sql broadcast hashjoin示例(java api)

Dataset的groupBy agg示例

Dataset<Row> resultDs = dsParsed
.groupBy("enodeb_id", "ecell_id")
.agg(
    functions.first("scan_start_time").alias("scan_start_time1"),
    functions.first("insert_time").alias("insert_time1"),
    functions.first("mr_type").alias("mr_type1"),
    functions.first("mr_ltescphr").alias("mr_ltescphr1"),
    functions.first("mr_ltescpuschprbnum").alias("mr_ltescpuschprbnum1"),
    functions.count("enodeb_id").alias("rows1"))
.selectExpr(
    "ecell_id", 
    "enodeb_id",
    "scan_start_time1 as scan_start_time",
    "insert_time1 as insert_time",
    "mr_type1 as mr_type",
    "mr_ltescphr1 as mr_ltescphr",
    "mr_ltescpuschprbnum1 as mr_ltescpuschprbnum",
    "rows1 as rows");

Dataset Join示例：

        Dataset<Row> ncRes = sparkSession.read().option("delimiter", "|").option("header", true).csv("/user/csv");
        Dataset<Row> mro=sparkSession.sql("。。。");

        Dataset<Row> ncJoinMro = ncRes
                .join(mro, mro.col("id").equalTo(ncRes.col("id")).and(mro.col("calid").equalTo(ncRes.col("calid"))), "left_outer")
                .select(ncRes.col("id").as("int_id"), 
                        mro.col("vendor_id"),
                         。。。
                 );

join condition另外一种方式:

leftDfWithWatermark.join(rightDfWithWatermark, 
　　expr(""" leftDfId = rightDfId AND leftDfTime >= rightDfTime AND leftDfTime <= rightDfTime + interval 1 hour"""), 
　　joinType = "leftOuter" )

BroadcastHashJoin示例：

package com.dx.testbroadcast;

import org.apache.spark.SparkConf;
import org.apache.spark.sql.Dataset;
import org.apache.spark.sql.Row;
import org.apache.spark.sql.SparkSession;
import org.apache.spark.sql.functions;

import java.io.*;

public class Test {
    public static void main(String[] args) {
        String personPath = "E:\\person.csv";
        String personOrderPath = "E:\\personOrder.csv";
        //writeToPersion(personPath);
        //writeToPersionOrder(personOrderPath);

        SparkConf conf = new SparkConf();
        SparkSession sparkSession = SparkSession.builder().config(conf).appName("test-broadcast-app").master("local[*]").getOrCreate();

        Dataset<Row> person = sparkSession.read()
                .option("header", "true")
                .option("inferSchema", "true") //是否自动推到内容的类型
                .option("delimiter", ",").csv(personPath).as("person");
        person.printSchema();

        Dataset<Row> personOrder = sparkSession.read()
                .option("header", "true")
                .option("inferSchema", "true") //是否自动推到内容的类型
                .option("delimiter", ",").csv(personOrderPath).as("personOrder");
        personOrder.printSchema();

        // Default `inner`. Must be one of:`inner`, `cross`, `outer`, `full`, `full_outer`, `left`, `left_outer`,`right`, `right_outer`, `left_semi`, `left_anti`.
        Dataset<Row> resultDs = personOrder.join(functions.broadcast(person), personOrder.col("personid").equalTo(person.col("id")),"left");
        resultDs.explain();
        resultDs.show(10);
    }

    private static void writeToPersion(String personPath) {
        BufferedWriter personWriter = null;
        try {
            personWriter = new BufferedWriter(new FileWriter(personPath));
            personWriter.write("id,name,age,address\r\n");
            for (int i = 0; i < 10000; i++) {
                personWriter.write("" + i + ",person-" + i + "," + i + ",address-address-address-address-address-address-address" + i + "\r\n");
            }
        } catch (Exception e) {
            e.printStackTrace();
        } finally {
            if (personWriter != null) {
                try {
                    personWriter.close();
                } catch (IOException e) {
                    e.printStackTrace();
                }
            }
        }
    }

    private static void writeToPersionOrder(String personOrderPath) {
        BufferedWriter personWriter = null;
        try {
            personWriter = new BufferedWriter(new FileWriter(personOrderPath));
            personWriter.write("personid,name,age,address\r\n");
            for (int i = 0; i < 1000000; i++) {
                personWriter.write("" + i + ",person-" + i + "," + i + ",address-address-address-address-address-address-address" + i + "\r\n");
            }
        } catch (Exception e) {
            e.printStackTrace();
        } finally {
            if (personWriter != null) {
                try {
                    personWriter.close();
                } catch (IOException e) {
                    e.printStackTrace();
                }
            }
        }
    }
}

打印结果：

== Physical Plan ==
*(2) BroadcastHashJoin [personid#28], [id#10], LeftOuter, BuildRight
:- *(2) FileScan csv [personid#28,name#29,age#30,address#31] Batched: false, Format: CSV, Location: InMemoryFileIndex[file:/E:/personOrder.csv], PartitionFilters: [], PushedFilters: [], ReadSchema: struct<personid:int,name:string,age:int,address:string>
+- BroadcastExchange HashedRelationBroadcastMode(List(cast(input[0, int, true] as bigint)))
   +- *(1) Project [id#10, name#11, age#12, address#13]
      +- *(1) Filter isnotnull(id#10)
         +- *(1) FileScan csv [id#10,name#11,age#12,address#13] Batched: false, Format: CSV, Location: InMemoryFileIndex[file:/E:/person.csv], PartitionFilters: [], PushedFilters: [IsNotNull(id)], ReadSchema: struct<id:int,name:string,age:int,address:string>

+--------+--------+---+--------------------+---+--------+---+--------------------+
|personid|    name|age|             address| id|    name|age|             address|
+--------+--------+---+--------------------+---+--------+---+--------------------+
|       0|person-0|  0|address-address-a...|  0|person-0|  0|address-address-a...|
|       1|person-1|  1|address-address-a...|  1|person-1|  1|address-address-a...|
|       2|person-2|  2|address-address-a...|  2|person-2|  2|address-address-a...|
|       3|person-3|  3|address-address-a...|  3|person-3|  3|address-address-a...|
|       4|person-4|  4|address-address-a...|  4|person-4|  4|address-address-a...|
|       5|person-5|  5|address-address-a...|  5|person-5|  5|address-address-a...|
|       6|person-6|  6|address-address-a...|  6|person-6|  6|address-address-a...|
|       7|person-7|  7|address-address-a...|  7|person-7|  7|address-address-a...|
|       8|person-8|  8|address-address-a...|  8|person-8|  8|address-address-a...|
|       9|person-9|  9|address-address-a...|  9|person-9|  9|address-address-a...|
+--------+--------+---+--------------------+---+--------+---+--------------------+
only showing top 10 rows

SparkSQL Broadcast HashJoin

        person.createOrReplaceTempView("temp_person");
        personOrder.createOrReplaceTempView("temp_person_order");

        Dataset<Row> sqlResult = sparkSession.sql(
                " SELECT /*+ BROADCAST (t11) */" +
                " t11.id,t11.name,t11.age,t11.address," +
                " t10.personid as person_id,t10.name as persion_order_name" +
                " FROM temp_person_order as t10 " +
                " inner join temp_person as t11" +
                " on t11.id = t10.personid ");
        sqlResult.show(10);
        sqlResult.explain();

打印日志

+---+--------+---+--------------------+---------+------------------+
| id|    name|age|             address|person_id|persion_order_name|
+---+--------+---+--------------------+---------+------------------+
|  0|person-0|  0|address-address-a...|        0|          person-0|
|  1|person-1|  1|address-address-a...|        1|          person-1|
|  2|person-2|  2|address-address-a...|        2|          person-2|
|  3|person-3|  3|address-address-a...|        3|          person-3|
|  4|person-4|  4|address-address-a...|        4|          person-4|
|  5|person-5|  5|address-address-a...|        5|          person-5|
|  6|person-6|  6|address-address-a...|        6|          person-6|
|  7|person-7|  7|address-address-a...|        7|          person-7|
|  8|person-8|  8|address-address-a...|        8|          person-8|
|  9|person-9|  9|address-address-a...|        9|          person-9|
+---+--------+---+--------------------+---------+------------------+
only showing top 10 rows

19/06/24 09:35:50 INFO FileSourceStrategy: Pruning directories with: 
19/06/24 09:35:50 INFO FileSourceStrategy: Post-Scan Filters: isnotnull(personid#28)
19/06/24 09:35:50 INFO FileSourceStrategy: Output Data Schema: struct<personid: int, name: string>
19/06/24 09:35:50 INFO FileSourceScanExec: Pushed Filters: IsNotNull(personid)
19/06/24 09:35:50 INFO FileSourceStrategy: Pruning directories with: 
19/06/24 09:35:50 INFO FileSourceStrategy: Post-Scan Filters: isnotnull(id#10)
19/06/24 09:35:50 INFO FileSourceStrategy: Output Data Schema: struct<id: int, name: string, age: int, address: string ... 2 more fields>
19/06/24 09:35:50 INFO FileSourceScanExec: Pushed Filters: IsNotNull(id)
== Physical Plan ==
*(2) Project [id#10, name#11, age#12, address#13, personid#28 AS person_id#94, name#29 AS persion_order_name#95]
+- *(2) BroadcastHashJoin [personid#28], [id#10], Inner, BuildRight
   :- *(2) Project [personid#28, name#29]
   :  +- *(2) Filter isnotnull(personid#28)
   :     +- *(2) FileScan csv [personid#28,name#29] Batched: false, Format: CSV, Location: InMemoryFileIndex[file:/E:/personOrder.csv], PartitionFilters: [], PushedFilters: [IsNotNull(personid)], ReadSchema: struct<personid:int,name:string>
   +- BroadcastExchange HashedRelationBroadcastMode(List(cast(input[0, int, true] as bigint)))
      +- *(1) Project [id#10, name#11, age#12, address#13]
         +- *(1) Filter isnotnull(id#10)
            +- *(1) FileScan csv [id#10,name#11,age#12,address#13] Batched: false, Format: CSV, Location: InMemoryFileIndex[file:/E:/person.csv], PartitionFilters: [], PushedFilters: [IsNotNull(id)], ReadSchema: struct<id:int,name:string,age:int,address:string>
19/06/24 09:35:50 INFO SparkContext: Invoking stop() from shutdown hook

posted @ 2018-10-12 10:04 cctext 阅读(3943) 评论(0) 编辑收藏举报

会员力量，点亮园子希望

刷新页面返回顶部

yy

基础才是编程人员应该深入研究的问题，警告自己问题解决不了时，多从运行原理底层研究后再考虑方案。

Spark中使用Dataset的groupBy/agg/join/broadcast hasjoin/sql broadcast hashjoin示例(java api)

Dataset的groupBy agg示例

Dataset Join示例：

BroadcastHashJoin示例：

SparkSQL Broadcast HashJoin

公告