第1章 Hadoop HA高可用
1.1 HA概述
(1)所谓HA(High Availablity),即高可用(7 * 24小时不中断服务)。
(2)实现高可用最关键的策略是消除单点故障。HA严格来说应该分成各个组件的HA机制:HDFS的HA和YARN的HA。
(3)NameNode主要在以下两个方面影响HDFS集群
NameNode机器发生意外,如宕机,集群将无法使用,直到管理员重启。
NameNode机器需要升级,包括软件、硬件升级,此时集群也将无法使用。
HDFS HA功能通过配置多个NameNodes(Active/Standby)实现在集群中对NameNode的热备来解决上述问题。如果出现故障,如机器崩溃或机器需要升级维护,
这时可通过此种方式将NameNode很快的切换到另外一台机器。
1.2 HDFS-HA核心问题
Edits文件管理系统解决namenode数据同步问题
Zookeeper通知
ZKFC:负责盯着NN,作为NN和Zookeeper的连接者,如果NN死了,ZKFC会通知Zookeeper
Zookeeper保存数据太少,不能代替Edits文件管理系统
Hadoop2.x最多有两个nn,hadoop3.x可以有多个nn
1)怎么保证三台namenode的数据一致
(1)Fsimage:让一台nn生成数据,让其他机器nn同步。
(2)Edits:需要引进新的模块JournalNode来保证edtis的文件的数据一致性。
2)怎么让同时只有一台nn是active,其他所有是standby的
(1)手动分配
(2)自动分配
3)2nn在ha架构中并不存在,定期合并fsimage和edtis的活谁来干
由standby的nn来干。
4)如果nn真的发生了问题,怎么让其他的nn上位干活
(1)手动故障转移
(2)自动故障转移
1.3 HDFS-HA手动模式
1.3.1 环境准备
(1)修改IP
(2)修改主机名及主机名和IP地址的映射
(3)关闭防火墙
(4)ssh免密登录
(5)安装JDK,配置环境变量等
1.3.2 规划集群
1.3.3 配置HDFS-HA集群
1)官方地址:http://hadoop.apache.org/
2)在opt目录下创建一个ha文件夹
[sarah@hadoop102 ~]$ cd /opt
[sarah@hadoop102 opt]$ sudo mkdir ha
[sarah@hadoop102 opt]$ sudo chown sarah:sarah /opt/ha
3)将/opt/module/下的 hadoop-3.1.3拷贝到/opt/ha目录下(记得删除data 和 log目录)
[sarah@hadoop102 opt]$ cp -r /opt/module/hadoop-3.1.3 /opt/ha/
4)配置core-site.xml
<configuration>
<!-- 把多个NameNode的地址组装成一个集群mycluster -->
<property>
<name>fs.defaultFS</name>
<value>hdfs://mycluster</value>
</property>
<!-- 指定hadoop运行时产生文件的存储目录 -->
<property>
<name>hadoop.tmp.dir</name>
<value>/opt/ha/hadoop-3.1.3/data</value>
</property>
</configuration>
5)配置hdfs-site.xml
<configuration>
<!-- NameNode数据存储目录 -->
<property>
<name>dfs.namenode.name.dir</name>
<value>file://${hadoop.tmp.dir}/name</value>
</property>
<!-- DataNode数据存储目录 -->
<property>
<name>dfs.datanode.data.dir</name>
<value>file://${hadoop.tmp.dir}/data</value>
</property>
<!-- JournalNode数据存储目录 -->
<property>
<name>dfs.journalnode.edits.dir</name>
<value>${hadoop.tmp.dir}/jn</value>
</property>
<!-- 完全分布式集群名称 -->
<property>
<name>dfs.nameservices</name>
<value>mycluster</value>
</property>
<!-- 集群中NameNode节点都有哪些 -->
<property>
<name>dfs.ha.namenodes.mycluster</name>
<value>nn1,nn2,nn3</value>
</property>
<!-- NameNode的RPC通信地址 -->
<property>
<name>dfs.namenode.rpc-address.mycluster.nn1</name>
<value>hadoop102:8020</value>
</property>
<property>
<name>dfs.namenode.rpc-address.mycluster.nn2</name>
<value>hadoop103:8020</value>
</property>
<property>
<name>dfs.namenode.rpc-address.mycluster.nn3</name>
<value>hadoop104:8020</value>
</property>
<!-- NameNode的http通信地址 -->
<property>
<name>dfs.namenode.http-address.mycluster.nn1</name>
<value>hadoop102:9870</value>
</property>
<property>
<name>dfs.namenode.http-address.mycluster.nn2</name>
<value>hadoop103:9870</value>
</property>
<property>
<name>dfs.namenode.http-address.mycluster.nn3</name>
<value>hadoop104:9870</value>
</property>
<!-- 指定NameNode元数据在JournalNode上的存放位置 -->
<property>
<name>dfs.namenode.shared.edits.dir</name>
<value>qjournal://hadoop102:8485;hadoop103:8485;hadoop104:8485/mycluster</value>
</property>
<!-- 访问代理类:client用于确定哪个NameNode为Active -->
<property>
<name>dfs.client.failover.proxy.provider.mycluster</name>
<value>org.apache.hadoop.hdfs.server.namenode.ha.ConfiguredFailoverProxyProvider</value>
</property>
<!-- 配置隔离机制,即同一时刻只能有一台服务器对外响应 -->
<property>
<name>dfs.ha.fencing.methods</name>
<value>sshfence</value>
</property>
<!-- 使用隔离机制时需要ssh秘钥登录-->
<property>
<name>dfs.ha.fencing.ssh.private-key-files</name>
<value>/home/sarah/.ssh/id_rsa</value>
</property>
</configuration>
6)分发配置好的hadoop环境到其他节点
1.3.4 启动HDFS-HA集群
1)将HADOOP_HOME环境变量更改到HA目录(三台机器)
[sarah@hadoop102 ~]$ sudo vim /etc/profile.d/my_env.sh
将HADOOP_HOME部分改为如下
#HADOOP_HOME
export HADOOP_HOME=/opt/ha/hadoop-3.1.3
export PATH=$PATH:$HADOOP_HOME/bin
export PATH=$PATH:$HADOOP_HOME/sbin
去三台机器上source环境变量
[sarah@hadoop102 ~]$source /etc/profile
2)在各个JournalNode节点上,输入以下命令启动journalnode服务
[sarah@hadoop102 ~]$ hdfs --daemon start journalnode
[sarah@hadoop103 ~]$ hdfs --daemon start journalnode
[sarah@hadoop104 ~]$ hdfs --daemon start journalnode
3)在[nn1]上,对其进行格式化,并启动
[sarah@hadoop102 ~]$ hdfs namenode -format
[sarah@hadoop102 ~]$ hdfs --daemon start namenode
4)在[nn2]和[nn3]上,同步nn1的元数据信息
[sarah@hadoop103 ~]$ hdfs namenode -bootstrapStandby
[sarah@hadoop104 ~]$ hdfs namenode -bootstrapStandby
5)启动[nn2]和[nn3]
[sarah@hadoop103 ~]$ hdfs --daemon start namenode
[sarah@hadoop104 ~]$ hdfs --daemon start namenode
图 hadoop102(standby)
图 hadoop103(standby)
7)在所有节点上,启动datanode
[sarah@hadoop102 ~]$ hdfs --daemon start datanode
[sarah@hadoop103 ~]$ hdfs --daemon start datanode
[sarah@hadoop104 ~]$ hdfs --daemon start datanode
8)将[nn1]切换为Active
[sarah@hadoop102 ~]$ hdfs haadmin -transitionToActive nn1
9)查看是否Active
[sarah@hadoop102 ~]$ hdfs haadmin -getServiceState nn1
1.4 HDFS-HA 自动模式
1.4.1 HDFS-HA自动故障转移工作机制
自动故障转移为HDFS部署增加了两个新组件:ZooKeeper和ZKFailoverController(ZKFC)进程,如图所示。ZooKeeper是维护少量协调数据,通知客户端这些数据的改变和监视客户端故障的高可用服务。
1.4.2 HDFS-HA自动故障转移的集群规划
1.4.3 配置HDFS-HA自动故障转移
1)具体配置
(1)在hdfs-site.xml中增加
<!-- 启用nn故障自动转移 -->
<property>
<name>dfs.ha.automatic-failover.enabled</name>
<value>true</value>
</property>
(2)在core-site.xml文件中增加
<property>
<name>ha.zookeeper.quorum</name>
<value>hadoop102:2181,hadoop103:2181,hadoop104:2181</value>
</property>
(3)修改后分发配置文件
[sarah@hadoop102 etc]$ pwd
/opt/ha/hadoop-3.1.3/etc
[sarah@hadoop102 etc]$ xsync hadoop/
2)启动
(1)关闭所有HDFS服务:
[sarah@hadoop102 ~]$ stop-dfs.sh
(2)启动Zookeeper集群:
[sarah@hadoop102 ~]$ zkServer.sh start
[sarah@hadoop103 ~]$ zkServer.sh start
[sarah@hadoop104 ~]$ zkServer.sh start
(3)启动Zookeeper以后,然后再初始化HA在Zookeeper中状态:
[sarah@hadoop102 ~]$ hdfs zkfc -formatZK
(4)启动HDFS服务:
[sarah@hadoop102 ~]$ start-dfs.sh
(5)可以去zkCli.sh客户端查看Namenode选举锁节点内容:
[zk: localhost:2181(CONNECTED) 7] get -s /hadoop-ha/mycluster/ActiveStandbyElectorLock
myclusternn2 hadoop103 �>(�>
cZxid = 0x10000000b
ctime = Tue Jul 14 17:00:13 CST 2020
mZxid = 0x10000000b
mtime = Tue Jul 14 17:00:13 CST 2020
pZxid = 0x10000000b
cversion = 0
dataVersion = 0
aclVersion = 0
ephemeralOwner = 0x40000da2eb70000
dataLength = 33
numChildren = 0
3)验证
(1)将Active NameNode进程kill,查看网页端三台Namenode的状态变化
[sarah@hadoop102 ~]$ kill -9 namenode的进程id
1.4.3 常见问题1–解决NN连接不上JN的问题
自动故障转移配置好以后,然后使用start-dfs.sh群起脚本启动hdfs集群,有可能会遇到NameNode起来一会后,进程自动关闭的问题。查看NameNode日志,报错信息如下:
2020-08-17 10:11:40,658 INFO org.apache.hadoop.ipc.Client: Retrying connect to server: hadoop104/192.168.6.104:8485. Already tried 0 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS)
2020-08-17 10:11:40,659 INFO org.apache.hadoop.ipc.Client: Retrying connect to server: hadoop102/192.168.6.102:8485. Already tried 0 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS)
2020-08-17 10:11:40,659 INFO org.apache.hadoop.ipc.Client: Retrying connect to server: hadoop103/192.168.6.103:8485. Already tried 0 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS)
… …
2020-08-17 10:11:49,669 INFO org.apache.hadoop.ipc.Client: Retrying connect to server: hadoop102/192.168.6.102:8485. Already tried 9 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS)
2020-08-17 10:11:49,673 INFO org.apache.hadoop.ipc.Client: Retrying connect to server: hadoop104/192.168.6.104:8485. Already tried 9 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS)
2020-08-17 10:11:49,676 INFO org.apache.hadoop.ipc.Client: Retrying connect to server: hadoop103/192.168.6.103:8485. Already tried 9 time(s); retry policy is RetryUpToMaximumCountWithFixedSleep(maxRetries=10, sleepTime=1000 MILLISECONDS)
2020-08-17 10:11:49,678 WARN org.apache.hadoop.hdfs.server.namenode.FSEditLog: Unable to determine input streams from QJM to [192.168.6.102:8485, 192.168.6.103:8485, 192.168.6.104:8485]. Skipping.
org.apache.hadoop.hdfs.qjournal.client.QuorumException: Got too many exceptions to achieve quorum size 2/3. 3 exceptions thrown:
192.168.6.103:8485: Call From hadoop102/192.168.6.102 to hadoop103:8485 failed on connection exception: java.net.ConnectException: 拒绝连接; For more details see: http://wiki.apache.org/hadoop/ConnectionRefused
192.168.6.102:8485: Call From hadoop102/192.168.6.102 to hadoop102:8485 failed on connection exception: java.net.ConnectException: 拒绝连接; For more details see: http://wiki.apache.org/hadoop/ConnectionRefused
192.168.6.104:8485: Call From hadoop102/192.168.6.102 to hadoop104:8485 failed on connection exception: java.net.ConnectException: 拒绝连接; For more details see: http://wiki.apache.org/hadoop/ConnectionRefused
查看报错日志,可分析出报错原因是因为NameNode连接不上JournalNode,而利用jps命令查看到三台JN都已经正常启动,为什么NN还是无法正常连接到JN呢?这是因为start-dfs.sh群起脚本默认的启动顺序是先启动NN,再启动DN,然后再启动JN,并且默认的rpc连接参数是重试次数为10,每次重试的间隔是1s,也就是说启动完NN以后的10s中内**,JN还启动不起来,NN就会报错了。**
core-default.xml里面有两个参数如下:
<!-- NN连接JN重试次数,默认是10次 -->
<property>
<name>ipc.client.connect.max.retries</name>
<value>10</value>
</property>
<!-- 重试时间间隔,默认1s -->
<property>
<name>ipc.client.connect.retry.interval</name>
<value>1000</value>
</property>
解决方案:遇到上述问题后,可以稍等片刻,等JN成功启动后,手动启动下三台NN:
[sarah@hadoop102 ~]$ hdfs --daemon start namenode
[sarah@hadoop103 ~]$ hdfs --daemon start namenode
[sarah@hadoop104 ~]$ hdfs --daemon start namenode
也可以在core-site.xml里面适当调大上面的两个参数:
<!-- NN连接JN重试次数,默认是10次 -->
<property>
<name>ipc.client.connect.max.retries</name>
<value>20</value>
</property>
<!-- 重试时间间隔,默认1s -->
<property>
<name>ipc.client.connect.retry.interval</name>
<value>5000</value>
</property>
1.5 YARN-HA配置
1.5.1 YARN-HA工作机制
1)官方文档:
http://hadoop.apache.org/docs/r3.1.3/hadoop-yarn/hadoop-yarn-site/ResourceManagerHA.html
2)YARN-HA工作机制
1.5.2 配置YARN-HA集群
1)环境准备
(1)修改IP
(2)修改主机名及主机名和IP地址的映射
(3)关闭防火墙
(4)ssh免密登录
(5)安装JDK,配置环境变量等
(6)配置Zookeeper集群
2)规划集群
3)核心问题
(1)如果当前active rm挂了,其他rm怎么将其他standby rm上位
核心原理跟HDFS一样,利用了zk的临时节点。
(2)当前rm上有很多的计算程序在等待运行,其他的rm怎么将这些程序接手过来接着跑
rm会将当前的所有计算程序的状态存储在zk中,其他rm上位后会去读取,然后接着跑。
4)具体配置
(1)yarn-site.xml
<configuration>
<property>
<name>yarn.nodemanager.aux-services</name>
<value>mapreduce_shuffle</value>
</property>
<!-- 启用resourcemanager ha -->
<property>
<name>yarn.resourcemanager.ha.enabled</name>
<value>true</value>
</property>
<!-- 声明两台resourcemanager的地址 -->
<property>
<name>yarn.resourcemanager.cluster-id</name>
<value>cluster-yarn1</value>
</property>
<!--指定resourcemanager的逻辑列表-->
<property>
<name>yarn.resourcemanager.ha.rm-ids</name>
<value>rm1,rm2,rm3</value>
</property>
<!-- ========== rm1的配置 ========== -->
<!-- 指定rm1的主机名 -->
<property>
<name>yarn.resourcemanager.hostname.rm1</name>
<value>hadoop102</value>
</property>
<!-- 指定rm1的web端地址 -->
<property>
<name>yarn.resourcemanager.webapp.address.rm1</name>
<value>hadoop102:8088</value>
</property>
<!-- 指定rm1的内部通信地址 -->
<property>
<name>yarn.resourcemanager.address.rm1</name>
<value>hadoop102:8032</value>
</property>
<!-- 指定AM向rm1申请资源的地址 -->
<property>
<name>yarn.resourcemanager.scheduler.address.rm1</name>
<value>hadoop102:8030</value>
</property>
<!-- 指定供NM连接的地址 -->
<property>
<name>yarn.resourcemanager.resource-tracker.address.rm1</name>
<value>hadoop102:8031</value>
</property>
<!-- ========== rm2的配置 ========== -->
<!-- 指定rm2的主机名 -->
<property>
<name>yarn.resourcemanager.hostname.rm2</name>
<value>hadoop103</value>
</property>
<property>
<name>yarn.resourcemanager.webapp.address.rm2</name>
<value>hadoop103:8088</value>
</property>
<property>
<name>yarn.resourcemanager.address.rm2</name>
<value>hadoop103:8032</value>
</property>
<property>
<name>yarn.resourcemanager.scheduler.address.rm2</name>
<value>hadoop103:8030</value>
</property>
<property>
<name>yarn.resourcemanager.resource-tracker.address.rm2</name>
<value>hadoop103:8031</value>
</property>
<!-- ========== rm3的配置 ========== -->
<!-- 指定rm1的主机名 -->
<property>
<name>yarn.resourcemanager.hostname.rm3</name>
<value>hadoop104</value>
</property>
<!-- 指定rm1的web端地址 -->
<property>
<name>yarn.resourcemanager.webapp.address.rm3</name>
<value>hadoop104:8088</value>
</property>
<!-- 指定rm1的内部通信地址 -->
<property>
<name>yarn.resourcemanager.address.rm3</name>
<value>hadoop104:8032</value>
</property>
<!-- 指定AM向rm1申请资源的地址 -->
<property>
<name>yarn.resourcemanager.scheduler.address.rm3</name>
<value>hadoop104:8030</value>
</property>
<!-- 指定供NM连接的地址 -->
<property>
<name>yarn.resourcemanager.resource-tracker.address.rm3</name>
<value>hadoop104:8031</value>
</property>
<!-- 指定zookeeper集群的地址 -->
<property>
<name>yarn.resourcemanager.zk-address</name>
<value>hadoop102:2181,hadoop103:2181,hadoop104:2181</value>
</property>
<!-- 启用自动恢复 -->
<property>
<name>yarn.resourcemanager.recovery.enabled</name>
<value>true</value>
</property>
<!-- 指定resourcemanager的状态信息存储在zookeeper集群 -->
<property>
<name>yarn.resourcemanager.store.class</name> <value>org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore</value>
</property>
<!-- 环境变量的继承 -->
<property>
<name>yarn.nodemanager.env-whitelist</name>
<value>JAVA_HOME,HADOOP_COMMON_HOME,HADOOP_HDFS_HOME,HADOOP_CONF_DIR,CLASSPATH_PREPEND_DISTCACHE,HADOOP_YARN_HOME,HADOOP_MAPRED_HOME</value>
</property>
</configuration>
(2)同步更新其他节点的配置信息,分发配置文件
[sarah@hadoop102 etc]$ xsync hadoop/
4)启动YARN
(1)在有ResourceManager的节点启动。
[sarah@hadoop102 ~]$ start-yarn.sh
(2)查看服务状态
[sarah@hadoop102 ~]$ yarn rmadmin -getServiceState rm1
(3)可以去zkCli.sh客户端查看ResourceManager选举锁节点内容。
[sarah@hadoop102 ~]$ zkCli.sh
[zk: localhost:2181(CONNECTED) 16] get -s /yarn-leader-election/cluster-yarn1/ActiveStandbyElectorLock
cluster-yarn1rm1
cZxid = 0x100000022
ctime = Tue Jul 14 17:06:44 CST 2020
mZxid = 0x100000022
mtime = Tue Jul 14 17:06:44 CST 2020
pZxid = 0x100000022
cversion = 0
dataVersion = 0
aclVersion = 0
ephemeralOwner = 0x30000da33080005
dataLength = 20
numChildren = 0
(4)web端查看hadoop102:8088和hadoop103:8088的YARN的状态
1.6 HADOOP HA的最终规划
将整个ha搭建完成后,集群将形成以下模样