Hudi集成Flink(Flink操作hudi表)

一、安装部署Flink 1.12

Apache Flink是一个框架和分布式处理引擎，用于对无界和有界数据流进行有状态计算。Flink被设计在所有常见的集群环境中运行，以内存执行速度和任意规模来执行计算。

 

1.准备tar包

flink-1.13.1-bin-scala_2.12.tgz

2.解压

 tar -zxvf flink-1.13.1-bin-scala_2.12.tgz

3.添加Hadoop依赖jar包，放在flink的lib目录下

flink-shaded-hadoop-2-uber-2.8.0-10.0.jar

flink-sql-connector-kafka_2.12-1.13.1.jar

hudi-flink-bundle_2.12-0.10.1.jar

hive-exec-2.3.9.jar

 

4.启动HDFS集群

hadoop-daemon.sh start namenode

hadoop-daemon.sh start datanode

5.启动flink本地集群

/flink/bin/start-cluster.sh

可看到两个进程：TaskManagerRunner、StandaloneSessionClusterEntrypoint

 

停止命令

/flink/bin/stop-cluster.sh

 

6.Flink Web UI

http://localhost:8081/#/overview

 

7.执行官方示例

读取文本文件数据，进行词频统计WordCount，将结果打印控制台

/flink/bin/flink run /fline/examples/batch/WordCount.jar

 

二、Flink集成Hudi时，本质将集成jar包：hudi-flink-bundle_2.12-0.10.1.jar，放入Flink应用CLASSPATH下即可。

Flink SQL Connector支持Hudi作为Source和Sink时，两种方式将jar包放入CLASSPATH路径：

方式一：运行Flink SQL Client命令时，通过参数【-j xx.jar】指定jar包

flink/bin/sql-client.sh embedded -j …./hudi-flink-bundle_2.12-0.10.1.jar

方式二：将jar包直接放入Flink软件安装包lib目录下【$FLINK_HOME/lib】

 

修改conf/flink-conf.yaml

jobmanager.rpc.address: localhost
jobmanager.memory.process.size: 1600m
taskmanager.memory.process.size: 1728m
taskmanager.numberOfTaskSlots: 4
 
classloader.check-leaked-classloader: false
classloader.resolve-order: parent-first
 
execution.checkpointing.interval: 3000
state.checkpoints.dir: hdfs://localhost:9000/flink-checkpoints
state.savepoints.dir: hdfs://localhost:9000/flink-savepoints
state.backend.incremental: true 

 

由于Flink需要连接HDFS文件系统，所以需要设置HADOOP_CLASSPATH环境变量，再启动集群

 

三、启动Flink SQL Cli命令行

sql-client.sh embedded shell

 

设置分析结果展示模式为：set execution.result-mode=tableau;

设置检查点间隔：set execution.checkpointing.interval=3sec;

 

四、使用

1.创建表：test_flink_hudi_mor，数据存储到hudi表中，底层HDFS存储，表类型MOR

CREATE TABLE test_flink_hudi_mor(
    uuid VARCHAR(20),
    name VARCHAR(10),
    age INT,
    ts TIMESTAMP(3),
    `partition` VARCHAR(20)
)
PARTITIONED BY (`partition`)
WITH(
    'connector' = 'hudi',
    'path' = 'hdfs://localhost:9000/hudi-warehouse/test_flink_hudi_mor',
    'write.tasks' = '1',
    'compaction.tasks' = '1',
    'table.type' = 'MERGE_ON_READ'
);

 

connector：表连接器

path：数据存储路径

write.tasks：flink往hudi写数据时，task数量

compaction.tasks：往hudi写数据时，做合并的task数量

table.type：hudi表类型

 

Flink SQL> desc test_flink_hudi_mor;
>
+-----------+--------------+------+-----+--------+-----------+
|      name |         type | null | key | extras | watermark |
+-----------+--------------+------+-----+--------+-----------+
|      uuid |  VARCHAR(20) | true |     |        |           |
|      name |  VARCHAR(10) | true |     |        |           |
|       age |          INT | true |     |        |           |
|        ts | TIMESTAMP(3) | true |     |        |           |
| partition |  VARCHAR(20) | true |     |        |           |
+-----------+--------------+------+-----+--------+-----------+
5 rows in set

 

2.插入数据

INSERT INTO test_flink_hudi_mor VALUES ('id1','FZ',29, TIMESTAMP '1993-04-09 00:00:01', 'par1' );
 
INSERT INTO test_flink_hudi_mor VALUES
  ('id1','Danny',23,TIMESTAMP '1970-01-01 00:00:01','par1'),
  ('id2','Stephen',33,TIMESTAMP '1970-01-01 00:00:02','par1'),
  ('id3','Julian',53,TIMESTAMP '1970-01-01 00:00:03','par2'),
  ('id4','Fabian',31,TIMESTAMP '1970-01-01 00:00:04','par2'),
  ('id5','Sophia',18,TIMESTAMP '1970-01-01 00:00:05','par3'),
  ('id6','Emma',20,TIMESTAMP '1970-01-01 00:00:06','par3'),
  ('id7','Bob',44,TIMESTAMP '1970-01-01 00:00:07','par4'),
  ('id8','Han',56,TIMESTAMP '1970-01-01 00:00:08','par4');

重复insert，会更新，id1的值由 VALUES ('id1','FZ',29, TIMESTAMP '1993-04-09 00:00:01', 'par1’ ) 改为  ('id1','Danny',23,TIMESTAMP '1970-01-01 00:00:01','par1’)

因为是MOR表，先入log，还未合并成parquet文件，如下图：

 

四、Streaming query

1.创建表：test_flink_hudi_mor_2, 以流的方式查询读取，映射到前面表test_flink_hudi_mor

CREATE TABLE test_flink_hudi_mor_2(
    uuid VARCHAR(20),
    name VARCHAR(10),
    age INT,
    ts TIMESTAMP(3),
    `partition` VARCHAR(20)
)
PARTITIONED BY (`partition`)
WITH(
    'connector' = 'hudi',
    'path' = 'hdfs://localhost:9000/hudi-warehouse/test_flink_hudi_mor',   
    'table.type' = 'MERGE_ON_READ',
    'read.tasks' = '1',
    'read.streaming.enabled' = 'true',
    'read.streaming.start-commit' = '20220307211200',
    'read.streaming.check-interval' = '4'
); 

 

read.streaming.enabled设置为true，表名通过streaming的方式读取表数据

read.streaming.check-interval指定了source监控新的commits的间隔为4s

table.type设置表类型为MERGE_ON_READ

 

 

2.重新开启terminal启动flink SQL CLI，重新创建表：test_flink_hudi_mor，采用批batch模式插入一条数据

CREATE TABLE test_flink_hudi_mor(
    uuid VARCHAR(20),
    name VARCHAR(10),
    age INT,
    ts TIMESTAMP(3),
    `partition` VARCHAR(20)
)
PARTITIONED BY (`partition`)
WITH(
    'connector' = 'hudi',
    'path' = 'hdfs://localhost:9000/hudi-warehouse/test_flink_hudi_mor',
    'write.tasks' = '1',
    'compaction.tasks' = '1',
    'table.type' = 'MERGE_ON_READ'
);
 
INSERT INTO test_flink_hudi_mor VALUES ('id9','FZ',29, TIMESTAMP '1993-04-09 00:00:01', 'par5' );
INSERT INTO test_flink_hudi_mor VALUES ('id10','DX',28, TIMESTAMP '1994-06-02 00:00:01', 'par5' );