大数据下一代变革之必研究数据湖技术Hudi原理实战双管齐下-下

• 2022 年 11 月 29 日
• 笔记
• 大数据, 数据仓库

@

集成Spark开发

Spark编程读写示例

通过IDE如Idea编程实质上和前面的spark-shell和spark-sql相似，其他都是Spark编程的知识，下面以scala语言为示例，idea新建scala的maven项目

pom文件添加如下依赖

<project xmlns="//maven.apache.org/POM/4.0.0" xmlns:xsi="//www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="//maven.apache.org/POM/4.0.0 //maven.apache.org/maven-v4_0_0.xsd">
  <modelVersion>4.0.0</modelVersion>
  <groupId>cn.itxs</groupId>
  <artifactId>hoodie-spark-demo</artifactId>
  <version>1.0</version>

  <properties>
    <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
    <scala.version>2.12.10</scala.version>
    <scala.binary.version>2.12</scala.binary.version>
    <spark.version>3.3.0</spark.version>
    <hoodie.version>0.12.1</hoodie.version>
    <hadoop.version>3.3.4</hadoop.version>
  </properties>

  <dependencies>
    <dependency>
      <groupId>org.scala-lang</groupId>
      <artifactId>scala-library</artifactId>
      <version>${scala.version}</version>
    </dependency>

    <dependency>
      <groupId>org.apache.spark</groupId>
      <artifactId>spark-core_${scala.binary.version}</artifactId>
      <version>${spark.version}</version>
      <scope>provided</scope>
    </dependency>

    <dependency>
      <groupId>org.apache.spark</groupId>
      <artifactId>spark-sql_${scala.binary.version}</artifactId>
      <version>${spark.version}</version>
      <scope>provided</scope>
    </dependency>

    <dependency>
      <groupId>org.apache.spark</groupId>
      <artifactId>spark-hive_${scala.binary.version}</artifactId>
      <version>${spark.version}</version>
      <scope>provided</scope>
    </dependency>

    <dependency>
      <groupId>org.apache.hadoop</groupId>
      <artifactId>hadoop-client</artifactId>
      <version>${hadoop.version}</version>
      <scope>provided</scope>
    </dependency>

    <dependency>
      <groupId>org.apache.hudi</groupId>
      <artifactId>hudi-spark3.3-bundle_${scala.binary.version}</artifactId>
      <version>${hoodie.version}</version>
      <scope>provided</scope>
    </dependency>
  </dependencies>

  <build>
    <plugins>
      <plugin>
        <groupId>org.apache.maven.plugins</groupId>
        <artifactId>maven-compiler-plugin</artifactId>
        <version>3.10.1</version>
        <configuration>
          <source>1.8</source>
          <target>1.8</target>
          <encoding>${project.build.sourceEncoding}</encoding>
        </configuration>
      </plugin>
      <plugin>
        <groupId>org.scala-tools</groupId>
        <artifactId>maven-scala-plugin</artifactId>
        <version>2.15.2</version>
        <executions>
          <execution>
            <goals>
              <goal>compile</goal>
              <goal>testCompile</goal>
            </goals>
          </execution>
        </executions>
      </plugin>
      <plugin>
        <groupId>org.apache.maven.plugins</groupId>
        <artifactId>maven-shade-plugin</artifactId>
        <version>3.2.4</version>
        <executions>
          <execution>
            <phase>package</phase>
            <goals>
              <goal>shade</goal>
            </goals>
            <configuration>
              <filters>
                <filter>
                  <artifact>*:*</artifact>
                  <excludes>
                    <exclude>META-INF/*.SF</exclude>
                    <exclude>META-INF/*.DSA</exclude>
                    <exclude>META-INF/*.RSA</exclude>
                  </excludes>
                </filter>
              </filters>
            </configuration>
          </execution>
        </executions>
      </plugin>
    </plugins>
  </build>
</project>

创建常量对象

object Constant {
    val HUDI_STORAGE_PATH = "hdfs://192.168.5.53:9000/tmp/"
}

插入hudi数据

package cn.itxs

import org.apache.spark.sql.SparkSession
import org.apache.spark.SparkConf
import org.apache.hudi.QuickstartUtils._
import scala.collection.JavaConversions._
import org.apache.spark.sql.SaveMode._
import org.apache.hudi.DataSourceWriteOptions._
import org.apache.hudi.config.HoodieWriteConfig._

object InsertDemo {
  def main(args: Array[String]): Unit = {
    val sparkConf = new SparkConf()
      .setAppName(this.getClass.getSimpleName)
      .setMaster("local[*]")
      .set("spark.serializer", "org.apache.spark.serializer.KryoSerializer")

    val sparkSession = SparkSession.builder()
      .config(sparkConf)
      .enableHiveSupport()
      .getOrCreate()

    val tableName = "hudi_trips_cow_idea"
    val basePath =  Constant.HUDI_STORAGE_PATH+tableName
    val dataGen = new DataGenerator
    val inserts = convertToStringList(dataGen.generateInserts(10))

    val df = sparkSession.read.json(sparkSession.sparkContext.parallelize(inserts,2))
    df.write.format("hudi").
      options(getQuickstartWriteConfigs).
      option(PRECOMBINE_FIELD.key(), "ts").
      option(RECORDKEY_FIELD.key(), "uuid").
      option(PARTITIONPATH_FIELD.key(), "partitionpath").
      option(TBL_NAME.key(), tableName).
      mode(Overwrite).
      save(basePath)

    sparkSession.close()
  }
}

由于依赖中scope是配置为provided，因此运行配置中勾选下面这项

运行InsertDemo程序写入hudi数据

运行ReadDemo程序读取hudi数据

通过mvn clean package打包后上传运行

spark-submit \
--class cn.itxs.ReadDemo \
/home/commons/spark-3.3.0-bin-hadoop3/appjars/hoodie-spark-demo-1.0.jar

DeltaStreamer

HoodieDeltaStreamer实用程序(hudi-utilities-bundle的一部分)提供了从不同源(如DFS或Kafka)中获取的方法，具有以下功能。

• 从Kafka的新事件，从Sqoop的增量导入或输出HiveIncrementalPuller或DFS文件夹下的文件。
• 支持json, avro或自定义记录类型的传入数据。
• 管理检查点、回滚和恢复。
• 利用来自DFS或Confluent模式注册中心的Avro模式。
• 支持插入转换。

# 拷贝hudi-utilities-bundle_2.12-0.12.1.jar到spark的jars目录
cp /home/commons/hudi-release-0.12.1/packaging/hudi-utilities-bundle/target/hudi-utilities-bundle_2.12-0.12.1.jar jars/
# 查看帮助文档，参数非常多，可以在有需要使用的时候查阅
spark-submit --class org.apache.hudi.utilities.deltastreamer.HoodieDeltaStreamer /home/commons/spark-3.3.0-bin-hadoop3/jars/hudi-utilities-bundle_2.12-0.12.1.jar --help

该工具采用层次结构组成的属性文件，并具有提取数据、密钥生成和提供模式的可插入接口。在hudi-下提供了从kafka和dfs中摄取的示例配置

接下里以File Based Schema Provider和JsonKafkaSoiurce为示例演示如何使用

# 创建topic
bin/kafka-topics.sh --zookeeper zk1:2181,zk2:2181,zk3:2181 --create --partitions 1 --replication-factor 1 --topic data_test

然后编写demo程序持续向这个kafka的topic发送消息

# 创建一个配置文件目录
mkdir /home/commons/hudi-properties
# 拷贝示例配置文件
cp hudi-utilities/src/test/resources/delta-streamer-config/kafka-source.properties /home/commons/hudi-properties/
cp hudi-utilities/src/test/resources/delta-streamer-config/base.properties /home/commons/hudi-properties/

定义avro所需的schema文件包括source和target,创建source文件 vim source-json-schema.avsc

{
  "type" : "record",
  "name" : "Profiles",
  "fields" : [
      {
        "name" : "id",
        "type" : "long"
      }, {
        "name" : "name",
        "type" : "string"
      }, {
         "name" : "age",
         "type" : "int"
      },  {
        "name" : "partitions",
        "type" : "int"
      }
  ]
}

拷贝一份为target文件

cp source-json-schema.avsc target-json-schema.avsc

修改kafka-source.properties的配置如下

include=hdfs://hadoop2:9000/hudi-properties/base.properties
# Key fields, for kafka example
hoodie.datasource.write.recordkey.field=id
hoodie.datasource.write.partitionpath.field=partitions
# schema provider configs
#hoodie.deltastreamer.schemaprovider.registry.url=//localhost:8081/subjects/impressions-value/versions/latest
hoodie.deltastreamer.schemaprovider.source.schema.file=hdfs://hadoop2:9000/hudi-properties/source-json-schema.avsc
hoodie.deltastreamer.schemaprovider.target.schema.file=hdfs://hadoop2:9000/hudi-properties/target-json-schema.avsc
# Kafka Source
#hoodie.deltastreamer.source.kafka.topic=uber_trips
hoodie.deltastreamer.source.kafka.topic=data_test
#Kafka props
bootstrap.servers=kafka1:9092,kafka2:9092,kafka3:9092
auto.offset.reset=earliest
#schema.registry.url=//localhost:8081
group.id=mygroup

将本地hudi-properties文件夹上传到HDFS

cd ..
hdfs dfs -put hudi-properties/ /

# 运行导入命令
spark-submit \
--class org.apache.hudi.utilities.deltastreamer.HoodieDeltaStreamer \
/home/commons/spark-3.3.0-bin-hadoop3/jars/hudi-utilities-bundle_2.12-0.12.1.jar \
  --props hdfs://hadoop2:9000/hudi-properties/kafka-source.properties \
  --schemaprovider-class org.apache.hudi.utilities.schema.FilebasedSchemaProvider \
  --source-class org.apache.hudi.utilities.sources.JsonKafkaSource \
  --source-ordering-field id \
  --target-base-path hdfs://hadoop2:9000/tmp/hudi/user_test \
  --target-table user_test \
  --op BULK_INSERT \
  --table-type MERGE_ON_READ

查看hdfs目录已经有表目录和分区目录

通过spark-sql查询从kafka摄取的数据

use hudi_spark;
create table user_test using hudi
location 'hdfs://hadoop2:9000/tmp/hudi/user_test';
select * from user_test limit 10;

集成Flink

环境准备

# 解压进入flink目录，这里我就用之前flink的环境，详细可以查看之前关于flink的文章
cd /home/commons/flink-1.15.1
# 拷贝编译好的jar到flink的lib目录
cp /home/commons/hudi-release-0.12.1/packaging/hudi-flink-bundle/target/hudi-flink1.15-bundle-0.12.1.jar lib/

# 拷贝guava包，解决依赖冲突
cp /home/commons/hadoop/share/hadoop/common/lib/guava-27.0-jre.jar lib/
# 配置hadoop环境变量和启动hadoop
export HADOOP_CLASSPATH=`$HADOOP_HOME/bin/hadoop classpath`

sql-clent使用

启动

修改配置文件 vi conf/flink-conf.yaml

classloader.check-leaked-classloader: false
taskmanager.numberOfTaskSlots: 4
state.backend: rocksdb
state.checkpoints.dir: hdfs://hadoop2:9000/checkpoints/flink
state.backend.incremental: true
execution.checkpointing.interval: 5min

• local 模式

修改workers文件，也可以多配制几个（伪分布式或完全分布式），官方提供示例是4个

localhost
localhost
localhost

# 在本机上启动三个TaskManagerRunner和一个Standalone伪分布式集群
./bin/start-cluster.sh 
# 查看进程确认
jps -l

# 启动内嵌的flink sql客户端
./bin/sql-client.sh embedded
show databases;
show tables;

• yarn-session 模式

  • 解决依赖冲突问题

  # 拷贝jar到flink的lib目录
  cp /home/commons/hadoop/share/hadoop/mapreduce/hadoop-mapreduce-client-core-3.3.4.jar lib/
  

  • 启动yarn-session

  # 先停止上面启动Standalone伪分布式集群
  ./bin/stop-cluster.sh
  # 启动yarn-session分布式集群
  ./bin/yarn-session.sh --detached
  

  

  查看yarn上已经有一个Flink session集群job， ID为application_1669357770610_0015

  

  查看Flink的Web UI可用TaskSlots为0，可确认已切换为yarn管理资源非分配

  

  • 启动sql-client

  # 由于使用内嵌模式管理元数据，元数据是保存在内存中，关闭sql-client后则元数据也会消失，生产环境建议使用如Hive元数据管理方式，后面再做配置
  ./bin/sql-client.sh embedded -s yarn-session
  show databases;
  show tables;
  

插入数据

CREATE TABLE t1(
  uuid VARCHAR(20),
  name VARCHAR(10),
  age INT,
  ts TIMESTAMP(3),
  `partition` VARCHAR(20),
  PRIMARY KEY(uuid) NOT ENFORCED
)
PARTITIONED BY (`partition`)
WITH (
  'connector' = 'hudi',
  'path' = 'hdfs://hadoop1:9000/tmp/hudi_flink/t1',
  'table.type' = 'MERGE_ON_READ' -- 创建一个MERGE_ON_READ表，默认情况下是COPY_ON_WRITE表
);
-- 插入数据
INSERT INTO t1 VALUES
  ('id1','Danny',23,TIMESTAMP '2022-11-25 00:00:01','par1'),
  ('id2','Stephen',33,TIMESTAMP '2022-11-25 00:00:02','par1'),
  ('id3','Julian',53,TIMESTAMP '2022-11-25 00:00:03','par2'),
  ('id4','Fabian',31,TIMESTAMP '2022-11-25 00:00:04','par2'),
  ('id5','Sophia',18,TIMESTAMP '2022-11-25 00:00:05','par3'),
  ('id6','Emma',20,TIMESTAMP '2022-11-25 00:00:06','par3'),
  ('id7','Bob',44,TIMESTAMP '2022-11-25 00:00:07','par4'),
  ('id8','Han',56,TIMESTAMP '2022-11-25 00:00:08','par4');

查看Flink Web UI Job的信息

# 查询数据
select * from t1;

# 更新数据
INSERT INTO t1 VALUES
  ('id1','Danny',28,TIMESTAMP '2022-11-25 00:00:01','par1');
# 查询数据
select * from t1;

流式读取

-- 设置结果模式为tableau，在CLI中直接显示结果;另外还有table和changelog；changelog模式可以获取+I,-U之类动作数据；
set 'sql-client.execution.result-mode' = 'tableau';
CREATE TABLE sourceT (
  uuid varchar(20),
  name varchar(10),
  age int,
  ts timestamp(3),
  `partition` varchar(20),
  PRIMARY KEY(uuid) NOT ENFORCED
) WITH (
  'connector' = 'datagen',
  'rows-per-second' = '1'
);

CREATE TABLE t2 (
  uuid varchar(20),
  name varchar(10),
  age int,
  ts timestamp(3),
  `partition` varchar(20),
  PRIMARY KEY(uuid) NOT ENFORCED
)
WITH (
'connector' = 'hudi', 
'path' = 'hdfs://hadoop1:9000/tmp/hudi_flink/t2', 
'table.type' = 'MERGE_ON_READ',
'read.streaming.enabled' = 'true',
'read.streaming.check-interval' = '4'
);

insert into t2 select * from sourceT;
select * from t2;

Bucket索引

在0.11.0增加了一种高效、轻量级的索引类型bucket index，其为字节贡献回馈给hudi社区。

• Bucket Index是一种Hash分配方式，根据指定的索引字段，计算hash值，然后结合Bucket个数，均匀分配到具体的文件中。Bucket Index支持大数据量场景下的更新，Bucket Index也可以对数据进行分桶存储，但是对于桶数的计算是需要根据当前数据量的大小进行评估的，如果后续需要re-hash的话成本也会比较高。在这里我们预计通过建立Extensible Hash Index来提高哈希索引的可扩展能力。
• 要使用此索引，请将索引类型设置为BUCKET并设置hoodie.storage.layout.partitioner.class为org.apache.hudi.table.action.commit.SparkBucketIndexPartitioner。对于 Flink，设置index.type=BUCKET.
• 该方式相比于BloomIndex在元素定位性能高很多，缺点是Bucket个数无法动态扩展。另外Bucket不适合于COW表，否则会导致写放大更严重。
• 实时入湖写入的性能要求高的场景建议采用Bucket索引。

Hudi Catalog

前面基于内容管理hudi元数据的方式每次重启sql客户端就丢掉了，Hudi Catalog则是可以持久化元数据；Hudi Catalog支持多种模式，包括dfs和hms,hudi还可以直接集群hive使用，后续再一步步演示，现在先简单看下dfs模式的Hudi Catalog，先添加启动sql文件，vim conf/sql-client-init.sql

create catalog hudi_catalog 
with(
'type' = 'hudi',
'mode' = 'dfs',
'catalog.path'='/tmp/hudi_catalog'
);
use catalog hudi_catalog;

create catalog hudi_catalog 
with(
'type' = 'hudi',
'mode' = 'hms',
'hive.conf.dir'='/etc/hive/conf'
);

创建目录并启动，建表测试

hdfs dfs -mkdir /tmp/hudi_catalog
./bin/sql-client.sh embedded -i conf/sql-client-init.sql -s yarn-session

查看hdfs的数据如下，退出客户端后重新登录客户端还可以查到上面的hudi_catalog及其库和表的数据。

本人博客网站IT小神 www.itxiaoshen.com