Redis热迁移实战分享

jtyhrew222

Redis cluster使用slots来分配集群中的资源，因此官方提供了热迁移slots的方案，以便于迁移redis cluster节点中的信息。此方案不仅可以使用于节点迁移，也可以根据资源的不同，配置不同的slots数量。
集群原环境：
主机IP： 192.168.170.101
集群节点端口：10001-10006
集群当前主备关系：

迁移需求：现在有节点redis007,需要将redis002替换掉
迁移步骤：

• 首先验证环境当前的配置：
  
  

由上可以看出节点信息如下：
10001M<-10004S  slots=5461
10002M<-10005S      slots=5462
10003M<-10006S      slots=5461

• 现在来插入10W条数据
  
  

from rediscluster import StrictRedisCluster
redis_nodes = [
        {"host": "192.168.170.101", "port": "10001"},
        {"host": "192.168.170.101", "port": "10002"},
        {"host": "192.168.170.101", "port": "10003"},
        {"host": "192.168.170.101", "port": "10004"},
        {"host": "192.168.170.101", "port": "10005"},
        {"host": "192.168.170.101", "port": "10006"}
    ]
redis_conn = StrictRedisCluster(startup_nodes=redis_nodes, decode_responses=True)
for key in range(0, 100000):
    print key
    value = key
    key = 'zhang%s' % key
    redis_conn.set(key, value)

• 启动redis007，并将redis007加入集群环境：
  
  

./redis-server redis007.conf
[iyunv@lab001 redis]# redis-trib.rb add-node 192.168.170.101:10007 192.168.170.101:10001
>>> Adding node 192.168.170.101:10007 to cluster 192.168.170.101:10001
>>> Performing Cluster Check (using node 192.168.170.101:10001)
M: b30fdabd07a4bef611c160828965b91a1cdd462a 192.168.170.101:10001
   slots:0-5460 (5461 slots) master
   1 additional replica(s)
S: 74805544792cc22f09af941fa117b2974c9c0abf 192.168.170.101:10005
   slots: (0 slots) slave
   replicates 6036af6afc6567b74ce2fecc734a2d3908d561d1
M: 9c6a98926b23585e2eeb49ae17cc77521041d3ed 192.168.170.101:10003
   slots:10923-16383 (5461 slots) master
   1 additional replica(s)
S: 57a2e7313b952b38e32d6a1030d4812ea4decb5f 192.168.170.101:10006
   slots: (0 slots) slave
   replicates 9c6a98926b23585e2eeb49ae17cc77521041d3ed
S: 58b7b25501fdba04aef9b8cb47edef89092f10a7 192.168.170.101:10004
   slots: (0 slots) slave
   replicates b30fdabd07a4bef611c160828965b91a1cdd462a
M: 6036af6afc6567b74ce2fecc734a2d3908d561d1 192.168.170.101:10002
   slots:5461-10922 (5462 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.
>>> Send CLUSTER MEET to node 192.168.170.101:10007 to make it join the cluster.
[OK] New node added correctly.

• 查看当前节点状态信息
  
  

[iyunv@lab001 redis]# redis-trib.rb check 192.168.170.101:10001
>>> Performing Cluster Check (using node 192.168.170.101:10001)
M: b30fdabd07a4bef611c160828965b91a1cdd462a 192.168.170.101:10001
   slots:0-5460 (5461 slots) master
   1 additional replica(s)
M: 7c1a0acf84b2cf08876036b07518fb8f2ea9b8f8 192.168.170.101:10007
   slots: (0 slots) master
   0 additional replica(s)
S: 74805544792cc22f09af941fa117b2974c9c0abf 192.168.170.101:10005
   slots: (0 slots) slave
   replicates 6036af6afc6567b74ce2fecc734a2d3908d561d1
M: 9c6a98926b23585e2eeb49ae17cc77521041d3ed 192.168.170.101:10003
   slots:10923-16383 (5461 slots) master
   1 additional replica(s)
S: 57a2e7313b952b38e32d6a1030d4812ea4decb5f 192.168.170.101:10006
   slots: (0 slots) slave
   replicates 9c6a98926b23585e2eeb49ae17cc77521041d3ed
S: 58b7b25501fdba04aef9b8cb47edef89092f10a7 192.168.170.101:10004
   slots: (0 slots) slave
   replicates b30fdabd07a4bef611c160828965b91a1cdd462a
M: 6036af6afc6567b74ce2fecc734a2d3908d561d1 192.168.170.101:10002
   slots:5461-10922 (5462 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.

• 由上可以发现，redis007添加到节点当中，但当前未分配任何slots，这一步把redis002的slots完全迁移到redis007
  
  

由于过程较长，这里不再输出全部过程
[iyunv@lab001 redis]# redis-trib.rb reshard --from 6036af6afc6567b74ce2fecc734a2d3908d561d1 --to 7c1a0acf84b2cf08876036b07518fb8f2ea9b8f8 --slots 5462 --yes 192.168.170.101:10001
………………
Moving slot 10911 from 192.168.170.101:10002 to 192.168.170.101:10007: .....
Moving slot 10912 from 192.168.170.101:10002 to 192.168.170.101:10007: ..........
Moving slot 10913 from 192.168.170.101:10002 to 192.168.170.101:10007: ......
Moving slot 10914 from 192.168.170.101:10002 to 192.168.170.101:10007: .......
Moving slot 10915 from 192.168.170.101:10002 to 192.168.170.101:10007: .....
Moving slot 10916 from 192.168.170.101:10002 to 192.168.170.101:10007: .....
Moving slot 10917 from 192.168.170.101:10002 to 192.168.170.101:10007: ...
Moving slot 10918 from 192.168.170.101:10002 to 192.168.170.101:10007: ........
Moving slot 10919 from 192.168.170.101:10002 to 192.168.170.101:10007: .....
Moving slot 10920 from 192.168.170.101:10002 to 192.168.170.101:10007: .......
Moving slot 10921 from 192.168.170.101:10002 to 192.168.170.101:10007: ..
Moving slot 10922 from 192.168.170.101:10002 to 192.168.170.101:10007: ..........

• 再次查看当前节点信息
  
  

[iyunv@lab001 redis]# redis-trib.rb check 192.168.170.101:10001
>>> Performing Cluster Check (using node 192.168.170.101:10001)
M: b30fdabd07a4bef611c160828965b91a1cdd462a 192.168.170.101:10001
   slots:0-5460 (5461 slots) master
   1 additional replica(s)
M: 7c1a0acf84b2cf08876036b07518fb8f2ea9b8f8 192.168.170.101:10007
   slots:5461-10922 (5462 slots) master
   1 additional replica(s)
S: 74805544792cc22f09af941fa117b2974c9c0abf 192.168.170.101:10005
   slots: (0 slots) slave
   replicates 7c1a0acf84b2cf08876036b07518fb8f2ea9b8f8
M: 9c6a98926b23585e2eeb49ae17cc77521041d3ed 192.168.170.101:10003
   slots:10923-16383 (5461 slots) master
   1 additional replica(s)
S: 57a2e7313b952b38e32d6a1030d4812ea4decb5f 192.168.170.101:10006
   slots: (0 slots) slave
   replicates 9c6a98926b23585e2eeb49ae17cc77521041d3ed
S: 58b7b25501fdba04aef9b8cb47edef89092f10a7 192.168.170.101:10004
   slots: (0 slots) slave
   replicates b30fdabd07a4bef611c160828965b91a1cdd462a
M: 6036af6afc6567b74ce2fecc734a2d3908d561d1 192.168.170.101:10002
   slots: (0 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.
由此可以发现，redis002的所有slots迁移到redis007完毕，并且redis005自动将主从关系从002转移到了007

• 最后来验证下数据的完整性
  
  

from rediscluster import StrictRedisCluster

redis_nodes = [

        {"host": "192.168.170.101", "port": "10001"},

        {"host": "192.168.170.101", "port": "10002"},

        {"host": "192.168.170.101", "port": "10003"},

        {"host": "192.168.170.101", "port": "10004"},

        {"host": "192.168.170.101", "port": "10005"},

        {"host": "192.168.170.101", "port": "10006"}

redis_conn = StrictRedisCluster(startup_nodes=redis_nodes, decode_responses=True)

for key in range(0, 100000):
    #value = key
    key = 'zhang%s' % key
    #redis_conn.set(key, value)
    print redis_conn.get(key)

执行完毕，数据较验完成，数据完成完整迁移

• 最后，将redis002从集群中移除
  
  

[iyunv@lab001 redis]# redis-trib.rb del-node 192.168.170.101:10001 6036af6afc6567b74ce2fecc734a2d3908d561d1
>>> Removing node 6036af6afc6567b74ce2fecc734a2d3908d561d1 from cluster 192.168.170.101:10001
>>> Sending CLUSTER FORGET messages to the cluster...
>>> SHUTDOWN the node.
[2]   Done                    ./redis001/bin/redis-server redis00$i/redis00$i.conf

• 再次查看集群状态
  
  

[iyunv@lab001 redis]# redis-trib.rb check 192.168.170.101:10001
>>> Performing Cluster Check (using node 192.168.170.101:10001)
M: b30fdabd07a4bef611c160828965b91a1cdd462a 192.168.170.101:10001
   slots:0-5460 (5461 slots) master
   1 additional replica(s)
M: 7c1a0acf84b2cf08876036b07518fb8f2ea9b8f8 192.168.170.101:10007
   slots:5461-10922 (5462 slots) master
   1 additional replica(s)
S: 74805544792cc22f09af941fa117b2974c9c0abf 192.168.170.101:10005
   slots: (0 slots) slave
   replicates 7c1a0acf84b2cf08876036b07518fb8f2ea9b8f8
M: 9c6a98926b23585e2eeb49ae17cc77521041d3ed 192.168.170.101:10003
   slots:10923-16383 (5461 slots) master
   1 additional replica(s)
S: 57a2e7313b952b38e32d6a1030d4812ea4decb5f 192.168.170.101:10006
   slots: (0 slots) slave
   replicates 9c6a98926b23585e2eeb49ae17cc77521041d3ed
S: 58b7b25501fdba04aef9b8cb47edef89092f10a7 192.168.170.101:10004
   slots: (0 slots) slave
   replicates b30fdabd07a4bef611c160828965b91a1cdd462a
[OK] All nodes agree about slots configuration.
>>> Check for open slots...
>>> Check slots coverage...
[OK] All 16384 slots covered.
[iyunv@lab001 redis]# redis-trib.rb info 192.168.170.101:10001
192.168.170.101:10001 (b30fdabd...) -> 33357 keys | 5461 slots | 1 slaves.
192.168.170.101:10007 (7c1a0acf...) -> 33283 keys | 5462 slots | 1 slaves.
192.168.170.101:10003 (9c6a9892...) -> 33360 keys | 5461 slots | 1 slaves.
[OK] 100000 keys in 3 masters.
6.10 keys per slot on average.
总结：
1. slots迁移在日常工作中，通常用于节点迁移、扩展，也可以根据服务器本身的压力将slots迁移至性能较优的服务器
2. slots迁移过程当中，不影响数据的读写，这点已经做过实际的测试。
3. 迁移过程当中，应用配置应当至少包含一个集群master节点信息，否则有可能会造成数据访问异常的情况。迁移完成并配置同步更新至最新后，方可删除旧的空slots节点。

colxiaoyu

多谢分享

sdylag