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
4.0.0
cn.itxs
hoodie-spark-demo
1.0
UTF-8
2.12.10
2.12
3.3.0
0.12.1
3.3.4
org.scala-lang
scala-library
${scala.version}
org.apache.spark
spark-core_${scala.binary.version}
${spark.version}
provided
org.apache.spark
spark-sql_${scala.binary.version}
${spark.version}
provided
org.apache.spark
spark-hive_${scala.binary.version}
${spark.version}
provided
org.apache.hadoop
hadoop-client
${hadoop.version}
provided
org.apache.hudi
hudi-spark3.3-bundle_${scala.binary.version}
${hoodie.version}
provided
org.apache.maven.plugins
maven-compiler-plugin
3.10.1
1.8
${project.build.sourceEncoding}
org.scala-tools
maven-scala-plugin
2.15.2
compile
testCompile
org.apache.maven.plugins
maven-shade-plugin
3.2.4
package
shade
*:*
META-INF/*.SF
META-INF/*.DSA
META-INF/*.RSA
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
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。
# 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=http://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=http://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;
# 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`
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
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
# 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/
# 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
# 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;
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;
-- 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;
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。
org.apache.hudi.table.action.commit.SparkBucketIndexPartitioner
。对于 Flink,设置index.type=BUCKET.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;
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。