatement(relation: HiveTableRelation, _, _, _, _, _)
      if DDLUtils.isHiveTable(relation.tableMeta) && relation.tableMeta.stats.isEmpty =>
      i.copy(table = hiveTableWithStats(relation))
  }
}
  • HiveTableRelation

    /**

    • A LogicalPlan that represents a hive table.
    • TODO: remove this after we completely make hive as a data source.
      */
      case class HiveTableRelation(
      tableMeta: CatalogTable,
      dataCols: Seq[AttributeReference],
      partitionCols: Seq[AttributeReference],
      tableStats: Option[Statistics] = None,
      @transient prunedPartitions: Option[Seq[CatalogTablePartition]] = None)
相关推荐
武子康15 小时前
大数据-98 Spark 从 DStream 到 Structured Streaming:Spark 实时计算的演进
大数据·后端·spark
阿里云大数据AI技术15 小时前
2025云栖大会·大数据AI参会攻略请查收!
大数据·人工智能
代码匠心18 小时前
从零开始学Flink:数据源
java·大数据·后端·flink
Lx35220 小时前
复杂MapReduce作业设计:多阶段处理的最佳实践
大数据·hadoop
武子康1 天前
大数据-100 Spark DStream 转换操作全面总结:map、reduceByKey 到 transform 的实战案例
大数据·后端·spark
expect7g1 天前
Flink KeySelector
大数据·后端·flink
阿里云大数据AI技术2 天前
StarRocks 助力数禾科技构建实时数仓:从数据孤岛到智能决策
大数据
Lx3522 天前
Hadoop数据处理优化:减少Shuffle阶段的性能损耗
大数据·hadoop
努力的小郑2 天前
从一次分表实践谈起:我们真的需要复杂的分布式ID吗?
分布式·后端·面试
武子康2 天前
大数据-99 Spark Streaming 数据源全面总结:原理、应用 文件流、Socket、RDD队列流
大数据·后端·spark