Spark工程开发常用函数与方法(Scala语言)

用户3003813

发布于 2018-09-06 13:14:41

7420

发布于 2018-09-06 13:14:41

import org.apache.spark.{SparkContext, SparkConf} import org.apache.spark.sql.{SaveMode, DataFrame} import scala.collection.mutable.ArrayBuffer import main.asiainfo.coc.tools.Configure import org.apache.spark.sql.hive.HiveContext import java.sql.DriverManager import java.sql.Connection

1 连接前台数据源查询前台MYSQL中的数据

val DIM_COC_INDEX_INFO_DDL = s"""
CREATE TEMPORARY TABLE DIM_COC_INDEX_INFO
USING org.apache.spark.sql.jdbc
OPTIONS (
url '${mySQLUrl}',
dbtable 'DIM_COC_INDEX_INFO'
)""".stripMargin

sqlContext.sql(DIM_COC_INDEX_INFO_DDL)
val DIM_COC_INDEX_INFO = sql("SELECT * FROM DIM_COC_INDEX_INFO").cache()

2 在A表中筛选出 B表中获取的TARGET_TABLE_CODE 然后再按照DATA_SRC_CODE排序，查询出源表的集合

val sources = DIM_COC_INDEX_INFO.filter("TARGET_TABLE_CODE ='"+TARGET_TABLE_CODE+"'")
        .select("DATA_SRC_CODE").groupBy("DATA_SRC_CODE").agg(DIM_COC_INDEX_INFO("DATA_SRC_CODE")).collect

3 将表进行关联

resultIndexTableDF = resultIndexTableDF.join(SOURCE_TABLE,ALL_USERS.col(ALL_USER_JOIN_COLUMN_NAME) === SOURCE_TABLE.col(SOURCE_TABLE_JOIN_COLUMN_NAME),"left_outer")
resultIndexTableDF.dtypes.foreach(println)

4 根据条件筛选

val labels = CI_MDA_SYS_TABLE.join(CI_MDA_SYS_TABLE_COLUMN,CI_MDA_SYS_TABLE("TABLE_ID") === CI_MDA_SYS_TABLE_COLUMN("TABLE_ID"),"inner")
      .join(CI_LABEL_EXT_INFO,CI_MDA_SYS_TABLE_COLUMN("COLUMN_ID") === CI_LABEL_EXT_INFO("COLUMN_ID"),"inner")
      .join(CI_LABEL_INFO,CI_LABEL_EXT_INFO("LABEL_ID") === CI_LABEL_INFO("LABEL_ID"),"inner")
      .join(CI_APPROVE_STATUS,CI_LABEL_INFO("LABEL_ID") === CI_APPROVE_STATUS("RESOURCE_ID"),"inner")
      .filter(CI_APPROVE_STATUS("CURR_APPROVE_STATUS_ID") === CI_APPROVE_STATUS_SUCCESS_CODE
      and (CI_LABEL_INFO("DATA_STATUS_ID") === 1 || CI_LABEL_INFO("DATA_STATUS_ID") === 2)
      and (CI_LABEL_EXT_INFO("COUNT_RULES_CODE") isNotNull  //TODO   trim.length>0
      )
      and CI_MDA_SYS_TABLE("UPDATE_CYCLE") === TABLE_DATA_CYCLE
      ).cache()

5 根据某字段对表进行排序

    val labelTargetTables = labels.groupBy("CI_MDA_SYS_TABLE.TABLE_ID","CI_MDA_SYS_TABLE.TABLE_NAME").agg(labels("CI_MDA_SYS_TABLE.TABLE_ID"),labels("CI_MDA_SYS_TABLE.TABLE_NAME")).collect

6 创建parquet格式的表可使用schema.生成到指定的schema.

        sqlContext.sql("create table "+labelTargetTableName+" stored as parquet as select * from default."+labelTargetTableNameJson)

7 保存数据格式，可以指定生成的格式

 resultLabelTable.saveAsTable(tableName = labelTargetTableName, source="parquet", mode=SaveMode.Overwrite)

8 根据筛选查询出相应数据，由于cache方法并不属于action操作，接下来的操作需要这一步所执行的数据信息，所以这里使用collect方法,再执行遍历方法

      val r0000Labels = labelInThisTargetTable.filter("COUNT_RULES_CODE = 'R_00000'").select("CI_LABEL_INFO.LABEL_ID","COLUMN_NAME").collect
for(r0000Label <- r0000Labels){
   ........
}

本文参与腾讯云自媒体分享计划，分享自作者个人站点/博客。

原始发表：2015-10-10 ，如有侵权请联系 cloudcommunity@tencent.com 删除

spark