三台主机 centos6
已经完成的工作：
- 防火墙已关闭
- 主机名修改完毕，ssh免密登陆配置完成
- jdk已安装
- zookeeper已经部署并运行
- hadoop已经部署并运行
版本：flink-1.8.2-bin-scala_2.11
上传或下载flink，解压缩
[root@node01 software]# tar -zxvf flink-1.8.2-bin-scala_2.11.tgz -C /bigdata/application/
配置环境变量，建立软连接
将官网hadoop的jar包放入lib目录下
编辑flink-conf.yaml
jobmanager.rpc.address：值设置成你master节点的IP地址
taskmanager.heap.mb：每个TaskManager可用的总内存
taskmanager.numberOfTaskSlots：每台机器上可用CPU的总数
parallelism.default：每个Job运行时默认的并行度
taskmanager.tmp.dirs：临时目录
jobmanager.heap.mb：每个节点的JVM能够分配的最大内存
jobmanager.rpc.port: 6123
jobmanager.web.port: 8081
1################################################################################ 2# Licensed to the Apache Software Foundation (ASF) under one 3# or more contributor license agreements. See the NOTICE file 4# distributed with this work for additional information 5# regarding copyright ownership. The ASF licenses this file 6# to you under the Apache License, Version 2.0 (the 7# "License"); you may not use this file except in compliance 8# with the License. You may obtain a copy of the License at 9# 10# http://www.apache.org/licenses/LICENSE-2.0 11# 12# Unless required by applicable law or agreed to in writing, software 13# distributed under the License is distributed on an "AS IS" BASIS, 14# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 15# See the License for the specific language governing permissions and 16# limitations under the License. 17################################################################################ 18 19 20#============================================================================== 21# Common 22#============================================================================== 23 24# The external address of the host on which the JobManager runs and can be 25# reached by the TaskManagers and any clients which want to connect. This setting 26# is only used in Standalone mode and may be overwritten on the JobManager side 27# by specifying the --host <hostname> parameter of the bin/jobmanager.sh executable. 28# In high availability mode, if you use the bin/start-cluster.sh script and setup 29# the conf/masters file, this will be taken care of automatically. Yarn/Mesos 30# automatically configure the host name based on the hostname of the node where the 31# JobManager runs. 32 33jobmanager.rpc.address: node03 34 35# The RPC port where the JobManager is reachable. 36 37jobmanager.rpc.port: 6123 38 39 40# The heap size for the JobManager JVM 41 42jobmanager.heap.size: 1024m 43 44 45# The heap size for the TaskManager JVM 46 47taskmanager.heap.size: 1024m 48 49 50# The number of task slots that each TaskManager offers. Each slot runs one parallel pipeline. 51 52taskmanager.numberOfTaskSlots: 2 53 54# The parallelism used for programs that did not specify and other parallelism. 55 56parallelism.default: 2 57 58# The default file system scheme and authority. 59# 60# By default file paths without scheme are interpreted relative to the local 61# root file system 'file:///'. Use this to override the default and interpret 62# relative paths relative to a different file system, 63# for example 'hdfs://mynamenode:12345' 64# 65fs.default-scheme: hdfs://ns/ 66 67#============================================================================== 68# High Availability 69#============================================================================== 70 71# The high-availability mode. Possible options are 'NONE' or 'zookeeper'. 72# 73high-availability: zookeeper 74 75# The path where metadata for master recovery is persisted. While ZooKeeper stores 76# the small ground truth for checkpoint and leader election, this location stores 77# the larger objects, like persisted dataflow graphs. 78# 79# Must be a durable file system that is accessible from all nodes 80# (like HDFS, S3, Ceph, nfs, ...) 81# 82high-availability.storageDir: hdfs://ns/flink/ha/ 83 84 85 86# The list of ZooKeeper quorum peers that coordinate the high-availability 87# setup. This must be a list of the form: 88# "host1:clientPort,host2:clientPort,..." (default clientPort: 2181) 89# 90high-availability.zookeeper.quorum: node01:2181,node02:2181,node03:2181 91high-availability.zookeeper.path.root: /flink 92 93# ACL options are based on https://zookeeper.apache.org/doc/r3.1.2/zookeeperProgrammers.html#sc_BuiltinACLSchemes 94# It can be either "creator" (ZOO_CREATE_ALL_ACL) or "open" (ZOO_OPEN_ACL_UNSAFE) 95# The default value is "open" and it can be changed to "creator" if ZK security is enabled 96# 97# high-availability.zookeeper.client.acl: open 98 99#============================================================================== 100# Fault tolerance and checkpointing 101#============================================================================== 102 103# The backend that will be used to store operator state checkpoints if 104# checkpointing is enabled. 105# 106# Supported backends are 'jobmanager', 'filesystem', 'rocksdb', or the 107# <class-name-of-factory>. 108# 109state.backend: filesystem 110 111# Directory for checkpoints filesystem, when using any of the default bundled 112# state backends. 113# 114state.checkpoints.dir: hdfs://ns/flink-checkpoints 115 116# Default target directory for savepoints, optional. 117# 118state.savepoints.dir: hdfs://ns/flink-checkpoints 119 120# Flag to enable/disable incremental checkpoints for backends that 121# support incremental checkpoints (like the RocksDB state backend). 122# 123# state.backend.incremental: false 124 125#============================================================================== 126# Rest & web frontend 127#============================================================================== 128 129# The port to which the REST client connects to. If rest.bind-port has 130# not been specified, then the server will bind to this port as well. 131# 132rest.port: 8081 133 134# The address to which the REST client will connect to 135# 136#rest.address: 0.0.0.0 137 138# Port range for the REST and web server to bind to. 139# 140#rest.bind-port: 8080-8090 141 142# The address that the REST & web server binds to 143# 144#rest.bind-address: 0.0.0.0 145 146# Flag to specify whether job submission is enabled from the web-based 147# runtime monitor. Uncomment to disable. 148 149web.submit.enable: true 150 151#============================================================================== 152# Advanced 153#============================================================================== 154 155# Override the directories for temporary files. If not specified, the 156# system-specific Java temporary directory (java.io.tmpdir property) is taken. 157# 158# For framework setups on Yarn or Mesos, Flink will automatically pick up the 159# containers' temp directories without any need for configuration. 160# 161# Add a delimited list for multiple directories, using the system directory 162# delimiter (colon ':' on unix) or a comma, e.g.: 163# /data1/tmp:/data2/tmp:/data3/tmp 164# 165# Note: Each directory entry is read from and written to by a different I/O 166# thread. You can include the same directory multiple times in order to create 167# multiple I/O threads against that directory. This is for example relevant for 168# high-throughput RAIDs. 169# 170# io.tmp.dirs: /tmp 171 172# Specify whether TaskManager's managed memory should be allocated when starting 173# up (true) or when memory is requested. 174# 175# We recommend to set this value to 'true' only in setups for pure batch 176# processing (DataSet API). Streaming setups currently do not use the TaskManager's 177# managed memory: The 'rocksdb' state backend uses RocksDB's own memory management, 178# while the 'memory' and 'filesystem' backends explicitly keep data as objects 179# to save on serialization cost. 180# 181# taskmanager.memory.preallocate: false 182 183# The classloading resolve order. Possible values are 'child-first' (Flink's default) 184# and 'parent-first' (Java's default). 185# 186# Child first classloading allows users to use different dependency/library 187# versions in their application than those in the classpath. Switching back 188# to 'parent-first' may help with debugging dependency issues. 189# 190# classloader.resolve-order: child-first 191 192# The amount of memory going to the network stack. These numbers usually need 193# no tuning. Adjusting them may be necessary in case of an "Insufficient number 194# of network buffers" error. The default min is 64MB, the default max is 1GB. 195# 196# taskmanager.network.memory.fraction: 0.1 197# taskmanager.network.memory.min: 64mb 198# taskmanager.network.memory.max: 1gb 199 200#============================================================================== 201# Flink Cluster Security Configuration 202#============================================================================== 203 204# Kerberos authentication for various components - Hadoop, ZooKeeper, and connectors - 205# may be enabled in four steps: 206# 1. configure the local krb5.conf file 207# 2. provide Kerberos credentials (either a keytab or a ticket cache w/ kinit) 208# 3. make the credentials available to various JAAS login contexts 209# 4. configure the connector to use JAAS/SASL 210 211# The below configure how Kerberos credentials are provided. A keytab will be used instead of 212# a ticket cache if the keytab path and principal are set. 213 214# security.kerberos.login.use-ticket-cache: true 215# security.kerberos.login.keytab: /path/to/kerberos/keytab 216# security.kerberos.login.principal: flink-user 217 218# The configuration below defines which JAAS login contexts 219 220# security.kerberos.login.contexts: Client,KafkaClient 221 222#============================================================================== 223# ZK Security Configuration 224#============================================================================== 225 226# Below configurations are applicable if ZK ensemble is configured for security 227 228# Override below configuration to provide custom ZK service name if configured 229# zookeeper.sasl.service-name: zookeeper 230 231# The configuration below must match one of the values set in "security.kerberos.login.contexts" 232# zookeeper.sasl.login-context-name: Client 233 234#============================================================================== 235# HistoryServer 236#============================================================================== 237 238# The HistoryServer is started and stopped via bin/historyserver.sh (start|stop) 239 240# Directory to upload completed jobs to. Add this directory to the list of 241# monitored directories of the HistoryServer as well (see below). 242#jobmanager.archive.fs.dir: hdfs:///completed-jobs/ 243 244# The address under which the web-based HistoryServer listens. 245#historyserver.web.address: 0.0.0.0 246 247# The port under which the web-based HistoryServer listens. 248historyserver.web.port: 8082 249 250# Comma separated list of directories to monitor for completed jobs. 251#historyserver.archive.fs.dir: hdfs:///completed-jobs/ 252 253# Interval in milliseconds for refreshing the monitored directories. 254#historyserver.archive.fs.refresh-interval: 10000 255 256yarn.application-attempts: 10
编辑master文件
1node03:8086 2node01:8086
编辑slaves文件
1node01 2node02 3node03
编辑zoo.cfg文件
1################################################################################ 2# Licensed to the Apache Software Foundation (ASF) under one 3# or more contributor license agreements. See the NOTICE file 4# distributed with this work for additional information 5# regarding copyright ownership. The ASF licenses this file 6# to you under the Apache License, Version 2.0 (the 7# "License"); you may not use this file except in compliance 8# with the License. You may obtain a copy of the License at 9# 10# http://www.apache.org/licenses/LICENSE-2.0 11# 12# Unless required by applicable law or agreed to in writing, software 13# distributed under the License is distributed on an "AS IS" BASIS, 14# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 15# See the License for the specific language governing permissions and 16# limitations under the License. 17################################################################################ 18 19# The number of milliseconds of each tick 20tickTime=2000 21 22# The number of ticks that the initial synchronization phase can take 23initLimit=10 24 25# The number of ticks that can pass between sending a request and getting an acknowledgement 26syncLimit=5 27 28# The directory where the snapshot is stored. 29# dataDir=/tmp/zookeeper 30 31# The port at which the clients will connect 32clientPort=2181 33 34# ZooKeeper quorum peers 35server.1=node01:2888:3888 36server.2=node02:2888:3888 37server.3=node03:2888:3888 38# server.2=host:peer-port:leader-port
复制到各个节点，配置环境变量，软连接
启动
bin下通过start-cluster.sh启动
访问node03:8086
Flink On Yarn模式
flink on yarn
1.第一种方式：yarn-session.sh(开辟资源)+flink run(提交任务)
启动一个一直运行的flink集群
1# 下面的命令会申请5个taskmanager，每个2G内存和2个solt，超过集群总资源将会启动失败。 2./bin/yarn-session.sh -n 5 -tm 2048 -s 2 --nm leo-flink -d-n ,--container <arg> 分配多少个yarn容器（=taskmanager的数量）
-D <arg> 动态属性
-d, --detached 独立运行
-jm,--jobManagerMemory <arg> JobManager的内存 [in MB]
-nm,--name 在YARN上为一个自定义的应用设置一个名字
-q,--query 显示yarn中可用的资源 (内存, cpu核数)
-qu,--queue <arg> 指定YARN队列.
-s,--slots <arg> 每个TaskManager使用的slots(vcore)数量
-tm,--taskManagerMemory <arg> 每个TaskManager的内存 [in MB]
-z,--zookeeperNamespace <arg> 针对HA模式在zookeeper上创建NameSpace
请注意：
请注意：client必须要设置YARN_CONF_DIR或者HADOOP_CONF_DIR环境变量，通过这个环境变量来读取YARN和HDFS的配置信息，否则启动会失败。
经实验发现，其实如果配置的有HADOOP_HOME环境变量的话也是可以的（只是会出现警告）。HADOOP_HOME ，YARN_CONF_DIR，HADOOP_CONF_DIR 只要配置的有任何一个即可。运行结果如图：
yarn-flink
浏览器中访问 http://node4:45559
yarn-flink
yarn web-ui中
yarn-flink
部署长期运行的flink on yarn实例后，在flink web上看到的TaskManager以及Slots都为0。只有在提交任务的时候，才会依据分配资源给对应的任务执行。</p>
提交Job到长期运行的flink on yarn实例上：
./bin/flink run ./examples/batch/WordCount.jar -input hdfs://leo/test/test.txt -output hdfs://leo/flink-word-count通过web ui可以看到已经运行完成的任务：
task
2.第二种方式：flink run -m yarn-cluster(开辟资源+提交任务)
./bin/flink run -m yarn-cluster -yn 2 -yjm 1024 -ytm 1024 ./examples/batch/WordCount.jar -input hdfs://leo/test/test.txt -output hdfs://leo/test/flink-word-count2.txtyarn web ui上查看刚刚提交的任务已经执行成功
task
作者：NikolasNull
链接：https://www.jianshu.com/p/4dc0a980e7e9
来源：简书
著作权归作者所有。商业转载请联系作者获得授权，非商业转载请注明出处。
1[root@node03 bin]# start-cluster.sh 2Starting HA cluster with 2 masters. 3ssh: connect to host node03 port 22: No buffer space available 4Starting standalonesession daemon on host node01. 5Starting taskexecutor daemon on host node01. 6Starting taskexecutor daemon on host node02. 7ssh: connect to host node03 port 22: No buffer space available 8[root@node03 bin]# start-cluster.sh 9Starting HA cluster with 2 masters. 10ssh: connect to host node03 port 22: No buffer space available 11[INFO] 1 instance(s) of standalonesession are already running on node01. 12Starting standalonesession daemon on host node01. 13[INFO] 1 instance(s) of taskexecutor are already running on node01. 14Starting taskexecutor daemon on host node01. 15[INFO] 1 instance(s) of taskexecutor are already running on node02. 16Starting taskexecutor daemon on host node02. 17ssh: connect to host node03 port 22: No buffer space available 18[root@node03 bin]# echo 512 > /proc/sys/net/ipv4/neigh/default/gc_thresh1 19[root@node03 bin]# echo 2048 > /proc/sys/net/ipv4/neigh/default/gc_thresh2 20[root@node03 bin]# echo 4096 > /proc/sys/net/ipv4/neigh/default/gc_thresh3
ping 或者ssh 发生connect: No buffer space available 错误
如果遇到这种情况，一般说明你的本地服务器的arp表缓存太大，而服务器内核设定的回收条数太小，一直被回收造成的。
可以用一下命令扩大arp表可以缓存的记录条数：
1echo 512 > /proc/sys/net/ipv4/neigh/default/gc_thresh1 2echo 2048 > /proc/sys/net/ipv4/neigh/default/gc_thresh2 3echo 4096 > /proc/sys/net/ipv4/neigh/default/gc_thresh3这三个值缺省是128，512，1024，我用arp -an |wc -l 看到自己服务器的arp缓存表竟然有300多条记录，修改完成后马上就好了，最后记得把
这三条写入/etc/rc.local 文件中，每次重启都写入下，不然机器重启就又被还原至缺省值了。