赞
踩
一、Flink 专栏
Flink 专栏系统介绍某一知识点,并辅以具体的示例进行说明。
1、Flink 部署系列
本部分介绍Flink的部署、配置相关基础内容。
2、Flink基础系列
本部分介绍Flink 的基础部分,比如术语、架构、编程模型、编程指南、基本的datastream api用法、四大基石等内容。
3、Flik Table API和SQL基础系列
本部分介绍Flink Table Api和SQL的基本用法,比如Table API和SQL创建库、表用法、查询、窗口函数、catalog等等内容。
4、Flik Table API和SQL提高与应用系列
本部分是table api 和sql的应用部分,和实际的生产应用联系更为密切,以及有一定开发难度的内容。
5、Flink 监控系列
本部分和实际的运维、监控工作相关。
二、Flink 示例专栏
Flink 示例专栏是 Flink 专栏的辅助说明,一般不会介绍知识点的信息,更多的是提供一个一个可以具体使用的示例。本专栏不再分目录,通过链接即可看出介绍的内容。
两专栏的所有文章入口点击:Flink 系列文章汇总索引
本文详细的介绍了flink的Standalone独立集群模式和Standalone HA集群模式的部署、提交任务与验证,同时介绍了Flink on yarn的两种运行模式。
本文依赖环境是hadoop集群可用、zookeeper集群环境可用以及环境是免密登录的。
本文部分图片来源于互联网。
本文分为3个部分,即Standalone独立集群模式部署及验证、Standalone HA模式部署及验证以及Flink on yarn的2种任务提交方式。
Flink支持多种安装模式
在配置系统之前,请确保在每个节点上安装有以下软件:
服务器(Master):server1(服务器名称), JobManager(服务)
服务器(Slave):server2、server3、server4(服务器名称),TaskManager(服务)
以下操作是在server1上完成的。
更多配置参考:https://nightlies.apache.org/flink/flink-docs-release-1.12/zh/deployment/config.html
vim /usr/local/flink-1.13.5/conf/flink-conf.yaml
jobmanager.rpc.address: server1
#每台机器的可用 CPU 数
taskmanager.numberOfTaskSlots: 3
#每个 TaskManager 的可用内存值
taskmanager.memory.process.size: 4096m
web.submit.enable: true
#配置项来定义 Flink 允许在每个节点上分配的最大内存值,单位是 MB,如果不设置则使用默认值
jobmanager.memory.process.size 和 taskmanager.memory.process.size
#历史服务器(flink的historyserver)
jobmanager.archive.fs.dir: hdfs://server1:8020/flink/completed-jobs/
historyserver.web.address: server1
historyserver.web.port: 8082
historyserver.archive.fs.dir: hdfs://server1:8020/flink/completed-jobs/
vim /usr/local/flink-1.13.5/conf/masters
# 添加如下内容
server1:8081
vim /usr/local/flink-1.13.5/conf/workers
#添加如下内容
server2
server3
server4
cd /usr/local/flink-1.13.5
scp -r /usr/local/flink-1.13.5 server2:$PWD
scp -r /usr/local/flink-1.13.5 server3:$PWD
scp -r /usr/local/flink-1.13.5 server4:$PWD
#如果没有权限,则进行授权
chown -R alanchan:root /usr/local/flink-1.13.5
由于Flink没有集成hdfs,在配置历史服务时启动会出现如下异常
Caused by: org.apache.flink.core.fs.UnsupportedFileSystemSchemeException:
Could not find a file system implementation for scheme 'hdfs'. The scheme is not directly supported by Flink and no Hadoop file system to support this scheme could be loaded. For a full list of supported file systems, please see
Caused by: org.apache.flink.core.fs.UnsupportedFileSystemSchemeException: Hadoop is not in the classpath/dependencies.
官网给出的说明
解决办法:
export HADOOP_CONF_DIR=/usr/local/bigdata/hadoop-3.1.4/etc/hadoop
或
export HADOOP_CONF_DIR=${HADOOP_HOME}/etc/hadoop
#shell命令,用于获取配置的Hadoop类路径
export HADOOP_CLASSPATH=`hadoop classpath`
source /etc/profile
start-cluster.sh
historyserver.sh start
#1、启动flink集群
start-cluster.sh
stop-cluster.sh
#或者单独启动
jobmanager.sh ((start|start-foreground) cluster)|stop|stop-all
taskmanager.sh start|start-foreground|stop|stop-all
[alanchan@server1 bin]$ start-cluster.sh
Starting cluster.
Starting standalonesession daemon on host server1.
Starting taskexecutor daemon on host server2.
Starting taskexecutor daemon on host server3.
Starting taskexecutor daemon on host server4.
[alanchan@server1 bin]$ stop-cluster.sh
Stopping taskexecutor daemon (pid: 28258) on host server2.
Stopping taskexecutor daemon (pid: 26309) on host server3.
Stopping taskexecutor daemon (pid: 27911) on host server4.
Stopping standalonesession daemon (pid: 12782) on host server1.
#2、启动历史服务
historyserver.sh start
#控制台显示日志
historyserver.sh start-foreground
historyserver.sh stop
flink web:http://server1:8081/#/overview
历史服务:http://server1:8082/
提交作业与本地集群部署一致。
[alanchan@server1 bin]$ flink run ../examples/streaming/WordCount.jar
Executing WordCount example with default input data set.
Use --input to specify file input.
Printing result to stdout. Use --output to specify output path.
Job has been submitted with JobID 0f8618fbf173d4272cb41384af382a8d
Program execution finished
Job with JobID 0f8618fbf173d4272cb41384af382a8d has finished.
Job Runtime: 643 ms
通过zookeeper来管理多个jobmanager,本示例2个jobmanager。
在配置系统之前,请确保在每个节点上安装有以下软件:
服务器(Master):server1、server2(服务器名称), JobManager(服务)
服务器(Slave):server2、server3、server4(服务器名称),TaskManager(服务)
在部署该集群前,zookeeper集群已经部署好了,其三台服务器为server1、server2和server3,其端口是2118。
如果需要了解其部署参考链接:1、zookeeper3.7.1安装与验证
在部署该集群前,hadoop集群已经部署好了,其四台服务器为server1、server2、server3和server4,其中server1是namenode、其余的是datanode,其端口是默认。
如果需要了解其部署参考链接:1、hadoop3.1.4简单介绍及部署、简单验证
以下操作是在server1上完成的,有不是该情况的会说明。
该示例是在standalone独立集群基础上部署的,只改变其需要变化的部分,未变的部分不再赘述。
更多配置参考:https://nightlies.apache.org/flink/flink-docs-release-1.12/zh/deployment/config.html
#开启HA,使用文件系统作为快照存储
state.backend: filesystem
#启用检查点,可以将快照保存到HDFS
state.checkpoints.dir:hdfs://server2:8020/flink-checkpoints
#使用zookeeper搭建高可用
high-availability: zookeeper
#存储JobManager的元数据到HDFS
high-availability.storageDir: hdfs://server2:8020/flink/ha/
#配置ZK集群地址
high-availability.zookeeper.quorum: server1:2118,server2:2118,server3:2118
vim /usr/local/flink-1.13.5/conf/masters
server1:8081
server2:8081
scp -r /usr/local/flink-1.13.5/conf/flink-conf.yaml server2:/usr/local/flink-1.13.5/conf/
scp -r /usr/local/flink-1.13.5/conf/flink-conf.yaml server3:/usr/local/flink-1.13.5/conf/
scp -r /usr/local/flink-1.13.5/conf/flink-conf.yaml server4:/usr/local/flink-1.13.5/conf/
scp -r /usr/local/flink-1.13.5/conf/masters server2:/usr/local/flink-1.13.5/conf/
scp -r /usr/local/flink-1.13.5/conf/masters server3:/usr/local/flink-1.13.5/conf/
scp -r /usr/local/flink-1.13.5/conf/masters server4:/usr/local/flink-1.13.5/conf/
登录server2操作
jobmanager.rpc.address: server2
#启动zookeeper集群,更多命令参考zookeeper相关专栏
zkServer.sh start
zkServer.sh stop
#启动hadoop集群,更多命令参考hadoop相关专栏
start-all.sh
start-cluster.sh
historyserver.sh start
[alanchan@server1 bin]$ start-cluster.sh
Starting HA cluster with 2 masters.
Starting standalonesession daemon on host server1.
Starting standalonesession daemon on host server2.
Starting taskexecutor daemon on host server2.
Starting taskexecutor daemon on host server3.
Starting taskexecutor daemon on host server4.
[alanchan@server1 bin]$ historyserver.sh start
Starting historyserver daemon on host server1.
验证启动情况
flink web server1:http://server1:8081/#/overview
flink web server2:http://server2:8081/#/overview
历史服务:http://server1:8082/#/overview
验证HA情况
关闭一个jobmanager,再提交任务看是否正常即可
根据自己部署时候的节点规划进行验证,以下仅仅是本人的环境验证结果
[alanchan@server1 bin]$ jps
#hadoop
19938 DFSZKFailoverController
20643 ResourceManager
19076 NameNode
#flink
18596 StandaloneSessionClusterEntrypoint
19435 HistoryServer
#zookeeper
14143 QuorumPeerMain
[alanchan@server1 bin]$ flink run ../examples/streaming/WordCount.jar
Executing WordCount example with default input data set.
Use --input to specify file input.
Printing result to stdout. Use --output to specify output path.
Job has been submitted with JobID 0f8618fbf173d4272cb41384af382a8d
Program execution finished
Job with JobID 0f8618fbf173d4272cb41384af382a8d has finished.
Job Runtime: 643 ms
在实际使用中,更多的使用方式是Flink On Yarn模式。
1.Client上传jar包和配置文件到HDFS集群上
2.Client向Yarn ResourceManager提交任务并申请资源
3.ResourceManager分配Container资源并启动ApplicationMaster,然后AppMaster加载Flink的Jar包和配置构建环境,启动JobManager
JobManager和ApplicationMaster运行在同一个container上。一旦他们被成功启动,AppMaster就知道JobManager的地址(AM它自己所在的机器)。
它就会为TaskManager生成一个新的Flink配置文件(他们就可以连接到JobManager)。这个配置文件也被上传到HDFS上。此外,AppMaster容器也提供了Flink的web服务接口。
YARN所分配的所有端口都是临时端口,这允许用户并行执行多个Flink。
4.ApplicationMaster向ResourceManager申请工作资源,NodeManager加载Flink的Jar包和配置构建环境并启动TaskManager
5.TaskManager启动后向JobManager发送心跳包,并等待JobManager向其分配任务
优点:不需要每次递交作业申请资源,而是使用已经申请好的资源,从而提高执行效率
缺点:作业执行完成以后,资源不会被释放,因此一直会占用系统资源
应用场景:适合作业递交比较频繁的场景,小作业比较多的场景
优点:作业运行完成,资源会立刻被释放,不会一直占用系统资源
缺点:每次递交作业都需要申请资源,会影响执行效率,因为申请资源需要消耗时间
应用场景:适合作业比较少的场景、大作业的场景
该模式下分为2步,即使用yarn-session.sh申请资源,然后 flink run提交任务。
在server1上执行
#执行命令
/usr/local/flink-1.13.5/bin/yarn-session.sh -n 2 -tm 1024 -s 1 -d
#申请2个CPU、2g内存
# -n 表示申请2个容器,就是多少个taskmanager
# -tm 表示每个TaskManager的内存大小
# -s 表示每个TaskManager的slots数量
# -d 表示以后台程序方式运行
#出现如下异常
2023-07-05 05:53:19,879 ERROR org.apache.flink.yarn.cli.FlinkYarnSessionCli [] - Error while running the Flink session.
java.lang.NoClassDefFoundError: javax/ws/rs/ext/MessageBodyReader
at java.lang.ClassLoader.defineClass1(Native Method) ~[?:1.8.0_144]
at java.lang.ClassLoader.defineClass(ClassLoader.java:763) ~[?:1.8.0_144]
at java.security.SecureClassLoader.defineClass(SecureClassLoader.java:142) ~[?:1.8.0_144]
at java.net.URLClassLoader.defineClass(URLClassLoader.java:467) ~[?:1.8.0_144]
#在flink整个集群的lib文件夹中增加javax.ws.rs-api-2.0.jar(https://repo1.maven.org/maven2/javax/ws/rs/javax.ws.rs-api/2.0/javax.ws.rs-api-2.0.jar),重启flink后再次提交yarn-session即可
yarn链接地址:http://server1:8088/cluster
需要在server1上执行。
#多执行几次看看运行情况,或者运行其他的应用也可以
/usr/local/flink-1.13.5/bin/flink run /usr/local/flink-1.13.5/examples/batch/WordCount.jar
[alanchan@server2 conf]$ /usr/local/flink-1.13.5/bin/flink run /usr/local/flink-1.13.5/examples/batch/WordCount.jar
Executing WordCount example with default input data set.
Use --input to specify file input.
Printing result to stdout. Use --output to specify output path.
Job has been submitted with JobID de776dfd06c52ebeadb257fe5825f11d
Program execution finished
Job with JobID de776dfd06c52ebeadb257fe5825f11d has finished.
Job Runtime: 827 ms
Accumulator Results:
- c6a7c8cb676ec7110cb43a08e072e0e5 (java.util.ArrayList) [170 elements]
(a,5)
(action,1)
(after,1)
(against,1)
(all,2)
(and,12)
(arms,1)
......
通过上方的ApplicationMaster可以进入Flink的管理界面
点击applicationmaster链接进入如下页面,可以看到flink提交的任务执行情况
正常的关闭yarn的任务即可,比如就该示例关闭如下
[alanchan@server1 ~]$ yarn application -kill application_1688448920799_0001
2023-07-05 06:18:10,152 INFO client.AHSProxy: Connecting to Application History server at server1/server1:10200
Killing application application_1688448920799_0001
2023-07-05 06:18:10,485 INFO impl.YarnClientImpl: Killed application application_1688448920799_0001
yarn链接:http://server1:8088/cluster
可以发现已经将该任务关闭了
该种模式不需要多步骤,仅仅一个步骤即可。
/usr/local/flink-1.13.5/bin/flink run -m yarn-cluster -yjm 2048 -ytm 2048 /usr/local/flink-1.13.5/examples/batch/WordCount.jar
# -m jobmanager的地址
# -yjm 1024 指定jobmanager的内存信息
# -ytm 1024 指定taskmanager的内存信息
[alanchan@server1 bin]$ /usr/local/flink-1.13.5/bin/flink run -m yarn-cluster -yjm 2048 -ytm 2048 /usr/local/flink-1.13.5/examples/batch/WordCount.jar
Executing WordCount example with default input data set.
Use --input to specify file input.
Printing result to stdout. Use --output to specify output path.
2023-07-05 06:24:29,505 WARN org.apache.flink.yarn.configuration.YarnLogConfigUtil [] - The configuration directory ('/usr/local/flink-1.13.5/conf') already contains a LOG4J config file.If you want to use logback, then please delete or rename the log configuration file.
2023-07-05 06:24:29,807 INFO org.apache.hadoop.yarn.client.AHSProxy [] - Connecting to Application History server at server1/server1:10200
2023-07-05 06:24:29,815 INFO org.apache.flink.yarn.YarnClusterDescriptor [] - No path for the flink jar passed. Using the location of class org.apache.flink.yarn.YarnClusterDescriptor to locate the jar
2023-07-05 06:24:29,922 WARN org.apache.flink.yarn.YarnClusterDescriptor [] - Neither the HADOOP_CONF_DIR nor the YARN_CONF_DIR environment variable is set. The Flink YARN Client needs one of these to be set to properly load the Hadoop configuration for accessing YARN.
2023-07-05 06:24:29,945 INFO org.apache.flink.yarn.YarnClusterDescriptor [] - The configured JobManager memory is 2048 MB. YARN will allocate 10752 MB to make up an integer multiple of its minimum allocation memory (10752 MB, configured via 'yarn.scheduler.minimum-allocation-mb'). The extra 8704 MB may not be used by Flink.
2023-07-05 06:24:29,946 INFO org.apache.flink.yarn.YarnClusterDescriptor [] - The configured TaskManager memory is 2048 MB. YARN will allocate 10752 MB to make up an integer multiple of its minimum allocation memory (10752 MB, configured via 'yarn.scheduler.minimum-allocation-mb'). The extra 8704 MB may not be used by Flink.
2023-07-05 06:24:29,946 INFO org.apache.flink.yarn.YarnClusterDescriptor [] - Cluster specification: ClusterSpecification{masterMemoryMB=10752, taskManagerMemoryMB=2048, slotsPerTaskManager=3}
2023-07-05 06:24:30,298 WARN org.apache.hadoop.hdfs.shortcircuit.DomainSocketFactory [] - The short-circuit local reads feature cannot be used because libhadoop cannot be loaded.
2023-07-05 06:24:35,442 INFO org.apache.flink.yarn.YarnClusterDescriptor [] - Submitting application master application_1688448920799_0002
2023-07-05 06:24:35,667 INFO org.apache.hadoop.yarn.client.api.impl.YarnClientImpl [] - Submitted application application_1688448920799_0002
2023-07-05 06:24:35,667 INFO org.apache.flink.yarn.YarnClusterDescriptor [] - Waiting for the cluster to be allocated
2023-07-05 06:24:35,669 INFO org.apache.flink.yarn.YarnClusterDescriptor [] - Deploying cluster, current state ACCEPTED
2023-07-05 06:24:41,699 INFO org.apache.flink.yarn.YarnClusterDescriptor [] - YARN application has been deployed successfully.
2023-07-05 06:24:41,700 INFO org.apache.flink.yarn.YarnClusterDescriptor [] - Found Web Interface server4:45227 of application 'application_1688448920799_0002'.
Job has been submitted with JobID 835195679cf827d88f8d35f60f5a923d
Program execution finished
Job with JobID 835195679cf827d88f8d35f60f5a923d has finished.
Job Runtime: 13118 ms
Accumulator Results:
- 1d6bef2182d20bbd9f6c36ce34c28e8e (java.util.ArrayList) [170 elements]
(a,5)
(action,1)
(after,1)
(against,1)
(all,2)
(and,12)
......
yarn链接:http://server1:8088/cluster
提交作业后,yarn任务页面运行情况,其实是和yarn运行任何作业一样,也是state状态由accept变化成run的
作业运行完成后
作业运行完成后,点击history链接,进入下面一个页面。
以上,完成了flink的2种部署方式与验证,同时介绍了on yarn的2种运行模式。
Copyright © 2003-2013 www.wpsshop.cn 版权所有,并保留所有权利。