文章目錄
- redis叢集分片存儲
-
-
- 為什麼要分片存儲
- 官方叢集方案
- 搭建叢集
- 叢集關心的問題
- Java用戶端代碼
- 非官方叢集方案
-
redis叢集分片存儲
為什麼要分片存儲
假設公司使用者有3千萬,使用者基本資訊緩存到redis中,需要記憶體10G,如何設計redis的緩存架構?
- 3千萬使用者,各種業務場景對使用者資訊的通路量很大,單台redis執行個體的讀寫瓶頸凸顯。
- 單redis執行個體管理10G記憶體,必然影響處理效率。
- redis的記憶體需求可能超過機器的最大記憶體,一台機器不夠用。
是以就有了分片存儲的技術。
官方叢集方案
redis cluster是Redis的分布式叢集解決方案,在3.0版本推出後有效地解決了redis分布式方面的需求,實作了資料在多個redis節點之間自動分片,故障自動轉移,擴容機制等功能。
搭建叢集
-
環境資訊
centos6,redis5。
-
叢集的資訊
我會在192.168.1.120,192.168.1.122,192.168.1.125這三台機器上分别啟動兩個redis執行個體,端口分别是6379和6380。
- 搭建叢集前確定每台機器上都已經安裝過redis,我的每台機器上的redis安裝目錄為:/var/redis-5.0.5。
- 每台機器上準備兩個redis配置檔案,一個redis_6379.conf,一個redis_6380.conf。
# 配置檔案進行了精簡,完整配置可自行和官方提供的完整conf檔案進行對照。端口号自行對應修改 #背景啟動的意思 daemonize yes #端口号 port 6381 # IP綁定,redis不建議對公網開放,直接綁定0.0.0.0沒毛病 bind 0.0.0.0 # redis資料檔案存放的目錄 dir /usr/local/redis/data # 開啟AOF appendonly yes # 開啟叢集 cluster-enabled yes # 會自動生成在上面配置的dir目錄下 cluster-config-file nodes-6381.conf cluster-node-timeout 5000 # 這個檔案會自動生成 pidfile /var/run/redis_6381.pid
- 啟動6個redis執行個體,在三台機器的redis的安裝目錄的src下,分别運作如下兩條指令。
使用ps -ef | grep redis指令檢視叢集是否啟動正常。./redis-server ../redis_6379.conf ./redis-server ../redis_6380.conf
-
建立cluster,在redis安裝目錄的src下執行下面的指令。
在已經使用安裝并使用過的redis伺服器上運作此指令可能會抛出如下錯誤。參考https://blog.csdn.net/XIANZHIXIANZHIXIAN/article/details/82777767解決。
運作成功後會讓你确認叢集資訊。
[root@spark src]# ./redis-cli --cluster create 192.168.1.120:6379 192.168.1.120:6380 192.168.1.122:6379 192.168.1.122:6380 192.168.1.125:6379 192.168.1.125:6380 --cluster-replicas 1 >>> Performing hash slots allocation on 6 nodes... Master[0] -> Slots 0 - 5460 Master[1] -> Slots 5461 - 10922 Master[2] -> Slots 10923 - 16383 Adding replica 192.168.1.122:6380 to 192.168.1.120:6379 Adding replica 192.168.1.125:6380 to 192.168.1.122:6379 Adding replica 192.168.1.120:6380 to 192.168.1.125:6379 M: fd36242496e7a04d883857891ce425a0fc948783 192.168.1.120:6379 slots:[0-5460] (5461 slots) master S: 9ffd15c5674f4ff06ff4dca186fbf919a992c87b 192.168.1.120:6380 replicates ace69b140e315ce90ed6c19de1aaec73032b1ca6 M: 77e054f92028a383b69147f0c21e680ae664bb6b 192.168.1.122:6379 slots:[5461-10922] (5462 slots) master S: 1892d6272a759f454548ce97b0d09c580f687ef3 192.168.1.122:6380 replicates fd36242496e7a04d883857891ce425a0fc948783 M: ace69b140e315ce90ed6c19de1aaec73032b1ca6 192.168.1.125:6379 slots:[10923-16383] (5461 slots) master S: a4c630716cd013da8a2e2ce89a26127c57ff356e 192.168.1.125:6380 replicates 77e054f92028a383b69147f0c21e680ae664bb6b Can I set the above configuration? (type 'yes' to accept): y [root@spark src]# ./redis-cli -c -h 192.168.1.120 -p 6379 cluster nodes fd36242496e7a04d883857891ce425a0fc948783 :6379@16379 myself,master - 0 0 0 connected [root@spark src]# ./redis-cli -h 192.168.1.120 -p 6379 192.168.1.120:6379> cluster nodes fd36242496e7a04d883857891ce425a0fc948783 :6379@16379 myself,master - 0 0 0 connected 192.168.1.120:6379> [root@spark src]# ./redis-cli --cluster create 192.168.1.120:6379 192.168.1.120:6380 192.168.1.122:6379 192.168.1.122:6380 192.168.1.125:6379 192.168.1.125:6380 --cluster-replicas 1 >>> Performing hash slots allocation on 6 nodes... Master[0] -> Slots 0 - 5460 Master[1] -> Slots 5461 - 10922 Master[2] -> Slots 10923 - 16383 Adding replica 192.168.1.122:6380 to 192.168.1.120:6379 Adding replica 192.168.1.125:6380 to 192.168.1.122:6379 Adding replica 192.168.1.120:6380 to 192.168.1.125:6379 M: fd36242496e7a04d883857891ce425a0fc948783 192.168.1.120:6379 slots:[0-5460] (5461 slots) master S: 9ffd15c5674f4ff06ff4dca186fbf919a992c87b 192.168.1.120:6380 replicates ace69b140e315ce90ed6c19de1aaec73032b1ca6 M: 77e054f92028a383b69147f0c21e680ae664bb6b 192.168.1.122:6379 slots:[5461-10922] (5462 slots) master S: 1892d6272a759f454548ce97b0d09c580f687ef3 192.168.1.122:6380 replicates fd36242496e7a04d883857891ce425a0fc948783 M: ace69b140e315ce90ed6c19de1aaec73032b1ca6 192.168.1.125:6379 slots:[10923-16383] (5461 slots) master S: a4c630716cd013da8a2e2ce89a26127c57ff356e 192.168.1.125:6380 replicates 77e054f92028a383b69147f0c21e680ae664bb6b 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 192.168.1.120:6379) M: fd36242496e7a04d883857891ce425a0fc948783 192.168.1.120:6379 slots:[0-5460] (5461 slots) master 1 additional replica(s) M: ace69b140e315ce90ed6c19de1aaec73032b1ca6 192.168.1.125:6379 slots:[10923-16383] (5461 slots) master 1 additional replica(s) S: 9ffd15c5674f4ff06ff4dca186fbf919a992c87b 192.168.1.120:6380 slots: (0 slots) slave replicates ace69b140e315ce90ed6c19de1aaec73032b1ca6 S: a4c630716cd013da8a2e2ce89a26127c57ff356e 192.168.1.125:6380 slots: (0 slots) slave replicates 77e054f92028a383b69147f0c21e680ae664bb6b S: 1892d6272a759f454548ce97b0d09c580f687ef3 192.168.1.122:6380 slots: (0 slots) slave replicates fd36242496e7a04d883857891ce425a0fc948783 M: 77e054f92028a383b69147f0c21e680ae664bb6b 192.168.1.122:6379 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.
-
叢集檢驗和測試
檢視所有節點資訊。
[root@spark src]# ./redis-cli -c -h 192.168.1.120 -p 6379 cluster nodes ace69b140e315ce90ed6c19de1aaec73032b1ca6 192.168.1.125:6379@16379 master - 0 1561421002080 5 connected 10923-16383 9ffd15c5674f4ff06ff4dca186fbf919a992c87b 192.168.1.120:6380@16380 slave ace69b140e315ce90ed6c19de1aaec73032b1ca6 0 1561421000463 5 connected fd36242496e7a04d883857891ce425a0fc948783 192.168.1.120:6379@16379 myself,master - 0 1561421000000 1 connected 0-5460 a4c630716cd013da8a2e2ce89a26127c57ff356e 192.168.1.125:6380@16380 slave 77e054f92028a383b69147f0c21e680ae664bb6b 0 1561421000160 6 connected 1892d6272a759f454548ce97b0d09c580f687ef3 192.168.1.122:6380@16380 slave fd36242496e7a04d883857891ce425a0fc948783 0 1561420999891 4 connected 77e054f92028a383b69147f0c21e680ae664bb6b 192.168.1.122:6379@16379 master - 0 1561421001081 3 connected 5461-10922 # 開啟用戶端不加-c選項,如果目前資料不應該在此執行個體上會傳回MOVED ./redis-cli -h 192.168.1.120 -p 6379 192.168.1.120:6379> set a 1 (error) MOVED 15495 192.168.1.125:6379 192.168.1.120:6379> # 開啟的話會為目前的資料自動跳轉到對應的執行個體 [root@spark src]# ./redis-cli -h 192.168.1.120 -p 6379 192.168.1.120:6379> set a 1 (error) MOVED 15495 192.168.1.125:6379 192.168.1.120:6379> [root@spark src]# ./redis-cli -c -h 192.168.1.120 -p 6379 192.168.1.120:6379> set a 1 -> Redirected to slot [15495] located at 192.168.1.125:6379 OK 192.168.1.125:6379> get a "1" 192.168.1.125:6379> set hello tony -> Redirected to slot [866] located at 192.168.1.120:6379 OK 192.168.1.120:6379> get a -> Redirected to slot [15495] located at 192.168.1.125:6379 "1" 192.168.1.125:6379> cluster keyslot a (integer) 15495 192.168.1.125:6379>
- 叢集slot數量整理reshard
# 有時根據叢集中的機器的配置,我們可能希望有些性能高的機器上的槽多一些 # 就可以使用reshard,下面我們從192.168.1.120伺服器移動1000個slot到192.168.1.122 # 這台伺服器上 ./redis-cli --cluster reshard 192.168.1.120:6379 --cluster-from fd36242496e7a04d883857891ce425a0fc948783 --cluster-to 77e054f92028a383b69147f0c21e680ae664bb6b --cluster-slots 1000 --cluster-yes # 運作後會有如下輸出 Moving slot 714 from 192.168.1.120:6379 to 192.168.1.122:6379: Moving slot 715 from 192.168.1.120:6379 to 192.168.1.122:6379: # 重新檢查叢集 ./redis-cli --cluster check 192.168.1.120:6379 M: fd36242496e7a04d883857891ce425a0fc948783 192.168.1.120:6379 slots:[1000-5460] (4461 slots) master 1 additional replica(s) M: ace69b140e315ce90ed6c19de1aaec73032b1ca6 192.168.1.125:6379 slots:[10923-16383] (5461 slots) master 1 additional replica(s) S: 9ffd15c5674f4ff06ff4dca186fbf919a992c87b 192.168.1.120:6380 slots: (0 slots) slave replicates ace69b140e315ce90ed6c19de1aaec73032b1ca6 S: a4c630716cd013da8a2e2ce89a26127c57ff356e 192.168.1.125:6380 slots: (0 slots) slave replicates 77e054f92028a383b69147f0c21e680ae664bb6b S: 1892d6272a759f454548ce97b0d09c580f687ef3 192.168.1.122:6380 slots: (0 slots) slave replicates fd36242496e7a04d883857891ce425a0fc948783 M: 77e054f92028a383b69147f0c21e680ae664bb6b 192.168.1.122:6379 slots:[0-999],[5461-10922] (6462 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. # 可以看出從192.168.1.120移動了1000個slot到192.168.1.122
- 測試自動故障轉移
# cluster叢集不保證資料一緻,資料也可能丢失 # 首先是運作用戶端不斷的寫入或讀取資料,以便能夠發現問題 # 然後是模拟節點故障:找一個主節點關閉,主從故障切換的過程中,這個時間端的操作,用戶端而言,隻能是失敗 # 官方描述 https://redis.io/topics/cluster-spec There is always a window of time when it is possible to lose writes during partitions. 分區的時間視窗内總是有可能丢失寫操作。 # 将192.168.1.120:6379這個執行個體停掉,192.168.1.122:6380成為一個master ./redis-cli -c -h 192.168.1.120 -p 6379 192.168.1.120:6379> shutdown not connected> set a 1 [expiration EX seconds|PX milliseconds] [NX|XX] [root@spark src]# ./redis-cli --cluster check 192.168.1.120:6380 Could not connect to Redis at 192.168.1.120:6379: Connection refused 192.168.1.125:6379 (ace69b14...) -> 1 keys | 5461 slots | 1 slaves. 192.168.1.122:6380 (1892d627...) -> 0 keys | 4461 slots | 0 slaves. 192.168.1.122:6379 (77e054f9...) -> 1 keys | 6462 slots | 1 slaves. [OK] 2 keys in 3 masters. 0.00 keys per slot on average. >>> Performing Cluster Check (using node 192.168.1.120:6380) S: 9ffd15c5674f4ff06ff4dca186fbf919a992c87b 192.168.1.120:6380 slots: (0 slots) slave replicates ace69b140e315ce90ed6c19de1aaec73032b1ca6 M: ace69b140e315ce90ed6c19de1aaec73032b1ca6 192.168.1.125:6379 slots:[10923-16383] (5461 slots) master 1 additional replica(s) M: 1892d6272a759f454548ce97b0d09c580f687ef3 192.168.1.122:6380 slots:[1000-5460] (4461 slots) master M: 77e054f92028a383b69147f0c21e680ae664bb6b 192.168.1.122:6379 slots:[0-999],[5461-10922] (6462 slots) master 1 additional replica(s) S: a4c630716cd013da8a2e2ce89a26127c57ff356e 192.168.1.125:6380 slots: (0 slots) slave replicates 77e054f92028a383b69147f0c21e680ae664bb6b [OK] All nodes agree about slots configuration. >>> Check for open slots... >>> Check slots coverage... [OK] All 16384 slots covered. [root@spark src]# # 檢視此時的叢集資訊,192.168.1.120:6379下線了 [root@spark src]# ./redis-cli -c -h 192.168.1.120 -p 6380 192.168.1.120:6380> cluster nodes fd36242496e7a04d883857891ce425a0fc948783 192.168.1.120:6379@16379 master,fail - 1561422523649 1561422523348 1 disconnected ace69b140e315ce90ed6c19de1aaec73032b1ca6 192.168.1.125:6379@16379 master - 0 1561422691644 5 connected 10923-16383 1892d6272a759f454548ce97b0d09c580f687ef3 192.168.1.122:6380@16380 master - 0 1561422693669 8 connected 1000-5460 77e054f92028a383b69147f0c21e680ae664bb6b 192.168.1.122:6379@16379 master - 0 1561422693570 7 connected 0-999 5461-10922 9ffd15c5674f4ff06ff4dca186fbf919a992c87b 192.168.1.120:6380@16380 myself,slave ace69b140e315ce90ed6c19de1aaec73032b1ca6 0 1561422687000 2 connected a4c630716cd013da8a2e2ce89a26127c57ff356e 192.168.1.125:6380@16380 slave 77e054f92028a383b69147f0c21e680ae664bb6b 0 1561422692148 7 connected 192.168.1.120:6380> # 如果再把192.168.1.120:6379啟動起來,其會議salve的身份加入叢集 [root@spark src]# ./redis-cli --cluster check 192.168.1.120:6380 192.168.1.125:6379 (ace69b14...) -> 1 keys | 5461 slots | 1 slaves. 192.168.1.122:6380 (1892d627...) -> 0 keys | 4461 slots | 1 slaves. 192.168.1.122:6379 (77e054f9...) -> 1 keys | 6462 slots | 1 slaves. [OK] 2 keys in 3 masters. 0.00 keys per slot on average. >>> Performing Cluster Check (using node 192.168.1.120:6380) S: 9ffd15c5674f4ff06ff4dca186fbf919a992c87b 192.168.1.120:6380 slots: (0 slots) slave replicates ace69b140e315ce90ed6c19de1aaec73032b1ca6 S: fd36242496e7a04d883857891ce425a0fc948783 192.168.1.120:6379 slots: (0 slots) slave replicates 1892d6272a759f454548ce97b0d09c580f687ef3 M: ace69b140e315ce90ed6c19de1aaec73032b1ca6 192.168.1.125:6379 slots:[10923-16383] (5461 slots) master 1 additional replica(s) M: 1892d6272a759f454548ce97b0d09c580f687ef3 192.168.1.122:6380 slots:[1000-5460] (4461 slots) master 1 additional replica(s) M: 77e054f92028a383b69147f0c21e680ae664bb6b 192.168.1.122:6379 slots:[0-999],[5461-10922] (6462 slots) master 1 additional replica(s) S: a4c630716cd013da8a2e2ce89a26127c57ff356e 192.168.1.125:6380 slots: (0 slots) slave replicates 77e054f92028a383b69147f0c21e680ae664bb6b [OK] All nodes agree about slots configuration. >>> Check for open slots... >>> Check slots coverage... [OK] All 16384 slots covered.
- 手動故障轉移
# 可能某個節點需要維護(機器下線,硬體更新,系統版本調整等場景),需要手動的實作轉移 # 在slave節點上執行指令 CLUSTER FAILOVER # 相比較于自動故障轉移而言,手動故障轉移不會丢失資料
# 上面我們手動将192.168.1.120:6379上的redis服務停掉再起來後,它以slave的身份加入 # 叢集,現在我們想讓它重新成為master [root@spark src]# ./redis-cli -c -h 192.168.1.120 -p 6379 # 檢視目前叢集資訊 192.168.1.120:6379> cluster nodes 1892d6272a759f454548ce97b0d09c580f687ef3 192.168.1.122:6380@16380 master - 0 1561423733576 8 connected 1000-5460 ace69b140e315ce90ed6c19de1aaec73032b1ca6 192.168.1.125:6379@16379 slave 9ffd15c5674f4ff06ff4dca186fbf919a992c87b 0 1561423734623 9 connected a4c630716cd013da8a2e2ce89a26127c57ff356e 192.168.1.125:6380@16380 slave 77e054f92028a383b69147f0c21e680ae664bb6b 0 1561423734084 7 connected 9ffd15c5674f4ff06ff4dca186fbf919a992c87b 192.168.1.120:6380@16380 master - 0 1561423735107 9 connected 10923-16383 77e054f92028a383b69147f0c21e680ae664bb6b 192.168.1.122:6379@16379 master - 0 1561423734620 7 connected 0-999 5461-10922 fd36242496e7a04d883857891ce425a0fc948783 192.168.1.120:6379@16379 myself,slave 1892d6272a759f454548ce97b0d09c580f687ef3 0 1561423733000 1 connected # 執行手動故障轉移 192.168.1.120:6379> cluster failover OK # 檢查手動故障轉移效果,發現192.168.1.120:6379再次成為master 192.168.1.120:6379> cluster nodes 1892d6272a759f454548ce97b0d09c580f687ef3 192.168.1.122:6380@16380 slave fd36242496e7a04d883857891ce425a0fc948783 0 1561423750936 10 connected ace69b140e315ce90ed6c19de1aaec73032b1ca6 192.168.1.125:6379@16379 slave 9ffd15c5674f4ff06ff4dca186fbf919a992c87b 0 1561423751000 9 connected a4c630716cd013da8a2e2ce89a26127c57ff356e 192.168.1.125:6380@16380 slave 77e054f92028a383b69147f0c21e680ae664bb6b 0 1561423750000 7 connected 9ffd15c5674f4ff06ff4dca186fbf919a992c87b 192.168.1.120:6380@16380 master - 0 1561423751955 9 connected 10923-16383 77e054f92028a383b69147f0c21e680ae664bb6b 192.168.1.122:6379@16379 master - 0 1561423750432 7 connected 0-999 5461-10922 fd36242496e7a04d883857891ce425a0fc948783 192.168.1.120:6379@16379 myself,master - 0 1561423751000 10 connected 1000-5460 192.168.1.120:6379>
- 擴容
# 我們再在192.168.1.120這台伺服器上準備一個redis配置檔案,redis_6381.conf # 啟動新節點 ./redis-server ../redis_6381.conf # 以master身份加入叢集 ./redis-cli --cluster add-node 192.168.1.120:6381 192.168.1.120:6379 >>> Adding node 192.168.1.120:6381 to cluster 192.168.1.120:6379 >>> Performing Cluster Check (using node 192.168.1.120:6379) M: fd36242496e7a04d883857891ce425a0fc948783 192.168.1.120:6379 slots:[1000-5460] (4461 slots) master 1 additional replica(s) S: 1892d6272a759f454548ce97b0d09c580f687ef3 192.168.1.122:6380 slots: (0 slots) slave replicates fd36242496e7a04d883857891ce425a0fc948783 S: ace69b140e315ce90ed6c19de1aaec73032b1ca6 192.168.1.125:6379 slots: (0 slots) slave replicates 9ffd15c5674f4ff06ff4dca186fbf919a992c87b S: a4c630716cd013da8a2e2ce89a26127c57ff356e 192.168.1.125:6380 slots: (0 slots) slave replicates 77e054f92028a383b69147f0c21e680ae664bb6b M: 9ffd15c5674f4ff06ff4dca186fbf919a992c87b 192.168.1.120:6380 slots:[10923-16383] (5461 slots) master 1 additional replica(s) M: 77e054f92028a383b69147f0c21e680ae664bb6b 192.168.1.122:6379 slots:[0-999],[5461-10922] (6462 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 192.168.1.120:6381 to make it join the cluster. [OK] New node added correctly. # 檢視此時的叢集資訊,新建立的redis執行個體以master身份加入叢集 [root@spark src]# ./redis-cli -c -h 192.168.1.120 -p 6379 192.168.1.120:6379> cluster nodes 12593a9cfbdffd272547926b625acb09d306c7fc 192.168.1.120:6381@16381 master - 0 1561424739585 0 connected 1892d6272a759f454548ce97b0d09c580f687ef3 192.168.1.122:6380@16380 slave fd36242496e7a04d883857891ce425a0fc948783 0 1561424739488 10 connected ace69b140e315ce90ed6c19de1aaec73032b1ca6 192.168.1.125:6379@16379 slave 9ffd15c5674f4ff06ff4dca186fbf919a992c87b 0 1561424740528 9 connected a4c630716cd013da8a2e2ce89a26127c57ff356e 192.168.1.125:6380@16380 slave 77e054f92028a383b69147f0c21e680ae664bb6b 0 1561424740529 7 connected 9ffd15c5674f4ff06ff4dca186fbf919a992c87b 192.168.1.120:6380@16380 master - 0 1561424740628 9 connected 10923-16383 77e054f92028a383b69147f0c21e680ae664bb6b 192.168.1.122:6379@16379 master - 0 1561424740019 7 connected 0-999 5461-10922 fd36242496e7a04d883857891ce425a0fc948783 192.168.1.120:6379@16379 myself,master - 0 1561424738000 10 connected 1000-5460 192.168.1.120:6379> # 本質就是發送一個新節點通過CLUSTER MEET指令加入叢集 # 注意新加入的節點是沒有配置設定slot的,它不會存儲資料,我們使用之前需要給它配置設定slot # 參考第8點 # 以slave身份加入叢集 ./redis-cli --cluster add-node 192.168.1.120:6381 192.168.1.120:6379 --cluster-slave # 還可以手動指定要以slave身份加入叢集的節點的master節點,否則就選擇一個slave數量 # 最少的master ./redis-cli --cluster add-node 192.168.1.120:6381 192.168.1.120:6379 --cluster-slave --cluster-master-id <node-id> # 将空master轉換為slave cluster replicate <master-node-id>
- 縮容
# 注意:删除master的時候要把資料清空或者配置設定給其他主節點 # 下面我們将新加入的192.168.1.120:6381節點删除 [root@spark src]# ./redis-cli --cluster del-node 192.168.1.120:6381 12593a9cfbdffd272547926b625acb09d306c7fc >>> Removing node 12593a9cfbdffd272547926b625acb09d306c7fc from cluster 192.168.1.120:6381 >>> Sending CLUSTER FORGET messages to the cluster... >>> SHUTDOWN the node. # 檢視6381對應的redis程序是否還在 [root@spark src]# ps -ef | grep redis root 17475 17274 0 07:53 pts/1 00:00:34 ./redis-server 192.168.1.120:6380 [cluster] root 17503 17228 0 08:33 pts/0 00:00:18 ./redis-server 192.168.1.120:6379 [cluster] root 17571 17303 0 09:17 pts/3 00:00:00 grep redis [root@spark src]#
-
叢集關心的問題
-
增加了slot槽位的計算,是不是比單機性能差?
共16384個槽位,slots槽計算方式公開的,HASH_SLOT = CRC(key) mod 16384。為了避免每次都需要伺服器計算重定向,優秀的java用戶端都實作了本地計算,并且緩存伺服器slots配置設定,有變動時再更新本地内容,進而避免了多次重定向帶來的性能損耗。
-
redis叢集大小,到底可以裝多少資料?
理論是可以做到16384個槽,每個槽對應一個執行個體,但是redis官方建議是最大1000個執行個體。
-
叢集節點間是怎麼通信的?
每個redis叢集節點都有一個額外的TCP端口,每個節點使用TCP連接配接與其他節點連接配接。檢測和故障轉移這些步驟基本和哨兵模式類似。
-
ask和moved重定向的差別。
重定向包括兩種情況:
- 若确定slot不屬于目前節點,redis會傳回moved。
- 若目前redis節點正在處理slot遷移,則代表此處請求對應的key暫時不在此節點,傳回ask,告訴用戶端本次請求重定向。
-
資料傾斜和通路傾斜的問題。
傾斜導緻叢集中部分節點資料多,壓力大。解決方案分為前期和後期:
- 前期是業務層面提前預測,哪些是熱點key,在設計的過程中規避。
- 後期是slot遷移,盡量将壓力分攤(slot調整有自動reblance,reshard和手動)。
- slot手動遷移怎麼做?
- 在遷移目的節點執行cluster setslot IMPORTING 指令,指明需要遷移的slot和遷移源節點。
- 在遷移源節點執行cluster setslot MIGRATING 指令,指明需要遷移的slot和遷移目的節點。
- 在遷移源節點執行cluster getkeysinslot擷取該slot的key清單。
- 在遷移源節點對每個key執行migrate指令,該指令會同步把該key遷移到目的節點。
- 在遷移源節點傳回執行cluster getkeysinslot指令,直到該slot的清單為空。
- 在遷移源節點和目的節點執行cluster setslot NODE ,完成遷移操作。
- 節點之間會交換資訊,傳遞的消息包括槽的資訊,帶來帶寬消耗。避免使用大的叢集,可以分多個叢集。
- Pub/Sub釋出訂閱機制,對叢集内任意一個節點執行publish釋出消息,這個消息會在叢集中進行傳播,其他節點接收到釋出的消息。
- 讀寫分離。
- redis-cluster預設所有從節點上的讀寫,都會重定向到key對接槽的主節點上。
- 可以通過readonly設定目前連接配接可讀,通過readwrite取消目前連接配接的可讀狀态。
- 主從節點依然存在資料不一緻的問題。
Java用戶端代碼
- 配置Bean。
@Configuration // 在cluster環境下生效 @Profile("a7_cluster") class ClusterAppConfig { @Bean public JedisConnectionFactory redisConnectionFactory() { System.out.println("加載cluster環境下的redis client配置"); RedisClusterConfiguration redisClusterConfiguration = new RedisClusterConfiguration(Arrays.asList( "192.168.1.120:6379", "192.168.1.120:6380", "192.168.1.122:6379", "192.168.1.122:6380", "192.168.1.125:6379", "192.168.1.125:6380" )); // 自适應叢集變化 return new JedisConnectionFactory(redisClusterConfiguration); } }
- 業務service。
@Service @Profile("a7_cluster") public class ClusterService { @Autowired private StringRedisTemplate template; public void set(String userId, String userInfo) { template.opsForValue().set(userId, userInfo); } }
- 單元測試
@RunWith(SpringJUnit4ClassRunner.class) @ContextConfiguration("classpath:applicationContext.xml") @ActiveProfiles("a7_cluster") // 設定profile // 叢集對于用戶端而言,基本是無感覺的 public class ClusterServiceTests { @Autowired ClusterService clusterService; @Test public void setTest() { clusterService.set("tony", "hahhhhh"); clusterService.set("a", "1"); clusterService.set("foo", "bar"); } // 測試cluster叢集故障時的反應 @Test public void failoverTest() { while (true) { try { long i = System.currentTimeMillis(); clusterService.set("tony", i + ""); // delay 10ms TimeUnit.MILLISECONDS.sleep(10); } catch (Exception e) { System.out.println(e.getMessage()); } } } } 運作setTest後,它實際上存儲在192.168.1.120:6380 ./redis-cli -c -h 192.168.1.120 -p 6379 192.168.1.120:6379> get tony -> Redirected to slot [14405] located at 192.168.1.120:6380 "hahhhhh" 192.168.1.120:6380> get tongy -> Redirected to slot [9429] located at 192.168.1.122:6379 (nil) 192.168.1.122:6379> get tony -> Redirected to slot [14405] located at 192.168.1.120:6380 "hahhhhh" 192.168.1.120:6380 運作failoverTest後,停掉192.168.1.120:6380 192.168.1.120:6380> shutdown not connected> 對用戶端無影響
非官方叢集方案
- Codis,豌豆莢開源。
- twemproxy,推特開源。