简述

spark在2.2.0版本是不支持通过jdbc的方式直接访问hive数据的，需要修改部分源码实现spark直接通过jdbc的方式读取hive数据，就在之前写的文章中的方法二里。

https://blog.csdn.net/qq_42213403/article/details/117557610?spm=1001.2014.3001.5501

还有一种方法不用重写源码，是先通过jdbc获取数据，再用spark封装成dataframe的方式操作的

实现过程

首先使用jdbc查询的方式获取hive表数据

  def getResult()={
    val properties = new Properties
    properties.setProperty("url", "jdbc:hive2://192.168.5.61:10000/")
    properties.setProperty("user", "hive")
    properties.setProperty("password", "")
    properties.setProperty("driver", "org.apache.hive.jdbc.HiveDriver")
    val connection = getConnection(properties)
    val statement = connection.createStatement
    val resultSet = statement.executeQuery("select * from test.user_info")
    resultSet
  }

  def getConnection(prop: Properties): Connection = try {
    Class.forName(prop.getProperty("driver"))
    conn = DriverManager.getConnection(prop.getProperty("url"), prop.getProperty("user"), prop.getProperty("password"))
    conn
  } catch {
    case e: Exception =>
      e.printStackTrace()
      null
  }

把查出的ResultSet转换成DataFrame

  def createStructField(name:String,colType:String):StructField={
    colType match {
      case "java.lang.String" =>{StructField(name,StringType,true)}
      case "java.lang.Integer" =>{StructField(name,IntegerType,true)}
      case "java.lang.Long" =>{StructField(name,LongType,true)}
      case "java.lang.Boolean" =>{StructField(name,BooleanType,true)}
      case "java.lang.Double" =>{StructField(name,DoubleType,true)}
      case "java.lang.Float" =>{StructField(name,FloatType,true)}
      case "java.sql.Date" =>{StructField(name,DateType,true)}
      case "java.sql.Time" =>{StructField(name,TimestampType,true)}
      case "java.sql.Timestamp" =>{StructField(name,TimestampType,true)}
      case "java.math.BigDecimal" =>{StructField(name,DecimalType(10,0),true)}
    }
  }
  /**
   * 把查出的ResultSet转换成DataFrame
   */
  def createResultSetToDF(rs:ResultSet,sparkSession: SparkSession):DataFrame= {
    val rsmd = rs.getMetaData
    val columnTypeList = new util.ArrayList[String]
    val rowSchemaList = new util.ArrayList[StructField]
    for (i <- 1 to rsmd.getColumnCount) {
      var temp = rsmd.getColumnClassName(i)
      temp = temp.substring(temp.lastIndexOf(".") + 1)
      if ("Integer".equals(temp)) {
        temp = "Int";
      }
      columnTypeList.add(temp)
      rowSchemaList.add(createStructField(rsmd.getColumnName(i), rsmd.getColumnClassName(i)))
    }
    val rowSchema = StructType(rowSchemaList)
    //ResultSet反射类对象
    val rsClass = rs.getClass
    var count = 1
    val resultList = new util.ArrayList[Row]
    var totalDF = sparkSession.createDataFrame(new util.ArrayList[Row], rowSchema)
    while (rs.next()) {
      count = count + 1
      //      val temp = new util.ArrayList[Object]
      val buffer = new ArrayBuffer[Any]()
      for (i <- 0 to columnTypeList.size() - 1) {
        val method = rsClass.getMethod("get" + columnTypeList.get(i), "aa".getClass)
        buffer+=method.invoke(rs, rsmd.getColumnName(i + 1))
      }
      resultList.add(Row(buffer: _*))
      if (count % 100000 == 0) {
        val tempDF = sparkSession.createDataFrame(resultList, rowSchema)
        totalDF = totalDF.union(tempDF).distinct()
        resultList.clear()
      }
    }
    val tempDF = sparkSession.createDataFrame(resultList, rowSchema)
    totalDF = totalDF.union(tempDF)
    totalDF
  }

运行代码

    val spark = SparkSession.builder()
      .master("local[2]")
      .appName("test")
      .getOrCreate()
    
    val df = createResultSetToDF(getResult(),spark)
    df.show()

结果

完整代码

import java.sql.{Connection, DriverManager, ResultSet}
import java.util
import java.util.Properties

import org.apache.spark.sql.{DataFrame, Row, SparkSession}
import org.apache.spark.sql.types.{BooleanType, DateType, DecimalType, DoubleType, FloatType, IntegerType, LongType, StringType, StructField, StructType, TimestampType}

import scala.collection.mutable.ArrayBuffer

/**
 * @Author: fcy
 * @Date: 2021/6/4 4:16 下午
 */
object SparkJDBCHiveDataFrame {
  def main(args: Array[String]): Unit = {
    val spark = SparkSession.builder()
      .master("local[2]")
      .appName("test")
      .getOrCreate()

    val df = createResultSetToDF(getResult(),spark)
    df.show()
  }

  def getResult()={
    val properties = new Properties
    properties.setProperty("url", "jdbc:hive2://xxx:10000/")
    properties.setProperty("user", "hive")
    properties.setProperty("password", xxx)
    properties.setProperty("driver", "org.apache.hive.jdbc.HiveDriver")
    val connection = getConnection(properties)
    val statement = connection.createStatement
    val resultSet = statement.executeQuery("select * from test.user_info")
    resultSet
  }

  def getConnection(prop: Properties): Connection = try {
    Class.forName(prop.getProperty("driver"))
    conn = DriverManager.getConnection(prop.getProperty("url"), prop.getProperty("user"), prop.getProperty("password"))
    conn
  } catch {
    case e: Exception =>
      e.printStackTrace()
      null
  }


  def createStructField(name:String,colType:String):StructField={
    colType match {
      case "java.lang.String" =>{StructField(name,StringType,true)}
      case "java.lang.Integer" =>{StructField(name,IntegerType,true)}
      case "java.lang.Long" =>{StructField(name,LongType,true)}
      case "java.lang.Boolean" =>{StructField(name,BooleanType,true)}
      case "java.lang.Double" =>{StructField(name,DoubleType,true)}
      case "java.lang.Float" =>{StructField(name,FloatType,true)}
      case "java.sql.Date" =>{StructField(name,DateType,true)}
      case "java.sql.Time" =>{StructField(name,TimestampType,true)}
      case "java.sql.Timestamp" =>{StructField(name,TimestampType,true)}
      case "java.math.BigDecimal" =>{StructField(name,DecimalType(10,0),true)}
    }
  }
  /**
   * 把查出的ResultSet转换成DataFrame
   */
  def createResultSetToDF(rs:ResultSet,sparkSession: SparkSession):DataFrame= {
    val rsmd = rs.getMetaData
    val columnTypeList = new util.ArrayList[String]
    val rowSchemaList = new util.ArrayList[StructField]
    for (i <- 1 to rsmd.getColumnCount) {
      var temp = rsmd.getColumnClassName(i)
      temp = temp.substring(temp.lastIndexOf(".") + 1)
      if ("Integer".equals(temp)) {
        temp = "Int";
      }
      columnTypeList.add(temp)
      rowSchemaList.add(createStructField(rsmd.getColumnName(i), rsmd.getColumnClassName(i)))
    }
    val rowSchema = StructType(rowSchemaList)
    //ResultSet反射类对象
    val rsClass = rs.getClass
    var count = 1
    val resultList = new util.ArrayList[Row]
    var totalDF = sparkSession.createDataFrame(new util.ArrayList[Row], rowSchema)
    while (rs.next()) {
      count = count + 1
      //      val temp = new util.ArrayList[Object]
      val buffer = new ArrayBuffer[Any]()
      for (i <- 0 to columnTypeList.size() - 1) {
        val method = rsClass.getMethod("get" + columnTypeList.get(i), "aa".getClass)
        buffer+=method.invoke(rs, rsmd.getColumnName(i + 1))
      }
      resultList.add(Row(buffer: _*))
      if (count % 100000 == 0) {
        val tempDF = sparkSession.createDataFrame(resultList, rowSchema)
        totalDF = totalDF.union(tempDF).distinct()
        resultList.clear()
      }
    }
    val tempDF = sparkSession.createDataFrame(resultList, rowSchema)
    totalDF = totalDF.union(tempDF)
    totalDF
  }


}

欢迎留言讨论和指正

标签：case,jdbc,java,name,val,StructField,DataFrame,spark,true
来源： https://blog.csdn.net/qq_42213403/article/details/117564393