Spark 结构化流：JDBC 接收器中的主键

Question

我正在使用具有更新模式的结构化流从 kafka 主题读取数据流，然后进行一些转换。

然后我创建了一个 jdbc sink 以 Append 模式将数据推送到 mysql sink 中。 问题是我如何告诉我的接收器让它知道这是我的主键并基于它进行更新，以便我的表不应该有任何重复的行。

   val df: DataFrame = spark
  .readStream
  .format("kafka")
  .option("kafka.bootstrap.servers", "<List-here>")
  .option("subscribe", "emp-topic")
  .load()


  import spark.implicits._
  // value in kafka is bytes so cast it to String
  val empList: Dataset[Employee] = df.
  selectExpr("CAST(value AS STRING)")
  .map(row => Employee(row.getString(0)))

  // window aggregations on 1 min windows
  val aggregatedDf= ......

  // How to tell here that id is my primary key and do the update
  // based on id column
  aggregatedDf
  .writeStream
  .trigger(Trigger.ProcessingTime(60.seconds))
  .outputMode(OutputMode.Update)
  .foreachBatch { (batchDF: DataFrame, batchId: Long) =>
      batchDF
      .select("id", "name","salary","dept")
      .write.format("jdbc")
      .option("url", "jdbc:mysql://localhost/empDb")
      .option("driver","com.mysql.cj.jdbc.Driver")
      .option("dbtable", "empDf")
      .option("user", "root")
      .option("password", "root")
      .mode(SaveMode.Append)
      .save()
     }

Answer 1

一种方法是，可以将foreachPartition ON DUPLICATE KEY UPDATE与foreachPartition一起使用foreachPartition到此目的。

以下是伪代码片段

/**
    * Insert in to database using foreach partition.
    * @param dataframe : DataFrame
    * @param sqlDatabaseConnectionString
    * @param sqlTableName
    */
  def insertToTable(dataframe: DataFrame, sqlDatabaseConnectionString: String, sqlTableName: String): Unit = {

//numPartitions = number of simultaneous DB connections you can planning to give
datframe.repartition(numofpartitionsyouwant)

    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(",") + "'),")
          }

val sql =   s"""
               | INSERT INTO $sqlTableName  VALUES  
               | $tableHeader
               | ${insertString}
               | ON DUPLICATE KEY UPDATE 
               | yourprimarykeycolumn='${record.getAs[String]("key")}'
    sqlExecutorConnection.createStatement()
                .executeUpdate(sql)
          }
    sqlExecutorConnection.close() // close the connection
        }
      }

您可以使用preparestatement代替jdbc语句。

进一步阅读： SPARK SQL-使用DataFrames和JDBC更新MySql表

Answer 2

您知道吗，为什么我使用devdb的writestream和jdbc一起出现此错误？

java.lang.UnsupportedOperationException：数据源jdbc不支持流式写入

另外，我听说有一种解决方法是引入foreachBatch，我尝试使用.foreachBatch {（（batchDF：DataFrame，batchId：Long）=> batchDF .writeStream ....，但出现此错误：foreachBatch值不是组织成员。 apache.spark.sql.streaming.DataStreamWriter [org.apache.spark.sql.Row]