站長資訊網(wǎng)
Redis集群詳解
Redis有三種集群模式，分別是：
* 主從模式 * Sentinel模式 * Cluster模式
推薦（免費(fèi)）：redis
三種集群模式各有特點，關(guān)于Redis介紹可以參考這里：NoSQL（二)——Redis
Redis官網(wǎng)：https://redis.io/ ，最新版本5.0.4
主從模式
主從模式介紹
主從模式是三種模式中最簡單的，在主從復(fù)制中，數(shù)據(jù)庫分為兩類：主數(shù)據(jù)庫(master)和從數(shù)據(jù)庫(slave)。
其中主從復(fù)制有如下特點：
* 主數(shù)據(jù)庫可以進(jìn)行讀寫操作，當(dāng)讀寫操作導(dǎo)致數(shù)據(jù)變化時會自動將數(shù)據(jù)同步給從數(shù)據(jù)庫 * 從數(shù)據(jù)庫一般都是只讀的，并且接收主數(shù)據(jù)庫同步過來的數(shù)據(jù) * 一個master可以擁有多個slave，但是一個slave只能對應(yīng)一個master * slave掛了不影響其他slave的讀和master的讀和寫，重新啟動后會將數(shù)據(jù)從master同步過來 * master掛了以后，不影響slave的讀，但redis不再提供寫服務(wù)，master重啟后redis將重新對外提供寫服務(wù) * master掛了以后，不會在slave節(jié)點中重新選一個master
工作機(jī)制：
當(dāng)slave啟動后，主動向master發(fā)送SYNC命令。master接收到SYNC命令后在后臺保存快照（RDB持久化）和緩存保存快照這段時間的命令，然后將保存的快照文件和緩存的命令發(fā)送給slave。slave接收到快照文件和命令后加載快照文件和緩存的執(zhí)行命令。
復(fù)制初始化后，master每次接收到的寫命令都會同步發(fā)送給slave，保證主從數(shù)據(jù)一致性。
安全設(shè)置：
當(dāng)master節(jié)點設(shè)置密碼后，
客戶端訪問master需要密碼 啟動slave需要密碼，在配置文件中配置即可 客戶端訪問slave不需要密碼
缺點：
從上面可以看出，master節(jié)點在主從模式中唯一，若master掛掉，則redis無法對外提供寫服務(wù)。
主從模式搭建
- 環(huán)境準(zhǔn)備：
master節(jié)點 192.168.30.128 slave節(jié)點 192.168.30.129 slave節(jié)點 192.168.30.130
- 全部下載安裝：
# cd /software# wget http://download.redis.io/releases/redis-5.0.4.tar.gz# tar zxf redis-5.0.4.tar.gz && mv redis-5.0.4/ /usr/local/redis# cd /usr/local/redis && make && make install# echo $?0
- 全部配置成服務(wù)：
服務(wù)文件
# vim /usr/lib/systemd/system/redis.service[Unit]Description=Redis persistent key-value database After=network.target After=network-online.target Wants=network-online.target[Service]ExecStart=/usr/local/bin/redis-server /usr/local/redis/redis.conf --supervised systemd ExecStop=/usr/libexec/redis-shutdown Type=notify User=redis Group=redis RuntimeDirectory=redis RuntimeDirectoryMode=0755[Install]WantedBy=multi-user.target
shutdown腳本
# vim /usr/libexec/redis-shutdown#!/bin/bash## Wrapper to close properly redis and sentineltest x"$REDIS_DEBUG" != x && set -x REDIS_CLI=/usr/local/bin/redis-cli# Retrieve service nameSERVICE_NAME="$1"if [ -z "$SERVICE_NAME" ]; then SERVICE_NAME=redisfi# Get the proper config file based on service nameCONFIG_FILE="/usr/local/redis/$SERVICE_NAME.conf"# Use awk to retrieve host, port from config fileHOST=`awk '/^[[:blank:]]*bind/ { print $2 }' $CONFIG_FILE | tail -n1`PORT=`awk '/^[[:blank:]]*port/ { print $2 }' $CONFIG_FILE | tail -n1`PASS=`awk '/^[[:blank:]]*requirepass/ { print $2 }' $CONFIG_FILE | tail -n1`SOCK=`awk '/^[[:blank:]]*unixsockets/ { print $2 }' $CONFIG_FILE | tail -n1`# Just in case, use default host, portHOST=${HOST:-127.0.0.1}if [ "$SERVICE_NAME" = redis ]; then PORT=${PORT:-6379}else PORT=${PORT:-26739}fi# Setup additional parameters# e.g password-protected redis instances[ -z "$PASS" ] || ADDITIONAL_PARAMS="-a $PASS"# shutdown the service properlyif [ -e "$SOCK" ] ; then $REDIS_CLI -s $SOCK $ADDITIONAL_PARAMS shutdownelse $REDIS_CLI -h $HOST -p $PORT $ADDITIONAL_PARAMS shutdownfi
# chmod +x /usr/libexec/redis-shutdown# useradd -s /sbin/nologin redis# chown -R redis:redis /usr/local/redis# chown -R redis:redis /data/redis# yum install -y bash-completion && source /etc/profile #命令補(bǔ)全# systemctl daemon-reload# systemctl enable redis
- 修改配置：
192.168.30.128
# mkdir -p /data/redis# vim /usr/local/redis/redis.confbind 192.168.30.128 #監(jiān)聽ip，多個ip用空格分隔daemonize yes #允許后臺啟動logfile "/usr/local/redis/redis.log" #日志路徑dir /data/redis #數(shù)據(jù)庫備份文件存放目錄masterauth 123456 #slave連接master密碼，master可省略requirepass 123456 #設(shè)置master連接密碼，slave可省略appendonly yes #在/data/redis/目錄生成appendonly.aof文件，將每一次寫操作請求都追加到appendonly.aof 文件中# echo 'vm.overcommit_memory=1' >> /etc/sysctl.conf# sysctl -p
192.168.30.129
# mkdir -p /data/redis# vim /usr/local/redis/redis.confbind 192.168.30.129 daemonize yeslogfile "/usr/local/redis/redis.log"dir /data/redis replicaof 192.168.30.128 6379 masterauth 123456 requirepass 123456 appendonly yes# echo 'vm.overcommit_memory=1' >> /etc/sysctl.conf# sysctl -p
192.168.30.130
# mkdir -p /data/redis# vim /usr/local/redis/redis.confbind 192.168.30.130 daemonize yeslogfile "/usr/local/redis/redis.log"dir /data/redis replicaof 192.168.30.128 6379 masterauth 123456 requirepass 123456 appendonly yes# echo 'vm.overcommit_memory=1' >> /etc/sysctl.conf# sysctl -p
- 全部啟動redis：
# systemctl start redis
- 查看集群狀態(tài)：
# redis-cli -h 192.168.30.128 -a 123456Warning: Using a password with '-a' or '-u' option on the command line interface may not be safe. 192.168.30.128:6379> info replication# Replicationrole:master connected_slaves:2 slave0:ip=192.168.30.129,port=6379,state=online,offset=168,lag=1 slave1:ip=192.168.30.130,port=6379,state=online,offset=168,lag=1 master_replid:fb4941e02d5032ad74c6e2383211fc58963dbe90 master_replid2:0000000000000000000000000000000000000000 master_repl_offset:168 second_repl_offset:-1 repl_backlog_active:1 repl_backlog_size:1048576 repl_backlog_first_byte_offset:1 repl_backlog_histlen:168
# redis-cli -h 192.168.30.129 -a 123456 info replicationWarning: Using a password with '-a' or '-u' option on the command line interface may not be safe.# Replicationrole:slave master_host:192.168.30.128 master_port:6379 master_link_status:up master_last_io_seconds_ago:1 master_sync_in_progress:0 slave_repl_offset:196 slave_priority:100 slave_read_only:1 connected_slaves:0 master_replid:fb4941e02d5032ad74c6e2383211fc58963dbe90 master_replid2:0000000000000000000000000000000000000000 master_repl_offset:196 second_repl_offset:-1 repl_backlog_active:1 repl_backlog_size:1048576 repl_backlog_first_byte_offset:1 repl_backlog_histlen:196
- 數(shù)據(jù)演示：
192.168.30.128:6379> keys *(empty list or set)192.168.30.128:6379> set key1 100 OK 192.168.30.128:6379> set key2 lzx OK 192.168.30.128:6379> keys * 1) "key1"2) "key2"
# redis-cli -h 192.168.30.129 -a 123456Warning: Using a password with '-a' or '-u' option on the command line interface may not be safe. 192.168.30.129:6379> keys * 1) "key2"2) "key1"192.168.30.129:6379> CONFIG GET dir1) "dir"2) "/data/redis"192.168.30.129:6379> CONFIG GET dbfilename 1) "dbfilename"2) "dump.rdb"192.168.30.129:6379> get key1"100"192.168.30.129:6379> get key2"lzx"192.168.30.129:6379> set key3 aaa(error) READONLY You can't write against a read only replica.
# redis-cli -h 192.168.30.130 -a 123456Warning: Using a password with '-a' or '-u' option on the command line interface may not be safe. 192.168.30.130:6379> keys * 1) "key2"2) "key1"192.168.30.130:6379> CONFIG GET dir1) "dir"2) "/data/redis"192.168.30.130:6379> CONFIG GET dbfilename 1) "dbfilename"2) "dump.rdb"192.168.30.130:6379> get key1"100"192.168.30.130:6379> get key2"lzx"192.168.30.130:6379> set key3 aaa(error) READONLY You can't write against a read only replica.
可以看到，在master節(jié)點寫入的數(shù)據(jù)，很快就同步到slave節(jié)點上，而且在slave節(jié)點上無法寫入數(shù)據(jù)。
Sentinel模式
Sentinel模式介紹
主從模式的弊端就是不具備高可用性，當(dāng)master掛掉以后，Redis將不能再對外提供寫入操作，因此sentinel應(yīng)運(yùn)而生。
sentinel中文含義為哨兵，顧名思義，它的作用就是監(jiān)控redis集群的運(yùn)行狀況，特點如下：
* sentinel模式是建立在主從模式的基礎(chǔ)上，如果只有一個Redis節(jié)點，sentinel就沒有任何意義 * 當(dāng)master掛了以后，sentinel會在slave中選擇一個做為master，并修改它們的配置文件，其他slave的配置文件也會被修改，比如slaveof屬性會指向新的master * 當(dāng)master重新啟動后，它將不再是master而是做為slave接收新的master的同步數(shù)據(jù) * sentinel因為也是一個進(jìn)程有掛掉的可能，所以sentinel也會啟動多個形成一個sentinel集群 * 多sentinel配置的時候，sentinel之間也會自動監(jiān)控 * 當(dāng)主從模式配置密碼時，sentinel也會同步將配置信息修改到配置文件中，不需要擔(dān)心 * 一個sentinel或sentinel集群可以管理多個主從Redis，多個sentinel也可以監(jiān)控同一個redis * sentinel最好不要和Redis部署在同一臺機(jī)器，不然Redis的服務(wù)器掛了以后，sentinel也掛了
工作機(jī)制：
* 每個sentinel以每秒鐘一次的頻率向它所知的master，slave以及其他sentinel實例發(fā)送一個 PING 命令 * 如果一個實例距離最后一次有效回復(fù) PING 命令的時間超過 down-after-milliseconds 選項所指定的值， 則這個實例會被sentinel標(biāo)記為主觀下線。 * 如果一個master被標(biāo)記為主觀下線，則正在監(jiān)視這個master的所有sentinel要以每秒一次的頻率確認(rèn)master的確進(jìn)入了主觀下線狀態(tài) * 當(dāng)有足夠數(shù)量的sentinel（大于等于配置文件指定的值）在指定的時間范圍內(nèi)確認(rèn)master的確進(jìn)入了主觀下線狀態(tài)， 則master會被標(biāo)記為客觀下線 * 在一般情況下， 每個sentinel會以每 10 秒一次的頻率向它已知的所有master，slave發(fā)送 INFO 命令 * 當(dāng)master被sentinel標(biāo)記為客觀下線時，sentinel向下線的master的所有slave發(fā)送 INFO 命令的頻率會從 10 秒一次改為 1 秒一次 * 若沒有足夠數(shù)量的sentinel同意master已經(jīng)下線，master的客觀下線狀態(tài)就會被移除； 若master重新向sentinel的 PING 命令返回有效回復(fù)，master的主觀下線狀態(tài)就會被移除
當(dāng)使用sentinel模式的時候，客戶端就不要直接連接Redis，而是連接sentinel的ip和port，由sentinel來提供具體的可提供服務(wù)的Redis實現(xiàn)，這樣當(dāng)master節(jié)點掛掉以后，sentinel就會感知并將新的master節(jié)點提供給使用者。
Sentinel模式搭建
- 環(huán)境準(zhǔn)備：
master節(jié)點 192.168.30.128 sentinel端口：26379 slave節(jié)點 192.168.30.129 sentinel端口：26379 slave節(jié)點 192.168.30.130 sentinel端口：26379
- 修改配置：
前面已經(jīng)下載安裝了redis，這里省略，直接修改sentinel配置文件。
192.168.30.128
# vim /usr/local/redis/sentinel.confdaemonize yeslogfile "/usr/local/redis/sentinel.log"dir "/usr/local/redis/sentinel" #sentinel工作目錄sentinel monitor mymaster 192.168.30.128 6379 2 #判斷master失效至少需要2個sentinel同意，建議設(shè)置為n/2+1，n為sentinel個數(shù)sentinel auth-pass mymaster 123456 sentinel down-after-milliseconds mymaster 30000 #判斷master主觀下線時間，默認(rèn)30s
這里需要注意，sentinel auth-pass mymaster 123456需要配置在sentinel monitor mymaster 192.168.30.128 6379 2下面，否則啟動報錯：
# /usr/local/bin/redis-sentinel /usr/local/redis/sentinel.conf*** FATAL CONFIG FILE ERROR *** Reading the configuration file, at line 104>>> 'sentinel auth-pass mymaster 123456'No such master with specified name.
- 全部啟動sentinel：
# mkdir /usr/local/redis/sentinel && chown -R redis:redis /usr/local/redis# /usr/local/bin/redis-sentinel /usr/local/redis/sentinel.conf
- 任一主機(jī)查看日志：
# tail -f /usr/local/redis/sentinel.log21574:X 09 May 2019 15:32:04.298 # Sentinel ID is 30c417116a8edbab09708037366c4a7471beb77021574:X 09 May 2019 15:32:04.298 # +monitor master mymaster 192.168.30.128 6379 quorum 221574:X 09 May 2019 15:32:04.299 * +slave slave 192.168.30.129:6379 192.168.30.129 6379 @ mymaster 192.168.30.128 6379 21574:X 09 May 2019 15:32:04.300 * +slave slave 192.168.30.130:6379 192.168.30.130 6379 @ mymaster 192.168.30.128 6379 21574:X 09 May 2019 15:32:16.347 * +sentinel sentinel 79b8d61626afd4d059fb5a6a63393e9a1374e78f 192.168.30.129 26379 @ mymaster 192.168.30.128 6379 21574:X 09 May 2019 15:32:31.584 * +sentinel sentinel d7b429dcba792103ef0d80827dd0910bd9284d21 192.168.30.130 26379 @ mymaster 192.168.30.128 6379
Sentinel模式下的幾個事件：
· +reset-master ：主服務(wù)器已被重置。 · +slave ：一個新的從服務(wù)器已經(jīng)被 Sentinel 識別并關(guān)聯(lián)。 · +failover-state-reconf-slaves ：故障轉(zhuǎn)移狀態(tài)切換到了 reconf-slaves 狀態(tài)。 · +failover-detected ：另一個 Sentinel 開始了一次故障轉(zhuǎn)移操作，或者一個從服務(wù)器轉(zhuǎn)換成了主服務(wù)器。 · +slave-reconf-sent ：領(lǐng)頭（leader）的 Sentinel 向?qū)嵗l(fā)送了 [SLAVEOF](/commands/slaveof.html) 命令，為實例設(shè)置新的主服務(wù)器。 · +slave-reconf-inprog ：實例正在將自己設(shè)置為指定主服務(wù)器的從服務(wù)器，但相應(yīng)的同步過程仍未完成。 · +slave-reconf-done ：從服務(wù)器已經(jīng)成功完成對新主服務(wù)器的同步。 · -dup-sentinel ：對給定主服務(wù)器進(jìn)行監(jiān)視的一個或多個 Sentinel 已經(jīng)因為重復(fù)出現(xiàn)而被移除 —— 當(dāng) Sentinel 實例重啟的時候，就會出現(xiàn)這種情況。 · +sentinel ：一個監(jiān)視給定主服務(wù)器的新 Sentinel 已經(jīng)被識別并添加。 · +sdown ：給定的實例現(xiàn)在處于主觀下線狀態(tài)。 · -sdown ：給定的實例已經(jīng)不再處于主觀下線狀態(tài)。 · +odown ：給定的實例現(xiàn)在處于客觀下線狀態(tài)。 · -odown ：給定的實例已經(jīng)不再處于客觀下線狀態(tài)。 · +new-epoch ：當(dāng)前的紀(jì)元（epoch）已經(jīng)被更新。 · +try-failover ：一個新的故障遷移操作正在執(zhí)行中，等待被大多數(shù) Sentinel 選中（waiting to be elected by the majority）。 · +elected-leader ：贏得指定紀(jì)元的選舉，可以進(jìn)行故障遷移操作了。 · +failover-state-select-slave ：故障轉(zhuǎn)移操作現(xiàn)在處于 select-slave 狀態(tài) —— Sentinel 正在尋找可以升級為主服務(wù)器的從服務(wù)器。 · no-good-slave ：Sentinel 操作未能找到適合進(jìn)行升級的從服務(wù)器。Sentinel 會在一段時間之后再次嘗試尋找合適的從服務(wù)器來進(jìn)行升級，又或者直接放棄執(zhí)行故障轉(zhuǎn)移操作。 · selected-slave ：Sentinel 順利找到適合進(jìn)行升級的從服務(wù)器。 · failover-state-send-slaveof-noone ：Sentinel 正在將指定的從服務(wù)器升級為主服務(wù)器，等待升級功能完成。 · failover-end-for-timeout ：故障轉(zhuǎn)移因為超時而中止，不過最終所有從服務(wù)器都會開始復(fù)制新的主服務(wù)器（slaves will eventually be configured to replicate with the new master anyway）。 · failover-end ：故障轉(zhuǎn)移操作順利完成。所有從服務(wù)器都開始復(fù)制新的主服務(wù)器了。 · +switch-master ：配置變更，主服務(wù)器的 IP 和地址已經(jīng)改變。 這是絕大多數(shù)外部用戶都關(guān)心的信息。 · +tilt ：進(jìn)入 tilt 模式。 · -tilt ：退出 tilt 模式。
- master宕機(jī)演示：
192.168.30.128
# systemctl stop redis# tail -f /usr/local/redis/sentinel.log22428:X 09 May 2019 15:51:29.287 # +sdown master mymaster 192.168.30.128 637922428:X 09 May 2019 15:51:29.371 # +odown master mymaster 192.168.30.128 6379 #quorum 2/222428:X 09 May 2019 15:51:29.371 # +new-epoch 122428:X 09 May 2019 15:51:29.371 # +try-failover master mymaster 192.168.30.128 637922428:X 09 May 2019 15:51:29.385 # +vote-for-leader 30c417116a8edbab09708037366c4a7471beb770 122428:X 09 May 2019 15:51:29.403 # d7b429dcba792103ef0d80827dd0910bd9284d21 voted for 30c417116a8edbab09708037366c4a7471beb770 122428:X 09 May 2019 15:51:29.408 # 79b8d61626afd4d059fb5a6a63393e9a1374e78f voted for 30c417116a8edbab09708037366c4a7471beb770 122428:X 09 May 2019 15:51:29.451 # +elected-leader master mymaster 192.168.30.128 637922428:X 09 May 2019 15:51:29.451 # +failover-state-select-slave master mymaster 192.168.30.128 637922428:X 09 May 2019 15:51:29.528 # +selected-slave slave 192.168.30.129:6379 192.168.30.129 6379 @ mymaster 192.168.30.128 637922428:X 09 May 2019 15:51:29.528 * +failover-state-send-slaveof-noone slave 192.168.30.129:6379 192.168.30.129 6379 @ mymaster 192.168.30.128 6379 22428:X 09 May 2019 15:51:29.594 * +failover-state-wait-promotion slave 192.168.30.129:6379 192.168.30.129 6379 @ mymaster 192.168.30.128 6379 22428:X 09 May 2019 15:51:30.190 # +promoted-slave slave 192.168.30.129:6379 192.168.30.129 6379 @ mymaster 192.168.30.128 637922428:X 09 May 2019 15:51:30.190 # +failover-state-reconf-slaves master mymaster 192.168.30.128 637922428:X 09 May 2019 15:51:30.258 * +slave-reconf-sent slave 192.168.30.130:6379 192.168.30.130 6379 @ mymaster 192.168.30.128 6379 22428:X 09 May 2019 15:51:30.511 # -odown master mymaster 192.168.30.128 637922428:X 09 May 2019 15:51:31.233 * +slave-reconf-inprog slave 192.168.30.130:6379 192.168.30.130 6379 @ mymaster 192.168.30.128 6379 22428:X 09 May 2019 15:51:31.233 * +slave-reconf-done slave 192.168.30.130:6379 192.168.30.130 6379 @ mymaster 192.168.30.128 6379 22428:X 09 May 2019 15:51:31.297 # +failover-end master mymaster 192.168.30.128 637922428:X 09 May 2019 15:51:31.297 # +switch-master mymaster 192.168.30.128 6379 192.168.30.129 637922428:X 09 May 2019 15:51:31.298 * +slave slave 192.168.30.130:6379 192.168.30.130 6379 @ mymaster 192.168.30.129 6379 22428:X 09 May 2019 15:51:31.298 * +slave slave 192.168.30.128:6379 192.168.30.128 6379 @ mymaster 192.168.30.129 6379 22428:X 09 May 2019 15:52:31.307 # +sdown slave 192.168.30.128:6379 192.168.30.128 6379 @ mymaster 192.168.30.129 6379
從日志中可以看到，master已經(jīng)從192.168.30.128轉(zhuǎn)移到192.168.30.129上
192.168.30.129上查看集群信息
# /usr/local/bin/redis-cli -h 192.168.30.129 -p 6379 -a 123456Warning: Using a password with '-a' or '-u' option on the command line interface may not be safe. 192.168.30.129:6379> info replication# Replicationrole:master connected_slaves:1 slave0:ip=192.168.30.130,port=6379,state=online,offset=291039,lag=1 master_replid:757aff269236ed2707ba584a86a40716c1c76d74 master_replid2:47a862fc0ff20362be29096ecdcca6d432070ee9 master_repl_offset:291182 second_repl_offset:248123 repl_backlog_active:1 repl_backlog_size:1048576 repl_backlog_first_byte_offset:1 repl_backlog_histlen:291182 192.168.30.129:6379> set key4 linux OK
當(dāng)前集群中只有一個slave——192.168.30.130，master是192.168.30.129，且192.168.30.129具有寫權(quán)限。
192.168.30.130上查看redis的配置文件也可以看到replicaof 192.168.30.129 6379，這是sentinel在選舉master是做的修改。
重新把192.168.30.128上進(jìn)程啟動
# systemctl start redis# tail -f /usr/local/redis/sentinel.log22428:X 09 May 2019 15:51:31.297 # +switch-master mymaster 192.168.30.128 6379 192.168.30.129 637922428:X 09 May 2019 15:51:31.298 * +slave slave 192.168.30.130:6379 192.168.30.130 6379 @ mymaster 192.168.30.129 6379 22428:X 09 May 2019 15:51:31.298 * +slave slave 192.168.30.128:6379 192.168.30.128 6379 @ mymaster 192.168.30.129 6379 22428:X 09 May 2019 15:52:31.307 # +sdown slave 192.168.30.128:6379 192.168.30.128 6379 @ mymaster 192.168.30.129 637922428:X 09 May 2019 16:01:24.872 # -sdown slave 192.168.30.128:6379 192.168.30.128 6379 @ mymaster 192.168.30.129 6379
查看集群信息
# /usr/local/bin/redis-cli -h 192.168.30.128 -p 6379 -a 123456Warning: Using a password with '-a' or '-u' option on the command line interface may not be safe. 192.168.30.128:6379> info replication# Replicationrole:slave master_host:192.168.30.129 master_port:6379 master_link_status:up master_last_io_seconds_ago:0 master_sync_in_progress:0 slave_repl_offset:514774 slave_priority:100 slave_read_only:1 connected_slaves:0 master_replid:757aff269236ed2707ba584a86a40716c1c76d74 master_replid2:0000000000000000000000000000000000000000 master_repl_offset:514774 second_repl_offset:-1 repl_backlog_active:1 repl_backlog_size:1048576 repl_backlog_first_byte_offset:376528 repl_backlog_histlen:138247 192.168.30.128:6379> get key4"linux"192.168.30.128:6379> set key5(error) ERR wrong number of arguments for 'set' command
即使192.168.30.128重新啟動redis服務(wù)，也是作為slave加入redis集群，192.168.30.129仍然是master。
Cluster模式
Cluster模式介紹
sentinel模式基本可以滿足一般生產(chǎn)的需求，具備高可用性。但是當(dāng)數(shù)據(jù)量過大到一臺服務(wù)器存放不下的情況時，主從模式或sentinel模式就不能滿足需求了，這個時候需要對存儲的數(shù)據(jù)進(jìn)行分片，將數(shù)據(jù)存儲到多個Redis實例中。cluster模式的出現(xiàn)就是為了解決單機(jī)Redis容量有限的問題，將Redis的數(shù)據(jù)根據(jù)一定的規(guī)則分配到多臺機(jī)器。
cluster可以說是sentinel和主從模式的結(jié)合體，通過cluster可以實現(xiàn)主從和master重選功能，所以如果配置兩個副本三個分片的話，就需要六個Redis實例。因為Redis的數(shù)據(jù)是根據(jù)一定規(guī)則分配到cluster的不同機(jī)器的，當(dāng)數(shù)據(jù)量過大時，可以新增機(jī)器進(jìn)行擴(kuò)容。
使用集群，只需要將redis配置文件中的cluster-enable配置打開即可。每個集群中至少需要三個主數(shù)據(jù)庫才能正常運(yùn)行，新增節(jié)點非常方便。
cluster集群特點：
* 多個redis節(jié)點網(wǎng)絡(luò)互聯(lián)，數(shù)據(jù)共享 * 所有的節(jié)點都是一主一從（也可以是一主多從），其中從不提供服務(wù)，僅作為備用 * 不支持同時處理多個key（如MSET/MGET），因為redis需要把key均勻分布在各個節(jié)點上， 并發(fā)量很高的情況下同時創(chuàng)建key-value會降低性能并導(dǎo)致不可預(yù)測的行為 * 支持在線增加、刪除節(jié)點 * 客戶端可以連接任何一個主節(jié)點進(jìn)行讀寫
Cluster模式搭建
- 環(huán)境準(zhǔn)備：
三臺機(jī)器，分別開啟兩個redis服務(wù)（端口） 192.168.30.128 端口：7001,7002 192.168.30.129 端口：7003,7004 192.168.30.130 端口：7005,7006
- 修改配置文件：
192.168.30.128
# mkdir /usr/local/redis/cluster# cp /usr/local/redis/redis.conf /usr/local/redis/cluster/redis_7001.conf# cp /usr/local/redis/redis.conf /usr/local/redis/cluster/redis_7002.conf# chown -R redis:redis /usr/local/redis# mkdir -p /data/redis/cluster/{redis_7001,redis_7002} && chown -R redis:redis /data/redis
# vim /usr/local/redis/cluster/redis_7001.confbind 192.168.30.128 port 7001 daemonize yespidfile "/var/run/redis_7001.pid"logfile "/usr/local/redis/cluster/redis_7001.log"dir "/data/redis/cluster/redis_7001"#replicaof 192.168.30.129 6379masterauth 123456 requirepass 123456 appendonly yescluster-enabled yescluster-config-file nodes_7001.conf cluster-node-timeout 15000
# vim /usr/local/redis/cluster/redis_7002.confbind 192.168.30.128 port 7002 daemonize yespidfile "/var/run/redis_7002.pid"logfile "/usr/local/redis/cluster/redis_7002.log"dir "/data/redis/cluster/redis_7002"#replicaof 192.168.30.129 6379masterauth "123456"requirepass "123456"appendonly yescluster-enabled yescluster-config-file nodes_7002.conf cluster-node-timeout 15000
其它兩臺機(jī)器配置與192.168.30.128一致，此處省略
- 啟動redis服務(wù)：
# redis-server /usr/local/redis/cluster/redis_7001.conf# tail -f /usr/local/redis/cluster/redis_7001.log# redis-server /usr/local/redis/cluster/redis_7002.conf# tail -f /usr/local/redis/cluster/redis_7002.log
其它兩臺機(jī)器啟動與192.168.30.128一致，此處省略
- 安裝ruby并創(chuàng)建集群（低版本）：
如果redis版本比較低，則需要安裝ruby。任選一臺機(jī)器安裝ruby即可
# yum -y groupinstall "Development Tools"# yum install -y gdbm-devel libdb4-devel libffi-devel libyaml libyaml-devel ncurses-devel openssl-devel readline-devel tcl-devel# mkdir -p ~/rpmbuild/{BUILD,BUILDROOT,RPMS,SOURCES,SPECS,SRPMS}# wget http://cache.ruby-lang.org/pub/ruby/2.2/ruby-2.2.3.tar.gz -P ~/rpmbuild/SOURCES# wget http://raw.githubusercontent.com/tjinjin/automate-ruby-rpm/master/ruby22x.spec -P ~/rpmbuild/SPECS# rpmbuild -bb ~/rpmbuild/SPECS/ruby22x.spec# rpm -ivh ~/rpmbuild/RPMS/x86_64/ruby-2.2.3-1.el7.x86_64.rpm# gem install redis #目的是安裝這個，用于配置集群
# cp /usr/local/redis/src/redis-trib.rb /usr/bin/# redis-trib.rb create --replicas 1 192.168.30.128:7001 192.168.30.128:7002 192.168.30.129:7003 192.168.30.129:7004 192.168.30.130:7005 192.168.30.130:7006
- 創(chuàng)建集群：
我這里是redis5.0.4，所以不需要安裝ruby，直接創(chuàng)建集群即可
# redis-cli -a 123456 --cluster create 192.168.30.128:7001 192.168.30.128:7002 192.168.30.129:7003 192.168.30.129:7004 192.168.30.130:7005 192.168.30.130:7006 --cluster-replicas 1Warning: Using a password with '-a' or '-u' option on the command line interface may not be safe.>>> Performing hash slots allocation on 6 nodes... Master[0] -> Slots 0 - 5460 Master[1] -> Slots 5461 - 10922 Master[2] -> Slots 10923 - 16383 Adding replica 192.168.30.129:7004 to 192.168.30.128:7001 Adding replica 192.168.30.130:7006 to 192.168.30.129:7003 Adding replica 192.168.30.128:7002 to 192.168.30.130:7005 M: 80c80a3f3e33872c047a8328ad579b9bea001ad8 192.168.30.128:7001 slots:[0-5460] (5461 slots) master S: b4d3eb411a7355d4767c6c23b4df69fa183ef8bc 192.168.30.128:7002 replicates 6788453ee9a8d7f72b1d45a9093838efd0e501f1 M: 4d74ec66e898bf09006dac86d4928f9fad81f373 192.168.30.129:7003 slots:[5461-10922] (5462 slots) master S: b6331cbc986794237c83ed2d5c30777c1551546e 192.168.30.129:7004 replicates 80c80a3f3e33872c047a8328ad579b9bea001ad8 M: 6788453ee9a8d7f72b1d45a9093838efd0e501f1 192.168.30.130:7005 slots:[10923-16383] (5461 slots) master S: 277daeb8660d5273b7c3e05c263f861ed5f17b92 192.168.30.130:7006 replicates 4d74ec66e898bf09006dac86d4928f9fad81f373 Can I set the above configuration? (type 'yes' to accept): yes #輸入yes，接受上面配置>>> Nodes configuration updated>>> Assign a different config epoch to each node>>> Sending CLUSTER MEET messages to join the cluster
可以看到，
192.168.30.128:7001是master，它的slave是192.168.30.129:7004； 192.168.30.129:7003是master，它的slave是192.168.30.130:7006； 192.168.30.130:7005是master，它的slave是192.168.30.128:7002
自動生成nodes.conf文件：
# ls /data/redis/cluster/redis_7001/appendonly.aof dump.rdb nodes-7001.conf# vim /data/redis/cluster/redis_7001/nodes-7001.conf 6788453ee9a8d7f72b1d45a9093838efd0e501f1 192.168.30.130:7005@17005 master - 0 1557454406312 5 connected 10923-16383 277daeb8660d5273b7c3e05c263f861ed5f17b92 192.168.30.130:7006@17006 slave 4d74ec66e898bf09006dac86d4928f9fad81f373 0 1557454407000 6 connected b4d3eb411a7355d4767c6c23b4df69fa183ef8bc 192.168.30.128:7002@17002 slave 6788453ee9a8d7f72b1d45a9093838efd0e501f1 0 1557454408371 5 connected 80c80a3f3e33872c047a8328ad579b9bea001ad8 192.168.30.128:7001@17001 myself,master - 0 1557454406000 1 connected 0-5460 b6331cbc986794237c83ed2d5c30777c1551546e 192.168.30.129:7004@17004 slave 80c80a3f3e33872c047a8328ad579b9bea001ad8 0 1557454407366 4 connected 4d74ec66e898bf09006dac86d4928f9fad81f373 192.168.30.129:7003@17003 master - 0 1557454407000 3 connected 5461-10922 vars currentEpoch 6 lastVoteEpoch 0
集群操作
- 登錄集群：
# redis-cli -c -h 192.168.30.128 -p 7001 -a 123456 # -c，使用集群方式登錄
- 查看集群信息：
192.168.30.128:7001> CLUSTER INFO #集群狀態(tài)cluster_state:ok cluster_slots_assigned:16384 cluster_slots_ok:16384 cluster_slots_pfail:0 cluster_slots_fail:0 cluster_known_nodes:6 cluster_size:3 cluster_current_epoch:6 cluster_my_epoch:1 cluster_stats_messages_ping_sent:580 cluster_stats_messages_pong_sent:551 cluster_stats_messages_sent:1131 cluster_stats_messages_ping_received:546 cluster_stats_messages_pong_received:580 cluster_stats_messages_meet_received:5 cluster_stats_messages_received:1131
- 列出節(jié)點信息：
192.168.30.128:7001> CLUSTER NODES #列出節(jié)點信息6788453ee9a8d7f72b1d45a9093838efd0e501f1 192.168.30.130:7005@17005 master - 0 1557455176000 5 connected 10923-16383 277daeb8660d5273b7c3e05c263f861ed5f17b92 192.168.30.130:7006@17006 slave 4d74ec66e898bf09006dac86d4928f9fad81f373 0 1557455174000 6 connected b4d3eb411a7355d4767c6c23b4df69fa183ef8bc 192.168.30.128:7002@17002 slave 6788453ee9a8d7f72b1d45a9093838efd0e501f1 0 1557455175000 5 connected 80c80a3f3e33872c047a8328ad579b9bea001ad8 192.168.30.128:7001@17001 myself,master - 0 1557455175000 1 connected 0-5460 b6331cbc986794237c83ed2d5c30777c1551546e 192.168.30.129:7004@17004 slave 80c80a3f3e33872c047a8328ad579b9bea001ad8 0 1557455174989 4 connected 4d74ec66e898bf09006dac86d4928f9fad81f373 192.168.30.129:7003@17003 master - 0 1557455175995 3 connected 5461-10922
這里與nodes.conf文件內(nèi)容相同
- 寫入數(shù)據(jù)：
192.168.30.128:7001> set key111 aaa -> Redirected to slot [13680] located at 192.168.30.130:7005 #說明數(shù)據(jù)到了192.168.30.130:7005上OK 192.168.30.130:7005> set key222 bbb -> Redirected to slot [2320] located at 192.168.30.128:7001 #說明數(shù)據(jù)到了192.168.30.128:7001上OK 192.168.30.128:7001> set key333 ccc -> Redirected to slot [7472] located at 192.168.30.129:7003 #說明數(shù)據(jù)到了192.168.30.129:7003上OK 192.168.30.129:7003> get key111 -> Redirected to slot [13680] located at 192.168.30.130:7005"aaa"192.168.30.130:7005> get key333 -> Redirected to slot [7472] located at 192.168.30.129:7003"ccc"192.168.30.129:7003>
可以看出redis cluster集群是去中心化的，每個節(jié)點都是平等的，連接哪個節(jié)點都可以獲取和設(shè)置數(shù)據(jù)。
當(dāng)然，平等指的是master節(jié)點，因為slave節(jié)點根本不提供服務(wù)，只是作為對應(yīng)master節(jié)點的一個備份。
- 增加節(jié)點：
192.168.30.129上增加一節(jié)點：
# cp /usr/local/redis/cluster/redis_7003.conf /usr/local/redis/cluster/redis_7007.conf# vim /usr/local/redis/cluster/redis_7007.confbind 192.168.30.129 port 7007 daemonize yespidfile "/var/run/redis_7007.pid"logfile "/usr/local/redis/cluster/redis_7007.log"dir "/data/redis/cluster/redis_7007"#replicaof 192.168.30.129 6379masterauth "123456"requirepass "123456"appendonly yescluster-enabled yescluster-config-file nodes_7007.conf cluster-node-timeout 15000# mkdir /data/redis/cluster/redis_7007# chown -R redis:redis /usr/local/redis && chown -R redis:redis /data/redis# redis-server /usr/local/redis/cluster/redis_7007.conf
192.168.30.130上增加一節(jié)點：
# cp /usr/local/redis/cluster/redis_7005.conf /usr/local/redis/cluster/redis_7008.conf# vim /usr/local/redis/cluster/redis_7007.confbind 192.168.30.130 port 7008 daemonize yespidfile "/var/run/redis_7008.pid"logfile "/usr/local/redis/cluster/redis_7008.log"dir "/data/redis/cluster/redis_7008"#replicaof 192.168.30.130 6379masterauth "123456"requirepass "123456"appendonly yescluster-enabled yescluster-config-file nodes_7008.conf cluster-node-timeout 15000# mkdir /data/redis/cluster/redis_7008# chown -R redis:redis /usr/local/redis && chown -R redis:redis /data/redis# redis-server /usr/local/redis/cluster/redis_7008.conf
集群中增加節(jié)點：
192.168.30.129:7003> CLUSTER MEET 192.168.30.129 7007 OK 192.168.30.129:7003> CLUSTER NODES 4d74ec66e898bf09006dac86d4928f9fad81f373 192.168.30.129:7003@17003 myself,master - 0 1557457361000 3 connected 5461-10922 80c80a3f3e33872c047a8328ad579b9bea001ad8 192.168.30.128:7001@17001 master - 0 1557457364746 1 connected 0-5460 277daeb8660d5273b7c3e05c263f861ed5f17b92 192.168.30.130:7006@17006 slave 4d74ec66e898bf09006dac86d4928f9fad81f373 0 1557457362000 6 connected b6331cbc986794237c83ed2d5c30777c1551546e 192.168.30.129:7004@17004 slave 80c80a3f3e33872c047a8328ad579b9bea001ad8 0 1557457363000 4 connected b4d3eb411a7355d4767c6c23b4df69fa183ef8bc 192.168.30.128:7002@17002 slave 6788453ee9a8d7f72b1d45a9093838efd0e501f1 0 1557457362000 5 connected e51ab166bc0f33026887bcf8eba0dff3d5b0bf14 192.168.30.129:7007@17007 master - 0 1557457362729 0 connected 6788453ee9a8d7f72b1d45a9093838efd0e501f1 192.168.30.130:7005@17005 master - 0 1557457363739 5 connected 10923-16383
192.168.30.129:7003> CLUSTER MEET 192.168.30.130 7008 OK 192.168.30.129:7003> CLUSTER NODES 4d74ec66e898bf09006dac86d4928f9fad81f373 192.168.30.129:7003@17003 myself,master - 0 1557457489000 3 connected 5461-10922 80c80a3f3e33872c047a8328ad579b9bea001ad8 192.168.30.128:7001@17001 master - 0 1557457489000 1 connected 0-5460 277daeb8660d5273b7c3e05c263f861ed5f17b92 192.168.30.130:7006@17006 slave 4d74ec66e898bf09006dac86d4928f9fad81f373 0 1557457489000 6 connected b6331cbc986794237c83ed2d5c30777c1551546e 192.168.30.129:7004@17004 slave 80c80a3f3e33872c047a8328ad579b9bea001ad8 0 1557457488000 4 connected b4d3eb411a7355d4767c6c23b4df69fa183ef8bc 192.168.30.128:7002@17002 slave 6788453ee9a8d7f72b1d45a9093838efd0e501f1 0 1557457489472 5 connected 1a1c7f02fce87530bd5abdfc98df1cffce4f1767 192.168.30.130:7008@17008 master - 0 1557457489259 0 connected e51ab166bc0f33026887bcf8eba0dff3d5b0bf14 192.168.30.129:7007@17007 master - 0 1557457489000 0 connected 6788453ee9a8d7f72b1d45a9093838efd0e501f1 192.168.30.130:7005@17005 master - 0 1557457490475 5 connected 10923-16383
可以看到，新增的節(jié)點都是以master身份加入集群的
- 更換節(jié)點身份：
將新增的192.168.30.130:7008節(jié)點身份改為192.168.30.129:7007的slave
# redis-cli -c -h 192.168.30.130 -p 7008 -a 123456 cluster replicate e51ab166bc0f33026887bcf8eba0dff3d5b0bf14
cluster replicate后面跟node_id，更改對應(yīng)節(jié)點身份。也可以登入集群更改
# redis-cli -c -h 192.168.30.130 -p 7008 -a 123456192.168.30.130:7008> CLUSTER REPLICATE e51ab166bc0f33026887bcf8eba0dff3d5b0bf14 OK 192.168.30.130:7008> CLUSTER NODES 277daeb8660d5273b7c3e05c263f861ed5f17b92 192.168.30.130:7006@17006 slave 4d74ec66e898bf09006dac86d4928f9fad81f373 0 1557458316881 3 connected 80c80a3f3e33872c047a8328ad579b9bea001ad8 192.168.30.128:7001@17001 master - 0 1557458314864 1 connected 0-5460 4d74ec66e898bf09006dac86d4928f9fad81f373 192.168.30.129:7003@17003 master - 0 1557458316000 3 connected 5461-10922 6788453ee9a8d7f72b1d45a9093838efd0e501f1 192.168.30.130:7005@17005 master - 0 1557458315872 5 connected 10923-16383 b4d3eb411a7355d4767c6c23b4df69fa183ef8bc 192.168.30.128:7002@17002 slave 6788453ee9a8d7f72b1d45a9093838efd0e501f1 0 1557458317890 5 connected 1a1c7f02fce87530bd5abdfc98df1cffce4f1767 192.168.30.130:7008@17008 myself,slave e51ab166bc0f33026887bcf8eba0dff3d5b0bf14 0 1557458315000 7 connected b6331cbc986794237c83ed2d5c30777c1551546e 192.168.30.129:7004@17004 slave 80c80a3f3e33872c047a8328ad579b9bea001ad8 0 1557458315000 1 connected e51ab166bc0f33026887bcf8eba0dff3d5b0bf14 192.168.30.129:7007@17007 master - 0 1557458314000 0 connected
查看相應(yīng)的nodes.conf文件，可以發(fā)現(xiàn)有更改，它記錄當(dāng)前集群的節(jié)點信息
# cat /data/redis/cluster/redis_7001/nodes-7001.conf1a1c7f02fce87530bd5abdfc98df1cffce4f1767 192.168.30.130:7008@17008 slave e51ab166bc0f33026887bcf8eba0dff3d5b0bf14 0 1557458236169 7 connected 6788453ee9a8d7f72b1d45a9093838efd0e501f1 192.168.30.130:7005@17005 master - 0 1557458235000 5 connected 10923-16383 277daeb8660d5273b7c3e05c263f861ed5f17b92 192.168.30.130:7006@17006 slave 4d74ec66e898bf09006dac86d4928f9fad81f373 0 1557458234103 6 connected b4d3eb411a7355d4767c6c23b4df69fa183ef8bc 192.168.30.128:7002@17002 slave 6788453ee9a8d7f72b1d45a9093838efd0e501f1 0 1557458235129 5 connected 80c80a3f3e33872c047a8328ad579b9bea001ad8 192.168.30.128:7001@17001 myself,master - 0 1557458234000 1 connected 0-5460 b6331cbc986794237c83ed2d5c30777c1551546e 192.168.30.129:7004@17004 slave 80c80a3f3e33872c047a8328ad579b9bea001ad8 0 1557458236000 4 connected e51ab166bc0f33026887bcf8eba0dff3d5b0bf14 192.168.30.129:7007@17007 master - 0 1557458236000 0 connected 4d74ec66e898bf09006dac86d4928f9fad81f373 192.168.30.129:7003@17003 master - 0 1557458233089 3 connected 5461-10922 vars currentEpoch 7 lastVoteEpoch 0
- 刪除節(jié)點：
192.168.30.130:7008> CLUSTER FORGET 1a1c7f02fce87530bd5abdfc98df1cffce4f1767(error) ERR I tried hard but I can't forget myself... #無法刪除登錄節(jié)點 192.168.30.130:7008> CLUSTER FORGET e51ab166bc0f33026887bcf8eba0dff3d5b0bf14 (error) ERR Can't forget my master! #不能刪除自己的master節(jié)點192.168.30.130:7008> CLUSTER FORGET 6788453ee9a8d7f72b1d45a9093838efd0e501f1 OK #可以刪除其它的master節(jié)點192.168.30.130:7008> CLUSTER NODES 277daeb8660d5273b7c3e05c263f861ed5f17b92 192.168.30.130:7006@17006 slave 4d74ec66e898bf09006dac86d4928f9fad81f373 0 1557458887328 3 connected 80c80a3f3e33872c047a8328ad579b9bea001ad8 192.168.30.128:7001@17001 master - 0 1557458887000 1 connected 0-5460 4d74ec66e898bf09006dac86d4928f9fad81f373 192.168.30.129:7003@17003 master - 0 1557458886000 3 connected 5461-10922 b4d3eb411a7355d4767c6c23b4df69fa183ef8bc 192.168.30.128:7002@17002 slave - 0 1557458888351 5 connected 1a1c7f02fce87530bd5abdfc98df1cffce4f1767 192.168.30.130:7008@17008 myself,slave e51ab166bc0f33026887bcf8eba0dff3d5b0bf14 0 1557458885000 7 connected b6331cbc986794237c83ed2d5c30777c1551546e 192.168.30.129:7004@17004 slave 80c80a3f3e33872c047a8328ad579b9bea001ad8 0 1557458883289 1 connected e51ab166bc0f33026887bcf8eba0dff3d5b0bf14 192.168.30.129:7007@17007 master - 0 1557458885310 0 connected 192.168.30.130:7008> CLUSTER FORGET b4d3eb411a7355d4767c6c23b4df69fa183ef8bc OK #可以刪除其它的slave節(jié)點192.168.30.130:7008> CLUSTER NODES 277daeb8660d5273b7c3e05c263f861ed5f17b92 192.168.30.130:7006@17006 slave 4d74ec66e898bf09006dac86d4928f9fad81f373 0 1557459031397 3 connected 80c80a3f3e33872c047a8328ad579b9bea001ad8 192.168.30.128:7001@17001 master - 0 1557459032407 1 connected 0-5460 4d74ec66e898bf09006dac86d4928f9fad81f373 192.168.30.129:7003@17003 master - 0 1557459035434 3 connected 5461-10922 6788453ee9a8d7f72b1d45a9093838efd0e501f1 192.168.30.130:7005@17005 master - 0 1557459034000 5 connected 10923-16383 1a1c7f02fce87530bd5abdfc98df1cffce4f1767 192.168.30.130:7008@17008 myself,slave e51ab166bc0f33026887bcf8eba0dff3d5b0bf14 0 1557459032000 7 connected b6331cbc986794237c83ed2d5c30777c1551546e 192.168.30.129:7004@17004 slave 80c80a3f3e33872c047a8328ad579b9bea001ad8 0 1557459034000 1 connected e51ab166bc0f33026887bcf8eba0dff3d5b0bf14 192.168.30.129:7007@17007 master - 0 1557459034427 0 connected
- 保存配置：
192.168.30.130:7008> CLUSTER SAVECONFIG #將節(jié)點配置信息保存到硬盤OK # cat /data/redis/cluster/redis_7001/nodes-7001.conf1a1c7f02fce87530bd5abdfc98df1cffce4f1767 192.168.30.130:7008@17008 slave e51ab166bc0f33026887bcf8eba0dff3d5b0bf14 0 1557458236169 7 connected 6788453ee9a8d7f72b1d45a9093838efd0e501f1 192.168.30.130:7005@17005 master - 0 1557458235000 5 connected 10923-16383 277daeb8660d5273b7c3e05c263f861ed5f17b92 192.168.30.130:7006@17006 slave 4d74ec66e898bf09006dac86d4928f9fad81f373 0 1557458234103 6 connected b4d3eb411a7355d4767c6c23b4df69fa183ef8bc 192.168.30.128:7002@17002 slave 6788453ee9a8d7f72b1d45a9093838efd0e501f1 0 1557458235129 5 connected 80c80a3f3e33872c047a8328ad579b9bea001ad8 192.168.30.128:7001@17001 myself,master - 0 1557458234000 1 connected 0-5460 b6331cbc986794237c83ed2d5c30777c1551546e 192.168.30.129:7004@17004 slave 80c80a3f3e33872c047a8328ad579b9bea001ad8 0 1557458236000 4 connected e51ab166bc0f33026887bcf8eba0dff3d5b0bf14 192.168.30.129:7007@17007 master - 0 1557458236000 0 connected 4d74ec66e898bf09006dac86d4928f9fad81f373 192.168.30.129:7003@17003 master - 0 1557458233089 3 connected 5461-10922 vars currentEpoch 7 lastVoteEpoch 0# redis-cli -c -h 192.168.30.130 -p 7008 -a 123456Warning: Using a password with '-a' or '-u' option on the command line interface may not be safe. 192.168.30.130:7008> CLUSTER NODES 277daeb8660d5273b7c3e05c263f861ed5f17b92 192.168.30.130:7006@17006 slave 4d74ec66e898bf09006dac86d4928f9fad81f373 0 1557459500741 3 connected 80c80a3f3e33872c047a8328ad579b9bea001ad8 192.168.30.128:7001@17001 master - 0 1557459500000 1 connected 0-5460 4d74ec66e898bf09006dac86d4928f9fad81f373 192.168.30.129:7003@17003 master - 0 1557459501000 3 connected 5461-10922 6788453ee9a8d7f72b1d45a9093838efd0e501f1 192.168.30.130:7005@17005 master - 0 1557459500000 5 connected 10923-16383 b4d3eb411a7355d4767c6c23b4df69fa183ef8bc 192.168.30.128:7002@17002 slave 6788453ee9a8d7f72b1d45a9093838efd0e501f1 0 1557459499737 5 connected 1a1c7f02fce87530bd5abdfc98df1cffce4f1767 192.168.30.130:7008@17008 myself,slave e51ab166bc0f33026887bcf8eba0dff3d5b0bf14 0 1557459499000 7 connected b6331cbc986794237c83ed2d5c30777c1551546e 192.168.30.129:7004@17004 slave 80c80a3f3e33872c047a8328ad579b9bea001ad8 0 1557459501750 1 connected e51ab166bc0f33026887bcf8eba0dff3d5b0bf14 192.168.30.129:7007@17007 master - 0 1557459498000 0 connected
可以看到，之前刪除的節(jié)點又恢復(fù)了，這是因為對應(yīng)的配置文件沒有刪除，執(zhí)行CLUSTER SAVECONFIG恢復(fù)。
- 模擬master節(jié)點掛掉：
192.168.30.128
# netstat -lntp |grep 7001tcp 0 0 192.168.30.128:17001 0.0.0.0:* LISTEN 6701/redis-server 1 tcp 0 0 192.168.30.128:7001 0.0.0.0:* LISTEN 6701/redis-server 1 # kill 6701
192.168.30.130:7008> CLUSTER NODES 277daeb8660d5273b7c3e05c263f861ed5f17b92 192.168.30.130:7006@17006 slave 4d74ec66e898bf09006dac86d4928f9fad81f373 0 1557461178000 3 connected 80c80a3f3e33872c047a8328ad579b9bea001ad8 192.168.30.128:7001@17001 master,fail - 1557460950483 1557460947145 1 disconnected 4d74ec66e898bf09006dac86d4928f9fad81f373 192.168.30.129:7003@17003 master - 0 1557461174922 3 connected 5461-10922 6788453ee9a8d7f72b1d45a9093838efd0e501f1 192.168.30.130:7005@17005 master - 0 1557461181003 5 connected 10923-16383 b4d3eb411a7355d4767c6c23b4df69fa183ef8bc 192.168.30.128:7002@17002 slave 6788453ee9a8d7f72b1d45a9093838efd0e501f1 0 1557461179993 5 connected 1a1c7f02fce87530bd5abdfc98df1cffce4f1767 192.168.30.130:7008@17008 myself,slave e51ab166bc0f33026887bcf8eba0dff3d5b0bf14 0 1557461176000 7 connected b6331cbc986794237c83ed2d5c30777c1551546e 192.168.30.129:7004@17004 master - 0 1557461178981 8 connected 0-5460 e51ab166bc0f33026887bcf8eba0dff3d5b0bf14 192.168.30.129:7007@17007 master - 0 1557461179000 0 connected
對應(yīng)7001的一行可以看到，master fail，狀態(tài)為disconnected；而對應(yīng)7004的一行，slave已經(jīng)變成master。
- 重新啟動7001節(jié)點：
# redis-server /usr/local/redis/cluster/redis_7001.conf192.168.30.130:7008> CLUSTER NODES 277daeb8660d5273b7c3e05c263f861ed5f17b92 192.168.30.130:7006@17006 slave 4d74ec66e898bf09006dac86d4928f9fad81f373 0 1557461307000 3 connected 80c80a3f3e33872c047a8328ad579b9bea001ad8 192.168.30.128:7001@17001 slave b6331cbc986794237c83ed2d5c30777c1551546e 0 1557461305441 8 connected 4d74ec66e898bf09006dac86d4928f9fad81f373 192.168.30.129:7003@17003 master - 0 1557461307962 3 connected 5461-10922 6788453ee9a8d7f72b1d45a9093838efd0e501f1 192.168.30.130:7005@17005 master - 0 1557461304935 5 connected 10923-16383 b4d3eb411a7355d4767c6c23b4df69fa183ef8bc 192.168.30.128:7002@17002 slave 6788453ee9a8d7f72b1d45a9093838efd0e501f1 0 1557461306000 5 connected 1a1c7f02fce87530bd5abdfc98df1cffce4f1767 192.168.30.130:7008@17008 myself,slave e51ab166bc0f33026887bcf8eba0dff3d5b0bf14 0 1557461305000 7 connected b6331cbc986794237c83ed2d5c30777c1551546e 192.168.30.129:7004@17004 master - 0 1557461308972 8 connected 0-5460 e51ab166bc0f33026887bcf8eba0dff3d5b0bf14 192.168.30.129:7007@17007 master - 0 1557461307000 0 connected
可以看到，7001節(jié)點啟動后為slave節(jié)點，并且是7004的slave節(jié)點。即master節(jié)點如果掛掉，它的slave節(jié)點變?yōu)樾耺aster節(jié)點繼續(xù)對外提供服務(wù)，而原來的master節(jié)點如果重啟，則變?yōu)樾耺aster節(jié)點的slave節(jié)點。
另外，如果這里是拿7007節(jié)點做測試的話，會發(fā)現(xiàn)7008節(jié)點并不會切換，這是因為7007節(jié)點上根本沒數(shù)據(jù)。集群數(shù)據(jù)被分為三份，采用哈希槽 (hash slot)的方式來分配16384個slot的話，它們?nèi)齻€節(jié)點分別承擔(dān)的slot 區(qū)間是：
節(jié)點7004覆蓋0－5460 節(jié)點7003覆蓋5461－10922 節(jié)點7005覆蓋10923－16383
