概述說明
# 聲明
1、方案A（Redis+Cluster）和方案B（Redis+Sentinel）都能實現Redis的高可用；
2、在多個redis實例正常啟動后，方案A進行Cluster配置，方案B進行Sentinel部署、配置等，均可實現高可用效果；
3、A方案中節點擔當了Sentinel功能，故Redis集群中不需要考慮Sentinel；
4、筆者此次實驗將A方案和B方案都基於同一個redis實例群，並且同時實現兩種方案，請讀者勿混淆概念，正常情況下，A或B選一即可！
# 說明
本次實驗采用c1、c2、c3三台虛擬機完成，每台服務器上都部署一個master、一個slave和一個sentinel。當某主節點的掛了，相應的從節點替位；當某主節點及主節點對應的從節點同時掛了，將造成數據的丟失！故生產環境都采用一主多從的集群模式！
搭建環境
# 服務器信息如下：
c1 192.168.10.11
c2 192.168.10.12
c3 192.168.10.13
# 每台需要部署redis的服務器上配置系統參數，執行以下腳本
# 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
說明：本次實驗在c1、c2、c3三台服務器上分別部署一個master和一個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
# 拷貝默認配置文件並根據需求修改，因實驗環境，故采用簡潔配置，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
# 啟動
[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服務器也進行以上操作，配置只需修改對應的ip即可！
cluster
說明：本次實驗主從關系如下：
c1 master --- c2 slave M/S 1
c2 master --- c3 slave M/S 2
c3 master --- c1 slave M/S 3
=================================================================================
說明：以下內容是編譯安裝ruby工具，以便能成功執行redis-trib.rb創建集群，；yum方式安裝ruby可能版本較低！
[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中的定義變量處加上如下紅色行，若存在相應變量的定義請注釋；
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
=================================================================================
# 創建集群，格式：主 主 主 從 從 從
# --replicas參數指定集群中每個主節點配備幾個從節點，這里設置為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.
# 查看節點信息
[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是官方推薦的高可用（HA）解決方案，本身也是一個獨立運行的進程。redis的sentinel系統用於管理多個redis服務器實例（instance）。
哨兵適用於非集群結構的redis環境，比如：redis主從環境。在redis集群中，節點擔當了哨兵的功能，所以redis集群不需要考慮sentinel。
為防止單點故障，可對sentinel進行集群化。其主要功能如下：
1）監控：sentinel不斷的檢查master和slave的活性；
2）通知：當發現redis節點故障，可通過API發出通知；
3）自動故障轉移：當一個master節點故障時，能夠從眾多slave中選舉一個作為新的master，同時其它slave節點也將自動將所追隨的master的地址改為新master的地址；
4）配置提供者：哨兵作為redis客戶端發現的權威來源：客戶端連接到哨兵請求當前可靠的master地址，若發生故障，哨兵將報告新地址。
配置
說明：本次實驗將三個哨兵分別部署在c1、c2、c3三台服務器上！
[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
# 啟動哨兵
[root@c1 ~]# /usr/local/redis/bin/redis-sentinel /etc/redis-cluster/sentinel.conf
# c2、c3服務器也進行以上操作，配置只需修改對應的端口即可！
說明：啟動哨兵后，配置文件會根據監控信息自動發生相應的變化，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
故障模擬
# 通過sentinel日志查看信息，圖中可看出192.168.10.11 7001為 master，其相應的從為192.168.10.12 7002;
# 手動停止master 192.168.10.11 7001，slave 192.168.10.12 7002升為主
[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
# 日志輸出如下：
# 登陸查看nodes信息
# /usr/local/redis/bin/redis-cli -h 192.168.10.13 -p 7001
# 將192.168.10.11 7001的redis啟動
[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 已經變為192.168.10.12 7002的slave
報錯說明
問題描述
在執行命令# /usr/local/redis/src/redis-trib.rb create --replicas 1 *** 創建集群時報錯如下：
>>> 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.
問題解決
1、將節點下aof、rdb等本地備份刪除；
2、刪除node集群配置文件，即redis.conf中cluster-config-file對應的文件；