spark 分析任务血缘关系

接到新的需求,在spark中增加埋点，解析任务的血缘关系，包括sql和代码方式，不包括中间临时视图(createOrReplaceTempView(XXX表))。

有位同学已经https://www.cnblogs.com/wuxilc/p/9326130.html 做了hive解析相关的，但是spark部分因为hive parseDriver解析不了。

还是在spark中搞搞吧。

spark  BaseSessionStateBuilder 类

  /**
   * Build the [[SessionState]].
   */
  def build(): SessionState = {
    new SessionState(
      session.sharedState,
      conf,
      experimentalMethods,
      functionRegistry,
      udfRegistration,
      () => catalog,
      sqlParser,
      () => analyzer,
      () => optimizer,
      planner,
      streamingQueryManager,
      listenerManager,
      () => resourceLoader,
      createQueryExecution,
      createClone)
  }
}

optimizer 是spark语法优化器，HiveSessionStateBilder 继承了BaseSessionStateBuilder类

 

 Spark 所有优化器都继承抽象类Optimizer

/**
 * Abstract class all optimizers should inherit of, contains the standard batches (extending
 * Optimizers can override this.
 */
abstract class Optimizer(sessionCatalog: SessionCatalog)
  extends RuleExecutor[LogicalPlan] {


在优化器 rule添加新的匹配规则

HiveSessionStateBuilder 类中修改

override lazy val optimizer: Optimizer = new SparkOptimizer(catalog, conf, experimentalMethods) {
  override def batches: Seq[Batch] = super.batches :+
    Batch("Determine stats of partitionedTable", Once,
      DeterminePartitionedTableStats(sparkSession)) :+
    Batch("Collect read and write tables", Once, DependencyCollect(sparkSession))
}

添加如下代码 匹配inset、create等语句解析出输入表输出表。

case class DependencyCollect(sparkSession: SparkSession) extends Rule[LogicalPlan] {
  def apply(plan: LogicalPlan): LogicalPlan = {
    if (sparkSession.sparkContext.conf.getBoolean("spark.collectDependencies", true)) {
      val readTables = mutable.HashSet[String]()
      val writeTables = mutable.HashSet[String]()
      plan transformDown {
        case a@InsertIntoHiveTable(table: CatalogTable,_,_,_,_,_) =>
          writeTables += s"${fillBlankDatabase(table)}.${table.identifier.table}"
          a
        case i@InsertIntoTable(table: HiveTableRelation, _, _, _, _) =>
          writeTables += s"${table.tableMeta.database}.${table.tableMeta.identifier.table}"
          i
        case c@CreateTable(table: CatalogTable, _, _) =>
          writeTables += s"${fillBlankDatabase(table)}.${table.identifier.table}"
          c
        case d@CreateTableCommand(table: CatalogTable, _) =>
          writeTables += s"${fillBlankDatabase(table)}.${table.identifier.table}"
          d
        case p@PhysicalOperation(_, _, table: HiveTableRelation) =>
          readTables += s"${table.tableMeta.database}.${table.tableMeta.identifier.table}"
          p
      }
      if (readTables.size > 0 || writeTables.size > 0) {
        logInfo(String.format("src table -> %s target table -> %s", readTables.mkString(","), writeTables.mkString(",")))
        AsyncExecution.AsycnHandle(new CallChain.Event(s"${readTables.mkString(",")}#${writeTables.mkString(",")}", AsyncExecution.getSparkAppName(sparkSession.sparkContext.conf), "bloodlineage"))
        sparkSession.sparkContext.listenerBus.post(DependencyEvent(readTables, writeTables))
      }
    }
    plan
  }

  private def fillBlankDatabase(table: CatalogTable): String = {
    var database = ""
    if (table.database.isEmpty) {
      database = sparkSession.sessionState.catalog.getCurrentDatabase
    } else {
      database = table.database
    }
    database
  }