HA 概述

  1. 所谓HA（High Available），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。Hadoop-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. Hadoop2.0之前，在HDFS集群中NameNode存在单点故障（SPOF）。

  4. 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Active/Standby两个NameNodes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。

HDFS-HA ⼯作机制

通过双NameNode消除单点故障（Active/Standby）

HDFS-HA⼯作要点

  1. 元数据管理⽅式需要改变

  • 内存中各⾃保存⼀份元数据；

  • Edits⽇志只有Active状态的NameNode节点可以做写操作；

  • 两个NameNode都可以读取Edits；

  • 共享的Edits放在⼀个共享存储中管理（qjournal和NFS两个主流实现）；

  1. 需要⼀个状态管理功能模块
  • 实现了⼀个zkfailover，常驻在每⼀个namenode所在的节点，每⼀个zkfailover负责监控⾃⼰所在NameNode节点，利⽤zk进⾏状态标识，当需要进⾏状态切换时，由zkfailover来负责切换，切换时需要防⽌brain split现象的发⽣（集群中出现两个Active的Namenode）。

  1. 必须保证两个NameNode之间能够ssh⽆密码登录

  2. 隔离（Fence），即同⼀时刻仅仅有⼀个NameNode对外提供服务

HDFS-HA⼯作机制

配置部署HDFS-HA进⾏⾃动故障转移。⾃动故障转移为HDFS部署增加了两个新组件：ZooKeeper和ZKFailoverController（ZKFC）进程，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。HA的⾃动故障转移依赖于ZooKeeper的以下功能：

  • 故障检测

    集群中的每个NameNode在ZooKeeper中维护了⼀个临时会话，如果机器崩溃，ZooKeeper中的会话将终⽌，ZooKeeper通知另⼀个NameNode需要触发故障转移。

  • 现役NameNode选择

    ZooKeeper提供了⼀个简单的机制⽤于唯⼀的选择⼀个节点为active状态。如果⽬前现役NameNode崩溃，另⼀个节点可能从ZooKeeper获得特殊的排外锁以表明它应该成为现役NameNode。

ZKFC是⾃动故障转移中的另⼀个新组件，是ZooKeeper的客户端，也监视和管理NameNode的状态。每个运⾏NameNode的主机也运⾏了⼀个ZKFC进程，ZKFC负责：

  • 健康监测

    ZKFC使⽤⼀个健康检查命令定期地ping与之在相同主机的NameNode，只要该NameNode及时地回复健康状态，ZKFC认为该节点是健康的。如果该节点崩溃，冻结或进⼊不健康状态，健康监测器标识该节点为⾮健康的。

  • ZooKeeper会话管理

    当本地NameNode是健康的，ZKFC保持⼀个在ZooKeeper中打开的会话。如果本地NameNode处于active状态，ZKFC也保持⼀个特殊的znode锁，该锁使⽤了ZooKeeper对短暂节点的⽀持，如果会话终⽌，锁节点将⾃动删除。

  • 基于ZooKeeper的选择

    如果本地NameNode是健康的，且ZKFC发现没有其它的节点当前持有znode锁，它将为⾃⼰获取该锁。如果成功，则它已经赢得了选择，并负责运⾏故障转移进程以使它的本地NameNode为Active。故障转移进程与前⾯描述的⼿动故障转移相似，⾸先如果必要保护之前的现役NameNode，然后本地NameNode转换为Active状态。


HDFS-HA集群配置

HDFS-HA官方文档：https://hadoop.apache.org/docs/stable/hadoop-project-dist/hadoop-hdfs/HDFSHighAvailabilityWithQJM.html

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

集群规划

Linux121 Linux122 Linux123
NameNode NameNode
JournalNode JournalNode JournalNode
DataNode DataNode DataNode
ZK ZK ZK
ResourceManager
NodeManager NodeManager NodeManager

配置HDFS-HA集群

  1. 停⽌原先HDFS集群
1
[root@Linux121 ~]# stop-dfs.sh
  1. 在所有节点，/opt/lagou/servers⽬录下创建⼀个ha⽂件夹
1
2
3
[root@Linux121 ~]# mkdir /opt/lagou/servers/ha
[root@Linux122 ~]# mkdir /opt/lagou/servers/ha
[root@Linux123 ~]# mkdir /opt/lagou/servers/ha
  1. 将/opt/lagou/servers/⽬录下的 hadoop-2.9.2拷⻉到ha⽬录下
1
[root@Linux121 ~]# cp -r /opt/lagou/servers/hadoop-2.9.2 /opt/lagou/servers/ha
  1. 删除原集群data⽬录
1
[root@Linux121 ~]# rm -rf /opt/lagou/servers/ha/hadoop-2.9.2/data
  1. 配置hdfs-site.xml
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
[root@Linux121 ~]# cd /opt/lagou/servers/ha/hadoop-2.9.2/etc/hadoop
[root@Linux121 hadoop]# vim hdfs-site.xml

<configuration>
  <property>
    <name>dfs.nameservices</name>
    <value>lagoucluster</value>
  </property>
  <property>
    <name>dfs.ha.namenodes.lagoucluster</name>
    <value>nn1,nn2</value>
  </property>
  <property>
    <name>dfs.namenode.rpc-address.lagoucluster.nn1</name>
    <value>Linux121:9000</value>
  </property>
  <property>
    <name>dfs.namenode.rpc-address.lagoucluster.nn2</name>
    <value>Linux122:9000</value>
  </property>
  <property>
    <name>dfs.namenode.http-address.lagoucluster.nn1</name>
    <value>linux121:50070</value>
  </property>
  <property>
    <name>dfs.namenode.http-address.lagoucluster.nn2</name>
    <value>linux122:50070</value>
  </property>
  <property>
    <name>dfs.namenode.shared.edits.dir</name>
    <value>qjournal://Linux121:8485;Linux122:8485;Linux123:8485/lagou</value>
  </property>
  <property>
    <name>dfs.client.failover.proxy.provider.lagoucluster</name>
    <value>org.apache.hadoop.hdfs.server.namenode.ha.ConfiguredFailoverProxyProvider</value>
  </property>
  <property>
    <name>dfs.ha.fencing.methods</name>
    <value>sshfence</value>
  </property>
  <property>
    <name>dfs.ha.fencing.ssh.private-key-files</name>
    <value>/root/.ssh/id_rsa</value>
  </property>
  <property>
    <name>dfs.journalnode.edits.dir</name>
    <value>/opt/journalnode</value>
  </property>
  <property>
    <name>dfs.ha.automatic-failover.enabled</name>
    <value>true</value>
  </property>
</configuration>
  1. 配置core-site.xml
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
[root@Linux121 hadoop]# vim core-site.xml

<configuration>
  <property>
    <name>fs.defaultFS</name>
    <value>hdfs://lagoucluster</value>
  </property>
  <property>
    <name>hadoop.tmp.dir</name>
    <value>/opt/lagou/servers/ha/hadoop-2.9.2/data/tmp</value>
  </property>
  <property>
    <name>ha.zookeeper.quorum</name>
    <value>Linux121:2181,Linux122:2181,Linux123:2181</value>
  </property>
</configuration>
  1. 拷⻉配置好的hadoop环境到其他节点
1
rsync-script /opt/lagou/servers/ha/hadoop-2.9.2/

启动HDFS-HA集群

  1. 启动zookeeper集群,并查看状态
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
[root@Linux123 ~]# zk.sh start
start zookeeper server...
ZooKeeper JMX enabled by default
Using config: /opt/lagou/servers/zookeeper-3.4.14/bin/../conf/zoo.cfg
Starting zookeeper ... STARTED
ZooKeeper JMX enabled by default
Using config: /opt/lagou/servers/zookeeper-3.4.14/bin/../conf/zoo.cfg
Starting zookeeper ... STARTED
ZooKeeper JMX enabled by default
Using config: /opt/lagou/servers/zookeeper-3.4.14/bin/../conf/zoo.cfg
Starting zookeeper ... STARTED

[root@Linux123 ~]# zk.sh status
start zookeeper server...
ZooKeeper JMX enabled by default
Using config: /opt/lagou/servers/zookeeper-3.4.14/bin/../conf/zoo.cfg
Mode: follower
ZooKeeper JMX enabled by default
Using config: /opt/lagou/servers/zookeeper-3.4.14/bin/../conf/zoo.cfg
Mode: leader
ZooKeeper JMX enabled by default
Using config: /opt/lagou/servers/zookeeper-3.4.14/bin/../conf/zoo.cfg
Mode: follower
  1. 在各个JournalNode节点上，输⼊以下命令启动journalnode服务(去往HA安装⽬录，不要使⽤环境变量中命令)
1
2
3
4
5
6
7
8
[root@Linux121 hadoop]# /opt/lagou/servers/ha/hadoop-2.9.2/sbin/hadoop-daemon.sh start journalnode
starting journalnode, logging to /opt/lagou/servers/ha/hadoop-2.9.2/logs/hadoop-root-journalnode-Linux121.out

[root@Linux122 ~]# /opt/lagou/servers/ha/hadoop-2.9.2/sbin/hadoop-daemon.sh start journalnode
starting journalnode, logging to /opt/lagou/servers/ha/hadoop-2.9.2/logs/hadoop-root-journalnode-Linux122.out

[root@Linux123 ~]# /opt/lagou/servers/ha/hadoop-2.9.2/sbin/hadoop-daemon.sh start journalnode
starting journalnode, logging to /opt/lagou/servers/ha/hadoop-2.9.2/logs/hadoop-root-journalnode-Linux123.out
  1. 在[nn1]上，对其进⾏格式化，并启动
1
2
3
[root@Linux121 hadoop]# /opt/lagou/servers/ha/hadoop-2.9.2/bin/hdfs namenode -format

[root@Linux121 hadoop]# /opt/lagou/servers/ha/hadoop-2.9.2/sbin/hadoop-daemon.sh start namenode
  1. 在[nn2]上，同步nn1的元数据信息
1
[root@Linux122 ~]# /opt/lagou/servers/ha/hadoop-2.9.2/bin/hdfs namenode -bootstrapStandby
  1. 在[nn1]上初始化zkfc
1
[root@Linux121 hadoop]# /opt/lagou/servers/ha/hadoop-2.9.2/bin/hdfs zkfc -formatZK
  1. 在[nn1]上，启动集群
1
2
3
4
5
6
7
8
9
10
11
12
13
14
[root@Linux121 hadoop]# /opt/lagou/servers/ha/hadoop-2.9.2/sbin/start-dfs.sh
Starting namenodes on [Linux121 Linux122]
Linux122: starting namenode, logging to /opt/lagou/servers/ha/hadoop-2.9.2/logs/hadoop-root-namenode-Linux122.out
Linux121: namenode running as process 33584. Stop it first.
Linux123: starting datanode, logging to /opt/lagou/servers/ha/hadoop-2.9.2/logs/hadoop-root-datanode-Linux123.out
Linux122: starting datanode, logging to /opt/lagou/servers/ha/hadoop-2.9.2/logs/hadoop-root-datanode-Linux122.out
Linux121: starting datanode, logging to /opt/lagou/servers/ha/hadoop-2.9.2/logs/hadoop-root-datanode-Linux121.out
Starting journal nodes [Linux121 Linux122 Linux123]
Linux121: journalnode running as process 30749. Stop it first.
Linux123: journalnode running as process 30055. Stop it first.
Linux122: journalnode running as process 30072. Stop it first.
Starting ZK Failover Controllers on NN hosts [Linux121 Linux122]
Linux122: starting zkfc, logging to /opt/lagou/servers/ha/hadoop-2.9.2/logs/hadoop-root-zkfc-Linux122.out
Linux121: starting zkfc, logging to /opt/lagou/servers/ha/hadoop-2.9.2/logs/hadoop-root-zkfc-Linux121.out
  1. 验证

访问http://linux121:50070/和http://linux122:50070/,如下图:



将Active NameNode进程kill,查看将 Active 和 Standby NameNode是否变化




重新启动原先Active NameNode,查看变化

1
2
[root@Linux121 hadoop]# /opt/lagou/servers/ha/hadoop-2.9.2/sbin/hadoop-daemon.sh start namenode
starting namenode, logging to /opt/lagou/servers/ha/hadoop-2.9.2/logs/hadoop-root-namenode-Linux121.out


YARN-HA配置

YARN-HA⼯作机制

YARN-HA官方文档：https://hadoop.apache.org/docs/stable/hadoop-yarn/hadoop-yarn-site/ResourceManagerHA.html


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

规划集群

Linux121 Linux122 Linux123
NameNode NameNode
JournalNode JournalNode JournalNode
DataNode DataNode DataNode
ZK ZK ZK
ResourceManager
NodeManager NodeManager NodeManager

配置 YARN-HA 集群

  1. 停⽌原先 YARN 集群
1
[root@Linux123 ~]# stop-yarn.sh
  1. 配置yarn-site.xml
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
[root@Linux121 ~]# cd /opt/lagou/servers/ha/hadoop-2.9.2/etc/hadoop
[root@Linux121 hadoop]# vim 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-yarn</value>
  </property>
  <property>
    <name>yarn.resourcemanager.ha.rm-ids</name>
    <value>rm1,rm2</value>
  </property>
  <property>
    <name>yarn.resourcemanager.hostname.rm1</name>
    <value>linux122</value>
  </property>
  <property>
    <name>yarn.resourcemanager.hostname.rm2</name>
    <value>linux123</value>
  </property>
  <!--指定zookeeper集群的地址-->
  <property>
    <name>yarn.resourcemanager.zk-address</name>
    <value>Linux121:2181,Linux122:2181,Linux123:2181</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>
</configuration>
  1. 同步更新其他节点的配置信息
1
[root@Linux121 hadoop]# rsync-script yarn-site.xml
  1. 启动yarn集群
1
2
3
4
5
6
[root@Linux123 ~]# /opt/lagou/servers/ha/hadoop-2.9.2/sbin/start-yarn.sh
starting yarn daemons
starting resourcemanager, logging to /opt/lagou/servers/ha/hadoop-2.9.2/logs/yarn-root-resourcemanager-Linux123.out
Linux123: starting nodemanager, logging to /opt/lagou/servers/ha/hadoop-2.9.2/logs/yarn-root-nodemanager-Linux123.out
Linux122: starting nodemanager, logging to /opt/lagou/servers/ha/hadoop-2.9.2/logs/yarn-root-nodemanager-Linux122.out
Linux121: starting nodemanager, logging to /opt/lagou/servers/ha/hadoop-2.9.2/logs/yarn-root-nodemanager-Linux121.out
  1. 访问http://linux123:8088/是否正常,如下:

  1. 启动Linux122上的resourcemanager,
1
2
[root@Linux122 ~]# /opt/lagou/servers/ha/hadoop-2.9.2/sbin/yarn-daemon.sh start resourcemanager
starting resourcemanager, logging to /opt/lagou/servers/ha/hadoop-2.9.2/logs/yarn-root-resourcemanager-Linux122.out

访问http://linux122:8088/会跳转到http://linux123:8088/,说明active的是Linux123

  1. 杀死Linux123上的resourcemanager
1
2
3
4
5
6
7
8
9
10
11
[root@Linux123 ~]# jps
64738 ResourceManager
64883 NodeManager
6595 -- process information unavailable
35926 DataNode
30055 JournalNode
26458 QuorumPeerMain
75674 Jps
6590 -- process information unavailable

[root@Linux123 ~]# kill -9 64738

访问http://linux122:8088/是正常的并且不会跳转到http://linux123:8088/,而http://linux123:8088/是无法访问的

  1. 启动Linux123上的resourcemanager
1
2
[root@Linux123 ~]# /opt/lagou/servers/ha/hadoop-2.9.2/sbin/yarn-daemon.sh start resourcemanager
starting resourcemanager, logging to /opt/lagou/servers/ha/hadoop-2.9.2/logs/yarn-root-resourcemanager-Linux123.out

访问http://linux123:8088/会跳转到http://linux122:8088/,说明active的是Linux122