概述说明
# 声明
1、方案A（Redis+Cluster）和方案B（Redis+Sentinel）都能实现Redis的高可用；
2、在多个redis实例正常启动后，方案A进行Cluster配置，方案B进行Sentinel部署、配置等，均可实现高可用效果；
3、A方案中节点担当了Sentinel功能，故Redis集群中不需要考虑Sentinel；
4、笔者此次实验将A方案和B方案都基于同一个redis实例群，并且同时实现两种方案，请读者勿混淆概念，正常情况下，A或B选一即可！
# 说明
本次实验采用c1、c2、c3三台虚拟机完成，每台服务器上都部署一个master、一个slave和一个sentinel。当某主节点的挂了，相应的从节点替位；当某主节点及主节点对应的从节点同时挂了，将造成数据的丢失！故生产环境都采用一主多从的集群模式！
搭建环境
# 服务器信息如下：
c1 192.168.10.11
c2 192.168.10.12
c3 192.168.10.13
# 每台需要部署redis的服务器上配置系统参数，执行以下脚本
# cat xitongcanshu.sh
1 #!/bin/bash 2 echo 'net.core.somaxconn=512' >> /etc/sysctl.conf 3 echo 'vm.overcommit_memory=1' >> /etc/sysctl.conf 4 echo never > /sys/kernel/mm/transparent_hugepage/enabled 5 echo 'echo never > /sys/kernel/mm/transparent_hugepage/enabled' >> /etc/rc.local 6 chmod +x /etc/rc.d/rc.local 7 sysctl -p 8 cat >> /etc/security/limits.conf << EOF 9 * soft nofile 65535 10 * hard nofile 65535 11 * soft nproc 65535 12 * hard nproc 65535 13 EOF 14 ulimit -SHn 65535 15 ulimit -n
redis
说明：本次实验在c1、c2、c3三台服务器上分别部署一个master和一个slave！
[root@c1 ~]# wget http://download.redis.io/releases/redis-4.0.9.tar.gz
[root@c1 ~]# tar xf redis-4.0.9.tar.gz -C /usr/local/
[root@c1 ~]# ln -sv /usr/local/redis-4.0.9 /usr/local/redis
[root@c1 ~]# yum -y install tcl openssl-devel zlib-devel
[root@c1 /usr/local/redis]# make && make PREFIX=/usr/local/redis-4.0.9/ install
[root@c1 /usr/local/redis]# make test
[root@c1 ~]# mkdir -pv /etc/redis-cluster/{7001,7002} /var/log/redis
# 拷贝默认配置文件并根据需求修改，因实验环境，故采用简洁配置，如下：
[root@c1 ~]# cat /etc/redis-cluster/7001/redis.conf
1 port 7001 2 bind 192.168.10.11 3 cluster-enabled yes 4 cluster-config-file /etc/redis-cluster/7001/nodes.conf 5 logfile /var/log/redis/redis_7001.log 6 cluster-node-timeout 5000 7 appendonly yes 8 daemonize yes
[root@c1 ~]# cat /etc/redis-cluster/7002/redis.conf
1 port 7002 2 bind 192.168.10.12 3 cluster-enabled yes 4 cluster-config-file /etc/redis-cluster/7002/nodes.conf 5 logfile /var/log/redis/redis_7002.log 6 cluster-node-timeout 5000 7 appendonly yes 8 daemonize yes
# 启动
[root@c1 ~]# /usr/local/redis/bin/redis-server /etc/redis-cluster/7001/redis.conf
[root@c1 ~]# /usr/local/redis/bin/redis-server /etc/redis-cluster/7002/redis.conf
# c2、c3服务器也进行以上操作，配置只需修改对应的ip即可！
cluster
说明：本次实验主从关系如下：
c1 master --- c2 slave M/S 1
c2 master --- c3 slave M/S 2
c3 master --- c1 slave M/S 3
=================================================================================
说明：以下内容是编译安装ruby工具，以便能成功执行redis-trib.rb创建集群，；yum方式安装ruby可能版本较低！
[root@c1 ~]# wget https://cache.ruby-lang.org/pub/ruby/2.5/ruby-2.5.1.tar.gz
[root@c1 ~]# wget https://rubygems.org/downloads/redis-4.0.0.gem
[root@c1 ~]# tar -xf ruby-2.5.1.tar.gz -C /usr/local/
[root@c1 /usr/local/ruby-2.5.1]# ./configure -prefix=/usr/local/ruby-2.5.1
[root@c1 /usr/local/ruby-2.5.1]# make && make install
[root@c1 ~]# ln -sv /usr/local/ruby-2.5.1/bin/gem /usr/bin/gem
[root@c1 ~]# ln -sv /usr/local/ruby-2.5.1/bin/ruby /usr/bin/ruby
# 在配置文件/usr/local/ruby-2.5.1/ext/openssl/Makefile和/usr/local/ruby-2.5.1/ext/zlib/Makefile中的定义变量处加上如下红色行，若存在相应变量的定义请注释；
srcdir = .
top_srcdir = ../..
topdir = /usr/local/ruby-2.5.1/include/ruby-2.5.0
[root@c1 /usr/local/ruby-2.5.1/ext/zlib]# ruby extconf.rb
[root@c1 /usr/local/ruby-2.5.1/ext/zlib]# make && make install
[root@c1 /usr/local/ruby-2.5.1/ext/openssl]# ruby extconf.rb
[root@c1 /usr/local/ruby-2.5.1/ext/openssl]# make && make install
[root@c1 ~]# gem install -l redis-4.0.0.gem
=================================================================================
# 创建集群，格式：主 主 主 从 从 从
# --replicas参数指定集群中每个主节点配备几个从节点，这里设置为1;
[root@c1 ~]# /usr/local/redis/src/redis-trib.rb create --replicas 1 192.168.10.11:7001 192.168.10.12:7001 192.168.10.13:7001 192.168.10.12:7002 192.168.10.13:7002 192.168.10.11:7002
>>> Creating cluster >>> Performing hash slots allocation on 6 nodes... Using 3 masters: 192.168.10.11:7001 192.168.10.12:7001 192.168.10.13:7001 Adding replica 192.168.10.12:7002 to 192.168.10.11:7001 Adding replica 192.168.10.13:7002 to 192.168.10.12:7001 Adding replica 192.168.10.11:7002 to 192.168.10.13:7001 M: 440541e2a3235205bf190336a1f37f127d18bf60 192.168.10.11:7001 slots:0-5460 (5461 slots) master M: c588a93825de6e0e6730a8bbb072684619201803 192.168.10.12:7001 slots:5461-10922 (5462 slots) master M: 9ba21cfda0fed2d9013103e934f199a247c378ef 192.168.10.13:7001 slots:10923-16383 (5461 slots) master S: f07abd56170635aaad5166bd38af9f7267834ca7 192.168.10.12:7002 replicates 440541e2a3235205bf190336a1f37f127d18bf60 S: 1aa03c91fc62ac72aeccf349d040f32ae190120b 192.168.10.13:7002 replicates c588a93825de6e0e6730a8bbb072684619201803 S: ff7e453f9ad5d2db2c7867893700fec033767bd9 192.168.10.11:7002 replicates 9ba21cfda0fed2d9013103e934f199a247c378ef Can I set the above configuration? (type 'yes' to accept): yes >>> Nodes configuration updated >>> Assign a different config epoch to each node >>> Sending CLUSTER MEET messages to join the cluster Waiting for the cluster to join.. >>> Performing Cluster Check (using node 192.168.10.11:7001) M: 440541e2a3235205bf190336a1f37f127d18bf60 192.168.10.11:7001 slots:0-5460 (5461 slots) master 1 additional replica(s) S: 1aa03c91fc62ac72aeccf349d040f32ae190120b 192.168.10.13:7002 slots: (0 slots) slave replicates c588a93825de6e0e6730a8bbb072684619201803 S: ff7e453f9ad5d2db2c7867893700fec033767bd9 192.168.10.11:7002 slots: (0 slots) slave replicates 9ba21cfda0fed2d9013103e934f199a247c378ef M: 9ba21cfda0fed2d9013103e934f199a247c378ef 192.168.10.13:7001 slots:10923-16383 (5461 slots) master 1 additional replica(s) M: c588a93825de6e0e6730a8bbb072684619201803 192.168.10.12:7001 slots:5461-10922 (5462 slots) master 1 additional replica(s) S: f07abd56170635aaad5166bd38af9f7267834ca7 192.168.10.12:7002 slots: (0 slots) slave replicates 440541e2a3235205bf190336a1f37f127d18bf60 [OK] All nodes agree about slots configuration. >>> Check for open slots... >>> Check slots coverage... [OK] All 16384 slots covered.
# 查看节点信息
[root@c1 ~]# /usr/local/redis/bin/redis-cli -h 192.168.10.13 -p 7001
192.168.10.13:7001> cluster nodes ff7e453f9ad5d2db2c7867893700fec033767bd9 192.168.10.11:7002@17002 slave 9ba21cfda0fed2d9013103e934f199a247c378ef 0 1527578162996 6 connected 1aa03c91fc62ac72aeccf349d040f32ae190120b 192.168.10.13:7002@17002 slave c588a93825de6e0e6730a8bbb072684619201803 0 1527578161483 5 connected 440541e2a3235205bf190336a1f37f127d18bf60 192.168.10.11:7001@17001 master - 0 1527578162000 1 connected 0-5460 f07abd56170635aaad5166bd38af9f7267834ca7 192.168.10.12:7002@17002 slave 440541e2a3235205bf190336a1f37f127d18bf60 0 1527578161000 4 connected c588a93825de6e0e6730a8bbb072684619201803 192.168.10.12:7001@17001 master - 0 1527578162491 2 connected 5461-10922 9ba21cfda0fed2d9013103e934f199a247c378ef 192.168.10.13:7001@17001 myself,master - 0 1527578162000 3 connected 10923-16383
sentinel
前言
Redis-Sentinel是官方推荐的高可用（HA）解决方案，本身也是一个独立运行的进程。redis的sentinel系统用于管理多个redis服务器实例（instance）。
哨兵适用于非集群结构的redis环境，比如：redis主从环境。在redis集群中，节点担当了哨兵的功能，所以redis集群不需要考虑sentinel。
为防止单点故障，可对sentinel进行集群化。其主要功能如下：
1）监控：sentinel不断的检查master和slave的活性；
2）通知：当发现redis节点故障，可通过API发出通知；
3）自动故障转移：当一个master节点故障时，能够从众多slave中选举一个作为新的master，同时其它slave节点也将自动将所追随的master的地址改为新master的地址；
4）配置提供者：哨兵作为redis客户端发现的权威来源：客户端连接到哨兵请求当前可靠的master地址，若发生故障，哨兵将报告新地址。
配置
说明：本次实验将三个哨兵分别部署在c1、c2、c3三台服务器上！
[root@c1 ~]# cp /usr/local/redis/sentinel.conf /etc/redis-cluster/
[root@c1 ~]# cat /etc/redis-cluster/sentinel.conf
protected-mode no port 27001 daemonize yes logfile "/var/log/sentinel.log" sentinel monitor mymaster1 192.168.10.11 7001 2 sentinel monitor mymaster2 192.168.10.12 7001 2 sentinel monitor mymaster3 192.168.10.13 7001 2 sentinel down-after-milliseconds mymaster1 10000 sentinel down-after-milliseconds mymaster2 10000 sentinel down-after-milliseconds mymaster3 10000 sentinel parallel-syncs mymaster1 1 sentinel parallel-syncs mymaster2 1 sentinel parallel-syncs mymaster3 1 sentinel failover-timeout mymaster1 15000 sentinel failover-timeout mymaster2 15000 sentinel failover-timeout mymaster3 15000
# 启动哨兵
[root@c1 ~]# /usr/local/redis/bin/redis-sentinel /etc/redis-cluster/sentinel.conf
# c2、c3服务器也进行以上操作，配置只需修改对应的端口即可！
说明：启动哨兵后，配置文件会根据监控信息自动发生相应的变化，如下：
[root@c1 ~]# cat /etc/redis-cluster/sentinel.conf
protected-mode no port 27001 daemonize yes logfile "/var/log/sentinel.log" sentinel myid e3733670b609b65e520b293789e4fbf10236089c sentinel monitor mymaster3 192.168.10.13 7001 2 sentinel down-after-milliseconds mymaster3 10000 sentinel failover-timeout mymaster3 15000 sentinel config-epoch mymaster3 0 sentinel leader-epoch mymaster3 0 sentinel known-slave mymaster3 192.168.10.11 7002 sentinel monitor mymaster1 192.168.10.11 7001 2 sentinel down-after-milliseconds mymaster1 10000 sentinel failover-timeout mymaster1 15000 sentinel config-epoch mymaster1 0 sentinel leader-epoch mymaster1 0 # Generated by CONFIG REWRITE dir "/etc/redis-cluster" sentinel known-slave mymaster1 192.168.10.12 7002 sentinel monitor mymaster2 192.168.10.12 7001 2 sentinel down-after-milliseconds mymaster2 10000 sentinel failover-timeout mymaster2 15000 sentinel config-epoch mymaster2 0 sentinel leader-epoch mymaster2 0 sentinel known-slave mymaster2 192.168.10.13 7002 sentinel current-epoch 0
故障模拟
# 通过sentinel日志查看信息，图中可看出192.168.10.11 7001为 master，其相应的从为192.168.10.12 7002;
# 手动停止master 192.168.10.11 7001，slave 192.168.10.12 7002升为主
[root@c1 ~]# ps -ef |grep redis
root 4243 1 0 03:05 ? 00:00:23 /usr/local/redis/bin/redis-server 192.168.10.11:7001 [cluster] root 4245 1 0 03:05 ? 00:00:23 /usr/local/redis/bin/redis-server 192.168.10.11:7002 [cluster] root 8472 1 1 03:45 ? 00:00:07 /usr/local/redis/bin/redis-sentinel *:27001 [sentinel]
[root@c1 ~]# kill 4243
# 日志输出如下：
# 登陆查看nodes信息
# /usr/local/redis/bin/redis-cli -h 192.168.10.13 -p 7001
# 将192.168.10.11 7001的redis启动
[root@c1 ~]# /usr/local/redis/bin/redis-server /etc/redis-cluster/7001/redis.conf
# sentinel刷新日志
# 再次查看nodes信息，192.168.10.11 7001 已经变为192.168.10.12 7002的slave
报错说明
问题描述
在执行命令# /usr/local/redis/src/redis-trib.rb create --replicas 1 *** 创建集群时报错如下：
>>> Creating cluster [ERR] Node 192.168.10.11:7001 is not empty. Either the node already knows other nodes (check with CLUSTER NODES) or contains some key in database 0.
问题解决
1、将节点下aof、rdb等本地备份删除；
2、删除node集群配置文件，即redis.conf中cluster-config-file对应的文件；