redis3最大的变化之一就是cluster功能的正式发布,以前要搞redis集群,得借助一致性hash来自己搞sharding,现在方便多了,直接上cluster功能就行了,而且还支持节点动态添加、HA、节点增减后缓存重新分布(resharding)。
下面是参考官方教程 在mac机上搭建cluster的过程:
一、下载最新版redis 编译
目前最新版是3.0.7,下载地址:
编译很简单,一个make命令即可,不清楚的同学,可参考我之前的笔记:
二、建6个目录
mkdir ~/app/redis-cluster/ #先建一个根目录mkdir 7000 7001 7002 7003 7004 7005
注:与大多数分布式中间件一样,redis的cluster也是依赖选举算法来保证集群的高可用,所以类似ZK一样,一般是奇数个节点(可以允许N/2以下的节点失效),再考虑到每个节点做Master-Slave互为备份,所以一个redis cluster集群最少也得6个节点。
然后把步骤1里编译好的redis,复制到这6个目录下。
三、配置文件
port 7000cluster-enabled yescluster-config-file nodes.confcluster-node-timeout 5000appendonly yes
把上面这段保存成redis-cluster.conf,放到每个目录的redis目录中,注意修改port端口,即7000目录下的port为7000,7001目录下的port为7001...
cluster-node-timeout 是集群中各节点相互通讯时,允许"失联"的最大毫秒数,上面的配置为5秒,如果超过5秒某个节点没向其它节点汇报成功,认为该节点挂了。
四、依次启动各个redis
在每个目录redis的src子目录下,输入:
./redis-server ../redis-cluster.conf
这样7000~7005这6个节点就启动了。
五、安装redis的ruby模块
brew updatebrew install rubysudo gem install redis #注:这个步骤建议翻^墙,不然你懂的
解释:虽然步骤4把6个redis server启动成功了,但是彼此之间是完全独立的,需要借助其它工具将其加入cluster,而这个工具就是redis提供的一个名为redis-trib.rb的ruby脚本(个人估计redis的作者比较偏爱ruby),mac自带了ruby2.0环境,但是没有redis模块,所以要安装这玩意儿,否则接下来的创建cluster将失败。
六、创建cluster
./redis-trib.rb create --replicas 1 127.0.0.1:7000 127.0.0.1:7001 \127.0.0.1:7002 127.0.0.1:7003 127.0.0.1:7004 127.0.0.1:7005
仍然保持在某个目录的src子目录下,运行上面这段shell脚本,cluster就创建成功了,replicas 1的意思,就是每个节点创建1个副本(即:slave),所以最终的结果,就是后面的127.0.0.1:7000~127.0.0.1:7005中,会有3个会指定成master,而其它3个会指定成slave。
注:利用redis-trib创建cluster的操作,只需要一次即可,假设系统关机,把所有6个节点全关闭后,下次重启后,即自动进入cluster模式,不用再次redis-trib.rb create。
此时,如何用ps查看redis进程,会看到每个进程后附带了cluster的字样
如果想知道,哪些端口的节点是master,哪些端口的节点是slave,可以用下面的命令:
./redis-trib.rb check 127.0.0.1:7000
输出结果如下:
>>> Performing Cluster Check (using node 127.0.0.1:7000)S: 0b7e0d5337e87ac7b59bba4c1248e5c9e8d1905e 127.0.0.1:7000 slots: (0 slots) slave replicates 38910c5baafea02c5303505acfd9bd331c608cfcM: e0e8dfddd4e9d855090d6efd18e55ea9c0e1f7aa 127.0.0.1:7001 slots:5461-10922 (5462 slots) master 1 additional replica(s)S: 88e16f91609c03277f2ee6ce5285932f58c221c1 127.0.0.1:7005 slots: (0 slots) slave replicates ec964a7c7cd53b986f54318a190c1426fc53a5faS: be7e9fd3b7d096b037306bc14e1017150fa59d7a 127.0.0.1:7004 slots: (0 slots) slave replicates e0e8dfddd4e9d855090d6efd18e55ea9c0e1f7aaM: 38910c5baafea02c5303505acfd9bd331c608cfc 127.0.0.1:7003 slots:0-5460 (5461 slots) master 1 additional replica(s)M: ec964a7c7cd53b986f54318a190c1426fc53a5fa 127.0.0.1:7002 slots:10923-16383 (5461 slots) master 1 additional replica(s)[OK] All nodes agree about slots configuration.>>> Check for open slots...>>> Check slots coverage...[OK] All 16384 slots covered.
从上面的输出,可以看出7000、7004、7005是slave,而7001、7003、7002是master(如果大家人为做过一些failover的测试,比如把某个节点手动停掉,再恢复,输出的结果可能与上面不太一样),除了check参数,还有一个常用的参数info
./redis-trib.rb info 127.0.0.1:7000
输出结果如下:
127.0.0.1:7001 (e0e8dfdd...) -> 2 keys | 5462 slots | 1 slaves.127.0.0.1:7003 (38910c5b...) -> 2 keys | 5461 slots | 1 slaves.127.0.0.1:7002 (ec964a7c...) -> 0 keys | 5461 slots | 1 slaves.[OK] 4 keys in 3 masters.0.00 keys per slot on average.
它会把所有的master信息输出,包括这个master上有几个缓存key,有几个slave,所有master上的keys合计,以及平均每个slot上有多少key,想了解更多redis-trib脚本的其它参数,可以用
./redis-trib.rb help
输出如下:
Usage: redis-tribcreate host1:port1 ... hostN:portN --replicas check host:port info host:port fix host:port --timeout reshard host:port --from --to --slots --yes --timeout --pipeline rebalance host:port --weight --auto-weights --use-empty-masters --timeout --simulate --pipeline --threshold add-node new_host:new_port existing_host:existing_port --slave --master-id del-node host:port node_id set-timeout host:port milliseconds call host:port command arg arg .. arg import host:port --from --copy --replace help (show this help) For check, fix, reshard, del-node, set-timeout you can specify the host and port of any working node in the cluster.
上面已经多次出现了slot这个词,略为解释一下:
如上图,redis-cluster把整个集群的存储空间划分为16384个slot(译为:插槽?),当6个节点分为3主3从时,相当于整个cluster中有3组HA的节点,3个master会平均分摊所有slot,每次向cluster中的key做操作时(比如:读取/写入缓存),redis会对key值做CRC32算法处理,得到一个数值,然后再对16384取模,通过余数判断该缓存项应该落在哪个slot上,确定了slot,也就确定了保存在哪个master节点上,当cluster扩容或删除节点时,只需要将slot重新分配即可(即:把部分slot从一些节点移动到其它节点)。
七、redis-cli客户端操作
./redis-cli -c -h localhost -p 7000
注意加参数-c,表示进入cluster模式,随便添加一个缓存试试:
localhost:7000> set user1 jimmy-> Redirected to slot [8106] located at 127.0.0.1:7001OK
注意第2行的输出,表示user1这个缓存通过计算后,落在8106这个slot上,最终定位在7001这个端口对应的节点上(解释:因为7000是slave,7001才是master,只有master才能写入),如果是在7001上重复上面的操作时,不会出现第2行(解释:7001是master,所以不存在redirect的过程)
➜ src ./redis-cli -c -h localhost -p 7001localhost:7001> set user1 yangOKlocalhost:7001>
八、FailOver测试
先用redis-trib.rb 查看下当前的主、从情况
➜ src ./redis-trib.rb check localhost:7000>>> Performing Cluster Check (using node localhost:7000)S: 0b7e0d5337e87ac7b59bba4c1248e5c9e8d1905e localhost:7000 slots: (0 slots) slave replicates 38910c5baafea02c5303505acfd9bd331c608cfcM: ec964a7c7cd53b986f54318a190c1426fc53a5fa 127.0.0.1:7002 slots:10923-16383 (5461 slots) master 1 additional replica(s)M: e0e8dfddd4e9d855090d6efd18e55ea9c0e1f7aa 127.0.0.1:7001 slots:5461-10922 (5462 slots) master 1 additional replica(s)S: be7e9fd3b7d096b037306bc14e1017150fa59d7a 127.0.0.1:7004 slots: (0 slots) slave replicates e0e8dfddd4e9d855090d6efd18e55ea9c0e1f7aaS: 88e16f91609c03277f2ee6ce5285932f58c221c1 127.0.0.1:7005 slots: (0 slots) slave replicates ec964a7c7cd53b986f54318a190c1426fc53a5faM: 38910c5baafea02c5303505acfd9bd331c608cfc 127.0.0.1:7003 slots:0-5460 (5461 slots) master 1 additional replica(s)[OK] All nodes agree about slots configuration.>>> Check for open slots...>>> Check slots coverage...[OK] All 16384 slots covered.
从输出上看7000是7003(38910c5baafea02c5303505acfd9bd331c608cfc)的slave,现在我们人工把7003的redis进程给kill掉,然后观察7000的终端输出:
872:S 21 Mar 10:55:55.663 * Connecting to MASTER 127.0.0.1:70033872:S 21 Mar 10:55:55.663 * MASTER <-> SLAVE sync started3872:S 21 Mar 10:55:55.663 # Error condition on socket for SYNC: Connection refused3872:S 21 Mar 10:55:55.771 * Marking node 38910c5baafea02c5303505acfd9bd331c608cfc as failing (quorum reached).3872:S 21 Mar 10:55:55.771 # Cluster state changed: fail3872:S 21 Mar 10:55:55.869 # Start of election delayed for 954 milliseconds (rank #0, offset 183).3872:S 21 Mar 10:55:56.703 * Connecting to MASTER 127.0.0.1:70033872:S 21 Mar 10:55:56.703 * MASTER <-> SLAVE sync started3872:S 21 Mar 10:55:56.703 # Error condition on socket for SYNC: Connection refused3872:S 21 Mar 10:55:56.909 # Starting a failover election for epoch 10.3872:S 21 Mar 10:55:56.911 # Failover election won: I'm the new master.3872:S 21 Mar 10:55:56.911 # configEpoch set to 10 after successful failover3872:M 21 Mar 10:55:56.911 * Discarding previously cached master state.3872:M 21 Mar 10:55:56.911 # Cluster state changed: ok
注意5,6,11这几行,第5行表明由于7003宕机,cluster状态已经切换到fail状态,第6行表示发起选举,第11行表示7000端口对应的节点当选为new master。
九、cluster 扩容
业务规模变大后,集群扩容是早晚的事情,下面演示如何再添加2个节点,先把7000复制二份,变成7006,7007,然后进入7006/7007目录redis的src子目录下
rm nodes.conf dump.rdb appendonly.aof
由于7000我们刚才启动过,里面有已经有一些数据了,所以要把数据文件,日志文件,以及cluster的nodes.conf文件删除,变成一个空的redis独立节点,否则无法加入cluster。
然后修改redis-cluster.conf
port 7000cluster-enabled yescluster-config-file "nodes.conf"cluster-node-timeout 10000appendonly yes# Generated by CONFIG REWRITEdir "/Users/yjmyzz/app/redis-cluster/7000/redis-3.0.7/src"
要修改的地方有二处,1是第一行的端口,改成与7006/7007匹配的端口,2是最后2行,这是7000运行后,自动添加的,把最后二行删除。
做完这些后,启动7006,7007这二个redis节点,此时这2个新节点与cluster没有任何关系,可以用下面的命令将7006做为master添加到cluster中。
./redis-trib.rb add-node 127.0.0.1:7006 127.0.0.1:7000
注:第1个参数为新节点的"IP:端口",第2个参数为集群中的任一有效的节点。
顺利的话,输出如下:
>>> Adding node 127.0.0.1:7006 to cluster 127.0.0.1:7000>>> Performing Cluster Check (using node 127.0.0.1:7000)M: 0b7e0d5337e87ac7b59bba4c1248e5c9e8d1905e 127.0.0.1:7000 slots:0-5460 (5461 slots) master 1 additional replica(s)M: e0e8dfddd4e9d855090d6efd18e55ea9c0e1f7aa 127.0.0.1:7001 slots:5461-10922 (5462 slots) master 1 additional replica(s)S: be7e9fd3b7d096b037306bc14e1017150fa59d7a 127.0.0.1:7004 slots: (0 slots) slave replicates e0e8dfddd4e9d855090d6efd18e55ea9c0e1f7aaM: ec964a7c7cd53b986f54318a190c1426fc53a5fa 127.0.0.1:7002 slots:10923-16383 (5461 slots) master 1 additional replica(s)S: 88e16f91609c03277f2ee6ce5285932f58c221c1 127.0.0.1:7005 slots: (0 slots) slave replicates ec964a7c7cd53b986f54318a190c1426fc53a5faS: 38910c5baafea02c5303505acfd9bd331c608cfc 127.0.0.1:7003 slots: (0 slots) slave replicates 0b7e0d5337e87ac7b59bba4c1248e5c9e8d1905e[OK] All nodes agree about slots configuration.>>> Check for open slots...>>> Check slots coverage...[OK] All 16384 slots covered.>>> Send CLUSTER MEET to node 127.0.0.1:7006 to make it join the cluster.[OK] New node added correctly.
可以再用check确认下状态:
➜ src ./redis-trib.rb check 127.0.0.1:7000>>> Performing Cluster Check (using node 127.0.0.1:7000)M: 0b7e0d5337e87ac7b59bba4c1248e5c9e8d1905e 127.0.0.1:7000 slots:0-5460 (5461 slots) master 1 additional replica(s)M: e0e8dfddd4e9d855090d6efd18e55ea9c0e1f7aa 127.0.0.1:7001 slots:5461-10922 (5462 slots) master 1 additional replica(s)S: be7e9fd3b7d096b037306bc14e1017150fa59d7a 127.0.0.1:7004 slots: (0 slots) slave replicates e0e8dfddd4e9d855090d6efd18e55ea9c0e1f7aaM: 226d1af3c95bf0798ea9fed86373b89347f889da 127.0.0.1:7006 slots: (0 slots) master 0 additional replica(s)M: ec964a7c7cd53b986f54318a190c1426fc53a5fa 127.0.0.1:7002 slots:10923-16383 (5461 slots) master 1 additional replica(s)S: 88e16f91609c03277f2ee6ce5285932f58c221c1 127.0.0.1:7005 slots: (0 slots) slave replicates ec964a7c7cd53b986f54318a190c1426fc53a5faS: 38910c5baafea02c5303505acfd9bd331c608cfc 127.0.0.1:7003 slots: (0 slots) slave replicates 0b7e0d5337e87ac7b59bba4c1248e5c9e8d1905e[OK] All nodes agree about slots configuration.>>> Check for open slots...>>> Check slots coverage...[OK] All 16384 slots covered.
12-14行说明7006已经是cluster的新master了,继续,用下面的命令把7007当成slave加入:
./redis-trib.rb add-node --slave --master-id 226d1af3c95bf0798ea9fed86373b89347f889da 127.0.0.1:7007 127.0.0.1:7000
这里多出了二个参数:--slave 表示准备将新节点当成slave加入,--master-id xxxxx 则是指定要当谁的slave,后面的xxx部分,即为前面check的输出结果中,7006的ID,完事之后,可以再次确认状态:
➜ src ./redis-trib.rb check 127.0.0.1:7000>>> Performing Cluster Check (using node 127.0.0.1:7000)M: 0b7e0d5337e87ac7b59bba4c1248e5c9e8d1905e 127.0.0.1:7000 slots:0-5460 (5461 slots) master 1 additional replica(s)S: 792bcccf35845c4922dd33d7f9827420ebb89bc9 127.0.0.1:7007 slots: (0 slots) slave replicates 226d1af3c95bf0798ea9fed86373b89347f889daM: e0e8dfddd4e9d855090d6efd18e55ea9c0e1f7aa 127.0.0.1:7001 slots:5461-10922 (5462 slots) master 1 additional replica(s)S: be7e9fd3b7d096b037306bc14e1017150fa59d7a 127.0.0.1:7004 slots: (0 slots) slave replicates e0e8dfddd4e9d855090d6efd18e55ea9c0e1f7aaM: 226d1af3c95bf0798ea9fed86373b89347f889da 127.0.0.1:7006 slots: (0 slots) master 1 additional replica(s)M: ec964a7c7cd53b986f54318a190c1426fc53a5fa 127.0.0.1:7002 slots:10923-16383 (5461 slots) master 1 additional replica(s)S: 88e16f91609c03277f2ee6ce5285932f58c221c1 127.0.0.1:7005 slots: (0 slots) slave replicates ec964a7c7cd53b986f54318a190c1426fc53a5faS: 38910c5baafea02c5303505acfd9bd331c608cfc 127.0.0.1:7003 slots: (0 slots) slave replicates 0b7e0d5337e87ac7b59bba4c1248e5c9e8d1905e[OK] All nodes agree about slots configuration.>>> Check for open slots...>>> Check slots coverage...[OK] All 16384 slots covered.
观察6-8行、15-17行,说明7007已经是7006的slave。
十、reshard 重新划分slot
增加新的节点之后,问题就来了,16384个slot已经被其它3组节点分完了,新节点没有slot,没办法存放缓存,所以需要将slot重新分布。
➜ src ./redis-trib.rb info 127.0.0.1:7000127.0.0.1:7000 (0b7e0d53...) -> 4 keys | 5461 slots | 1 slaves.127.0.0.1:7001 (e0e8dfdd...) -> 4 keys | 5462 slots | 1 slaves.127.0.0.1:7006 (226d1af3...) -> 0 keys | 0 slots | 1 slaves. #7006上完全没有slot127.0.0.1:7002 (ec964a7c...) -> 9 keys | 5461 slots | 1 slaves.[OK] 17 keys in 4 masters.0.00 keys per slot on average.
用下面的命令可以重新分配slot
./redis-trib.rb reshard 127.0.0.1:7000
reshard后面的IP:port,只要是在cluster中的有效节点即可。
➜ src ./redis-trib.rb reshard 127.0.0.1:7000>>> Performing Cluster Check (using node 127.0.0.1:7000)M: 0b7e0d5337e87ac7b59bba4c1248e5c9e8d1905e 127.0.0.1:7000 slots:1792-4095 (2304 slots) master 0 additional replica(s) ...[OK] All nodes agree about slots configuration.>>> Check for open slots...>>> Check slots coverage...[OK] All 16384 slots covered.How many slots do you want to move (from 1 to 16384)? 1000 #这里输入要移动多少slotWhat is the receiving node ID? 0b7e0d5337e87ac7b59bba4c1248e5c9e8d1905e #这里输入目标节点的idPlease enter all the source node IDs. Type 'all' to use all the nodes as source nodes for the hash slots. Type 'done' once you entered all the source nodes IDs.Source node #1:all #将所有node都当成源节点 ... Moving slot 4309 from ec964a7c7cd53b986f54318a190c1426fc53a5fa Moving slot 4310 from ec964a7c7cd53b986f54318a190c1426fc53a5fa Moving slot 4311 from ec964a7c7cd53b986f54318a190c1426fc53a5fa Moving slot 4312 from ec964a7c7cd53b986f54318a190c1426fc53a5fa Moving slot 4313 from ec964a7c7cd53b986f54318a190c1426fc53a5faDo you want to proceed with the proposed reshard plan (yes/no)? yes #确认执行
注:第一个交互询问,填写多少slot移动时,要好好想想,如果填成16384,则将所有slot都移动到一个固定节点上,会导致更加不均衡!建议每次移动500~1000,这样对线上的影响比较小。
另外在填写source node时,除了all之外,还可以直接填写源节点的id,即:
[OK] All 16384 slots covered.How many slots do you want to move (from 1 to 16384)? 300What is the receiving node ID? 0b7e0d5337e87ac7b59bba4c1248e5c9e8d1905ePlease enter all the source node IDs. Type 'all' to use all the nodes as source nodes for the hash slots. Type 'done' once you entered all the source nodes IDs.Source node #1:226d1af3c95bf0798ea9fed86373b89347f889da #这里填写源节点的idSource node #2:done #这里输入done表示,不再继续添加源节点了
reshard可以多次操作,直到达到期望的分布为止(注:个人觉得redis的reshard这里有点麻烦,要移动多少slot需要人工计算,如果能提供一个参数之类,让16384个slot自动平均分配就好了),调整完成后,可以再看看分布情况:
➜ src ./redis-trib.rb info 127.0.0.1:7000127.0.0.1:7000 (0b7e0d53...) -> 4 keys | 4072 slots | 0 slaves.127.0.0.1:7001 (e0e8dfdd...) -> 5 keys | 4099 slots | 0 slaves.127.0.0.1:7006 (226d1af3...) -> 5 keys | 4132 slots | 4 slaves.127.0.0.1:7002 (ec964a7c...) -> 3 keys | 4081 slots | 0 slaves.[OK] 17 keys in 4 masters.0.00 keys per slot on average.
十一、删除节点del-node
既然有扩容,就会有反向需求,某些节点不再需要时,可以用del-node删除,比如刚才我一阵乱倒腾后,发现7006已经有4个slave了,而其它master一个slave都没有,这明显不合理。
删除节点命令:
./redis-trib.rb del-node 127.0.0.1:7006 88e16f91609c03277f2ee6ce5285932f58c221c1
del-node后面的ip:port只要是cluster中有效节点即可,最后一个参数为目标节点的id,注意:只有slave节点和空的master节点可以删除,如果master非空,先用reshard把上面的slot移动到其它node后再删除,如果有一组master-slave节点,将master上所有slot移到其它节点,然后将master删除,剩下的slave会另寻他主,变成其它master的slave。
另外:删除节点的含义,不仅仅是从cluster中将这个节点移除,还会直接将目标节点的redis服务停止。
参考文章: