为了账号安全,请及时绑定邮箱和手机立即绑定

Flink Standalone Cluster 集群安装

标签:
Flink

Flink Standalone Cluster 集群安装


本文主要介绍如何将Flink以分布式模式运行在集群上(可能是异构的)。

环境准备

Flink 运行在所有类 UNIX 环境上,例如 Linux、Mac OS X 和 Cygwin(对于Windows),而且要求集群由一个master节点和一个或多个worker节点组成。在安装系统之前,确保每台机器上都已经安装了下面的软件:

  • Java 1.8.x或更高版本

  • ssh(Flink的脚本会用到sshd来管理远程组件)

如果你的集群还没有完全装好这些软件,你需要安装/升级它们。例如,在 Ubuntu Linux 上, 你可以执行下面的命令安装 ssh 和 Java :

sudo apt-get install ssh 
sudo apt-get install openjdk-8-jre

SSH免密码登录

译注:安装过Hadoop、Spark集群的用户应该对这段很熟悉,如果已经了解,可跳过。

为了能够启动/停止远程主机上的进程,master节点需要能免密登录所有worker节点。最方便的方式就是使用ssh的公钥验证了。要安装公钥验证,首先以最终会运行Flink的用户登录master节点。所有的worker节点上也必须要有同样的用户(例如:使用相同用户名的用户)。由于以前安装过ES,用的是es用户,所以本文会以 es用户为例。非常不建议使用 root 账户,这会有很多的安全问题。

当你用需要的用户登录了master节点,你就可以生成一对新的公钥/私钥。下面这段命令会在 ~/.ssh 目录下生成一对新的公钥/私钥。

ssh-keygen -b 2048 -P '' -f ~/.ssh/id_rsa

接下来,将公钥添加到用于认证的authorized_keys文件中:

cat ~/.ssh/id_rsa.pub >> ~/.ssh/authorized_keys

最后,将authorized_keys文件分发给集群中所有的worker节点,你可以重复地执行下面这段命令:

scp ~/.ssh/authorized_keys <worker>:~/.ssh/

将上面的<worker>替代成相应worker节点的IP/Hostname。完成了上述拷贝的工作,你应该就可以从master上免密登录其他机器了。

ssh <worker>

配置JAVA_HOME

Flink 需要master和worker节点都配置了JAVA_HOME环境变量。有两种方式可以配置。
一种是,你可以在conf/flink-conf.yaml中设置env.java.home配置项为Java的安装路径。
另一种是,sudo vi /etc/profile,在其中添加JAVA_HOME:

#javaexport JAVA_HOME=/usr/lib/jvm/java-8-openjdk-amd64export PATH=$JAVA_HOME/bin:$PATHexport CLASSPATH=.:$JAVA_HOME/lib:$JAVA_HOME/jre/lib:$CLASSPATH#nodeexport NODE_HOME=/usr/local/es/node-v9.11.1-linux-x64export PATH=$NODE_HOME/bin:$PATH#mavenexport MAVEN_HOME=/usr/local/software/maven-3.5.3export PATH=$MAVEN_HOME/bin:$PATH#hadoopexport HADOOP_HOME=/usr/local/software/hadoop-2.8.3export PATH=$PATH:$HADOOP_HOME/binexport PATH=$PATH:$HADOOP_HOME/sbinexport HADOOP_MAPRED_HOME=$HADOOP_HOMEexport HADOOP_COMMON_HOME=$HADOOP_HOMEexport HADOOP_HDFS_HOME=$HADOOP_HOMEexport YARN_HOME=$HADOOP_HOMEexport HADOOP_ROOT_LOGGER=INFO,consoleexport HADOOP_COMMON_LIB_NATIVE_DIR=$HADOOP_HOME/lib/nativeexport HADOOP_OPTS="-Djava.library.path=$HADOOP_HOME/lib"#hiveexport HIVE_HOME=/usr/local/software/hive-2.2.0export PATH=$HIVE_HOME/bin:$PATH

然后使环境变量生效,并验证 Java 是否安装成功

es@es1:/root$ java -version
openjdk version "1.8.0_162"OpenJDK Runtime Environment (build 1.8.0_162-8u162-b12-0ubuntu0.16.04.2-b12)
OpenJDK 64-Bit Server VM (build 25.162-b12, mixed mode)

安装Flink

进入下载页面。请选择一个与你的Hadoop版本相匹配的Flink包。如果你不打算使用Hadoop,选择任何版本都可以。
我这里下载的flink是flink-1.5.1,hadoop是hadoop-2.8.3。
在下载了最新的发布包后,拷贝到master节点上,并解压:

tar xzf flink-1.5.1/.tgzcd flink-1.5.1/

配置Flink

在解压完之后,你需要编辑conf/flink-conf.yaml配置Flink。

设置jobmanager.rpc.address配置项为你的master节点地址。另外为了明确 JVM 在每个节点上所能分配的最大内存,我们需要配置jobmanager.heap.mb和taskmanager.heap.mb,值的单位是 MB。如果对于某些worker节点,你想要分配更多的内存给Flink系统,你可以在相应节点上设置FLINK_TM_HEAP环境变量来覆盖默认的配置。
flink-conf.yaml配置如下:

es@es2:/usr/local/software/flink-1.5.1$ cat conf/flink-conf.yaml#################################################################################  Licensed to the Apache Software Foundation (ASF) under one#  or more contributor license agreements.  See the NOTICE file#  distributed with this work for additional information#  regarding copyright ownership.  The ASF licenses this file#  to you under the Apache License, Version 2.0 (the#  "License"); you may not use this file except in compliance#  with the License.  You may obtain a copy of the License at##      http://www.apache.org/licenses/LICENSE-2.0##  Unless required by applicable law or agreed to in writing, software#  distributed under the License is distributed on an "AS IS" BASIS,#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.#  See the License for the specific language governing permissions and# limitations under the License.#################################################################################==============================================================================# Common#==============================================================================# The external address of the host on which the JobManager runs and can be# reached by the TaskManagers and any clients which want to connect. This setting# is only used in Standalone mode and may be overwritten on the JobManager side# by specifying the --host <hostname> parameter of the bin/jobmanager.sh executable.# In high availability mode, if you use the bin/start-cluster.sh script and setup# the conf/masters file, this will be taken care of automatically. Yarn/Mesos# automatically configure the host name based on the hostname of the node where the# JobManager runs.#jobmanager.rpc.address: localhostjobmanager.rpc.address: es2# The RPC port where the JobManager is reachable.jobmanager.rpc.port: 6123# The heap size for the JobManager JVMjobmanager.heap.mb: 1024# The heap size for the TaskManager JVMtaskmanager.heap.mb: 1024# The number of task slots that each TaskManager offers. Each slot runs one parallel pipeline.taskmanager.numberOfTaskSlots: 1# The parallelism used for programs that did not specify and other parallelism.parallelism.default: 1# The default file system scheme and authority.## By default file paths without scheme are interpreted relative to the local# root file system 'file:///'. Use this to override the default and interpret# relative paths relative to a different file system,# for example 'hdfs://mynamenode:12345'## fs.default-scheme#==============================================================================# High Availability#==============================================================================# The high-availability mode. Possible options are 'NONE' or 'zookeeper'.## high-availability: zookeeper# The path where metadata for master recovery is persisted. While ZooKeeper stores# the small ground truth for checkpoint and leader election, this location stores# the larger objects, like persisted dataflow graphs.## Must be a durable file system that is accessible from all nodes# (like HDFS, S3, Ceph, nfs, ...)## high-availability.storageDir: hdfs:///flink/ha/# The list of ZooKeeper quorum peers that coordinate the high-availability# setup. This must be a list of the form:# "host1:clientPort,host2:clientPort,..." (default clientPort: 2181)## high-availability.zookeeper.quorum: localhost:2181# ACL options are based on https://zookeeper.apache.org/doc/r3.1.2/zookeeperProgrammers.html#sc_BuiltinACLSchemes# It can be either "creator" (ZOO_CREATE_ALL_ACL) or "open" (ZOO_OPEN_ACL_UNSAFE)# The default value is "open" and it can be changed to "creator" if ZK security is enabled## high-availability.zookeeper.client.acl: open#==============================================================================# Fault tolerance and checkpointing#==============================================================================# The backend that will be used to store operator state checkpoints if# checkpointing is enabled.## Supported backends are 'jobmanager', 'filesystem', 'rocksdb', or the# <class-name-of-factory>.## state.backend: filesystem# Directory for checkpoints filesystem, when using any of the default bundled# state backends.## state.checkpoints.dir: hdfs://namenode-host:port/flink-checkpoints# Default target directory for savepoints, optional.## state.savepoints.dir: hdfs://namenode-host:port/flink-checkpoints# Flag to enable/disable incremental checkpoints for backends that# support incremental checkpoints (like the RocksDB state backend).## state.backend.incremental: false#==============================================================================# Web Frontend#==============================================================================# The address under which the web-based runtime monitor listens.##jobmanager.web.address: 0.0.0.0# The port under which the web-based runtime monitor listens.# A value of -1 deactivates the web server.rest.port: 8081# Flag to specify whether job submission is enabled from the web-based# runtime monitor. Uncomment to disable.#jobmanager.web.submit.enable: false#==============================================================================# Advanced#==============================================================================# Override the directories for temporary files. If not specified, the# system-specific Java temporary directory (java.io.tmpdir property) is taken.## For framework setups on Yarn or Mesos, Flink will automatically pick up the# containers' temp directories without any need for configuration.## Add a delimited list for multiple directories, using the system directory# delimiter (colon ':' on unix) or a comma, e.g.:#     /data1/tmp:/data2/tmp:/data3/tmp## Note: Each directory entry is read from and written to by a different I/O# thread. You can include the same directory multiple times in order to create# multiple I/O threads against that directory. This is for example relevant for# high-throughput RAIDs.## io.tmp.dirs: /tmp# Specify whether TaskManager's managed memory should be allocated when starting# up (true) or when memory is requested.## We recommend to set this value to 'true' only in setups for pure batch# processing (DataSet API). Streaming setups currently do not use the TaskManager's# managed memory: The 'rocksdb' state backend uses RocksDB's own memory management,# while the 'memory' and 'filesystem' backends explicitly keep data as objects# to save on serialization cost.## taskmanager.memory.preallocate: false# The classloading resolve order. Possible values are 'child-first' (Flink's default)# and 'parent-first' (Java's default).## Child first classloading allows users to use different dependency/library# versions in their application than those in the classpath. Switching back# to 'parent-first' may help with debugging dependency issues.## classloader.resolve-order: child-first# The amount of memory going to the network stack. These numbers usually need# no tuning. Adjusting them may be necessary in case of an "Insufficient number# of network buffers" error. The default min is 64MB, teh default max is 1GB.## taskmanager.network.memory.fraction: 0.1# taskmanager.network.memory.min: 67108864# taskmanager.network.memory.max: 1073741824#==============================================================================# Flink Cluster Security Configuration#==============================================================================# Kerberos authentication for various components - Hadoop, ZooKeeper, and connectors -# may be enabled in four steps:# 1. configure the local krb5.conf file# 2. provide Kerberos credentials (either a keytab or a ticket cache w/ kinit)# 3. make the credentials available to various JAAS login contexts# 4. configure the connector to use JAAS/SASL# The below configure how Kerberos credentials are provided. A keytab will be used instead of# a ticket cache if the keytab path and principal are set.# security.kerberos.login.use-ticket-cache: true# security.kerberos.login.keytab: /path/to/kerberos/keytab# security.kerberos.login.principal: flink-user# The configuration below defines which JAAS login contexts# security.kerberos.login.contexts: Client,KafkaClient#==============================================================================# ZK Security Configuration#==============================================================================# Below configurations are applicable if ZK ensemble is configured for security# Override below configuration to provide custom ZK service name if configured# zookeeper.sasl.service-name: zookeeper# The configuration below must match one of the values set in "security.kerberos.login.contexts"# zookeeper.sasl.login-context-name: Client#==============================================================================# HistoryServer#==============================================================================# The HistoryServer is started and stopped via bin/historyserver.sh (start|stop)# Directory to upload completed jobs to. Add this directory to the list of# monitored directories of the HistoryServer as well (see below).#jobmanager.archive.fs.dir: hdfs:///completed-jobs/# The address under which the web-based HistoryServer listens.#historyserver.web.address: 0.0.0.0# The port under which the web-based HistoryServer listens.#historyserver.web.port: 8082# Comma separated list of directories to monitor for completed jobs.#historyserver.archive.fs.dir: hdfs:///completed-jobs/# Interval in milliseconds for refreshing the monitored directories.#historyserver.archive.fs.refresh-interval: 10000

最后,你需要提供一个集群中worker节点的列表。因此,就像配置HDFS,编辑conf/slaves文件,然后输入每个worker节点的 IP/Hostname。每一个worker结点之后都会运行一个 TaskManager。
每一条记录占一行,就像下面展示的一样:

es@es2:/usr/local/software/flink-1.5.1$ cat conf/slaves#localhostes1
es2

conf:

webp

conf

译注:conf/master文件是用来做JobManager HA的,在这里不需要配置

每一个worker节点上的 Flink 路径必须一致。你可以使用共享的 NSF 目录,或者拷贝整个 Flink 目录到各个worker节点。

cp -r /path/to/flink <worker>:/path/to/

请查阅配置页面了解更多关于Flink的配置。
特别的,这几个

  • TaskManager 总共能使用的内存大小(taskmanager.heap.mb)

  • 每一台机器上能使用的 CPU 个数(taskmanager.numberOfTaskSlots)

  • 集群中的总 CPU 个数(parallelism.default)

  • 临时目录(taskmanager.tmp.dirs)

是非常重要的配置项。

启动Flink

下面的脚本会在本地节点启动一个 JobManager,然后通过 SSH 连接所有的worker节点(slaves文件中所列的节点),并在每个节点上运行 TaskManager。现在你的 Flink 系统已经启动并运行了。跑在本地节点上的 JobManager 现在会在配置的 RPC 端口上监听并接收任务。

假定你在master节点上,并在Flink目录中:

bin/start-cluster.sh

master上启动的进程:


webp

slave上启动的进程:


webp

访问8081端口:


webp

可以看到两个taskManager都成功加入进来了。

要停止Flink,也有一个 stop-cluster.sh 脚本。

添加 JobManager/TaskManager 实例到集群中

你可以使用 bin/jobmanager.sh 和 bin/taskmanager 脚本来添加 JobManager 和 TaskManager 实例到你正在运行的集群中。

添加一个 JobManager

bin/jobmanager.sh (start cluster)|stop|stop-all

添加一个 TaskManager

bin/taskmanager.sh start|stop|stop-all

确保你是在需要启动/停止相应实例的节点上运行的这些脚本。



作者:it_zzy
链接:https://www.jianshu.com/p/7c142e48de61


点击查看更多内容
TA 点赞

若觉得本文不错,就分享一下吧!

评论

作者其他优质文章

正在加载中
  • 推荐
  • 评论
  • 收藏
  • 共同学习,写下你的评论
感谢您的支持,我会继续努力的~
扫码打赏,你说多少就多少
赞赏金额会直接到老师账户
支付方式
打开微信扫一扫,即可进行扫码打赏哦
今天注册有机会得

100积分直接送

付费专栏免费学

大额优惠券免费领

立即参与 放弃机会
意见反馈 帮助中心 APP下载
官方微信

举报

0/150
提交
取消