一、用Sparkcontext讀入文件，map逐行用Gson解析，輸出轉成一個caseclass類，填充各字段，輸出。

解析JSON這里沒有什么問題。

RDD覆蓋寫的時候碰到了一些問題 ：

1.直接saveAsTextFile沒有覆蓋true參數；

2.轉dataframe時，還得一個一個字段顯化才能轉成dataframe；

3.write時，一開始打算寫text，說字段里不能含有l(wèi)ong和int，換成string后，說不能有多個字段，只能有一個string。剛好用parquet存儲省空間，就存parquet了。

跑通代碼如下：

package XXX

import com.google.gson.JsonParser
import org.apache.spark.sql.SparkSession
import schema.caseClass1

object ParsecaseClass1Json {
  def main(args: Array[String]): Unit = {
    val inputPath = args(0)
    val outputPath = args(1)
    val sparkSession = SparkSession.builder().appName(this.getClass.getSimpleName).getOrCreate()
    import sparkSession.implicits._

    val lines = sparkSession.sparkContext.textFile(inputPath)
    val result = lines.map(=> parseJsonStr(str))

    val df = result.map(
      x => (
        x.adId
        , x.campaignId
        , x.settlementType
        , x.billingType
        , x.billingTypeCode
        , x.packageName
      ))
      .toDF()
    df.coalesce(1).write.format("parquet").mode("overwrite").save(outputPath)

  }

  def parseJsonStr(str: String): caseClass1 = {
      val inputJson = new JsonParser().parse(str).getAsJsonObject
      val object = new caseClass1

      //1.campaignId
      if (inputJson.has("campaign")) {
        val campaign = inputJson.getAsJsonObject("campaign")
        var campaignId: Long = 0
        if (campaign.has("id"))
          campaignId = campaign.getAsJsonPrimitive("id").getAsLong
        else if (campaign.has("campaignId")) {
          campaignId = campaign.getAsJsonPrimitive("campaignId").getAsLong
        }
        else {
          System.err.println("No campaignId, inputJson: {}", str)
        }
        object.campaignId = campaignId
      }

      //2.creativeId
      if (inputJson.has("creative")) {
        val creative = inputJson.getAsJsonObject("creative")
        var adId: Long = 0
        if (creative.has("id"))
          adId = creative.getAsJsonPrimitive("id").getAsLong
        else if (creative.has("creativeId"))
          adId = creative.getAsJsonPrimitive("creativeId").getAsLong
        else
          System.err.println("No adId, inputJson: {}", str)
        object.adId = adId

      }

      /*
      3.settlementType
      4.billingType
      5.billingTypeCode
      6.appId -> packageName
      */
      if (inputJson.has("group")) {
        val group = inputJson.getAsJsonObject("group")

        object.settlementType = group.getAsJsonPrimitive("settlementType").getAsString
        object.billingType = group.getAsJsonPrimitive("billingType").getAsString
        object.billingTypeCode = group.getAsJsonPrimitive("billingTypeCode").getAsInt
        object.packageName = group.getAsJsonPrimitive("appId").getAsString
      }
    System.err.println(object.toString)
    object
  }

}

caseClass如下文章來源地址http://www.zghlxwxcb.cn/news/detail-817176.html

package XXX;

import java.io.Serializable;

public class caseClass1 implements Serializable {

    private static final long serialVersionUID = ***;

    public long adId = 0;
    public long campaignId = 0;
    public String settlementType;
    public String billingType;
    public Integer billingTypeCode;
    public String packageName;

    @Override
    public String toString() {
        return "caseClass1{" +
                "adId=" + adId +
                ", campaignId=" + campaignId +
                ", settlementType='" + settlementType + '\'' +
                ", billingType='" + billingType + '\'' +
                ", billingTypeCode=" + billingTypeCode +
                ", packageName='" + packageName + '\'' +
                '}';
    }
    public caseClass1() {

    }
}

到了這里，關于Spark解析JSON文件，寫入hdfs的文章就介紹完了。如果您還想了解更多內容，請在右上角搜索TOY模板網以前的文章或繼續(xù)瀏覽下面的相關文章，希望大家以后多多支持TOY模板網！