Question

我正在尝试在Spark中创建一批Dataset行。为了保持发送到服务的记录数量，我想批量项目，以便我可以保持数据发送的速率。对于，

case class Person(name:String, address: String)
case class PersonBatch(personBatch: List[Person])

对于给定的Dataset[Person]，我想创建Dataset[PersonBatch]

例如，如果输入Dataset[Person]有100条记录，则输出Dataset应该类似于Dataset[PersonBatch]，其中每个PersonBatch都应该是n条记录的列表（人）

我已经尝试了这个但是它没有用。

object DataBatcher extends Logger {

  var batchList: ListBuffer[PersonBatch] = ListBuffer[PersonBatch]()
  var batchSize: Long = 500  //default batch size

  def addToBatchList(batch: PersonBatch): Unit = {
    batchList += batch
  }

  def clearBatchList(): Unit = {
    batchList.clear()
  }

  def createBatches(ds: Dataset[Person]): Dataset[PersonBatch] = {

    val dsCount = ds.count()
    logger.info(s"Count of dataset passed for creating batches : ${dsCount}")
    val batchElement = ListBuffer[Person]()
    val batch = PersonBatch(batchElement)
    ds.foreach(x => {
      batch.personBatch += x
      if(batch.personBatch.length == batchSize) {
        addToBatchList(batch)
        batch.requestBatch.clear()
      }
    })
    if(batch.personBatch.length > 0) {
      addToBatchList(batch)
      batch.personBatch.clear()
    }
    sparkSession.createDataset(batchList)
  }  
}

我想在Hadoop集群上运行这项工作。有人可以帮我这个吗？

Answer 1

rdd.iterator具有分组功能可能对您有用。

例如：

iter.grouped（BATCHSIZE）

示例代码片段，其中使用iter.grouped（batchsize）进行批量插入，其中1000和Im尝试插入数据库

   df.repartition(numofpartitionsyouwant) // numPartitions ~ number of simultaneous DB connections you can planning to give...
def insertToTable(sqlDatabaseConnectionString: String,
                  sqlTableName: String): Unit = {

  val tableHeader: String = dataFrame.columns.mkString(",")
  dataFrame.foreachPartition { partition =>
    //NOTE : EACH PARTITION ONE CONNECTION (more better way is to use connection pools)
    val sqlExecutorConnection: Connection =
      DriverManager.getConnection(sqlDatabaseConnectionString)
    //Batch size of 1000 is used since some databases cant use batch size more than 1000 for ex : Azure sql
    partition.grouped(1000).foreach { group =>
      val insertString: scala.collection.mutable.StringBuilder =
        new scala.collection.mutable.StringBuilder()

      group.foreach { record =>
        insertString.append("('" + record.mkString(",") + "'),")
      }

      sqlExecutorConnection
        .createStatement()
        .executeUpdate(f"INSERT INTO [$sqlTableName] ($tableHeader) VALUES "
          + insertString.stripSuffix(","))
    }

    sqlExecutorConnection.close() // close the connection so that connections wont exhaust.
  }
}

Answer 2

val tableHeader: String = dataFrame.columns.mkString(",")
dataFrame.foreachPartition((it: Iterator[Row]) => {
      println("partition index: " )
      val url = "jdbc:...+ "user=;password=;"
      val conn = DriverManager.getConnection(url)
      conn.setAutoCommit(true)
      val stmt = conn.createStatement()
      val batchSize = 10
      var i =0
      while (it.hasNext) {
        val row = it.next
        import java.sql.SQLException
        import java.sql.SQLIntegrityConstraintViolationException
        try {
          stmt.addBatch(" UPDATE TABLE SET STATUS = 0 , " +
            " DATE ='" + new java.sql.Timestamp(System.currentTimeMillis()) +"'" +
            " where id = " + row.getAs("IDNUM")  )
          i += 1
          if (  i  % batchSize == 0 ) {
            stmt.executeBatch
            conn.commit
          }
        } catch {
          case e: SQLIntegrityConstraintViolationException =>
          case e: SQLException =>
            e.printStackTrace()
        }
        finally {
             stmt.executeBatch
             conn.commit
        }

      }
      import java.util
      val ret = stmt.executeBatch
      System.out.println("Ret val: " + util.Arrays.toString(ret))
      System.out.println("Update count: " + stmt.getUpdateCount)
      conn.commit
      stmt.close

批处理数据集Spark scala

2 个答案: