kubernetesでRedisClusterを構築してみた


概要

kubernetesでredis cluster構築してみた備忘録

https://github.com/kelseyhightower/kubernetes-redis-cluster
ほぼこちらをそのまま使っている

  • Redis server version、3.2.0
  • データの永続化はしない
  • redisクラスタ基本master×3にレプリカ用 slave×3
redis (loadbalancer) 
  |
  |- [ redis-1(master) ] - [redis-4(slave)]
  |- [ redis-2(master) ] - [redis-5(slave)]
  |- [ redis-3(master) ] - [redis-6(slave)]
  • maxmemory-policy は、デフォルトvolatile-lru
`volatile-lru` remove the key with an expire set using an LRU algorithm

セッション保存時に、有効期限を設定する想定

redis-1 ~ redis6もreplicasetでpod作成しているので設定を変えれば複数pod起動可能。
1podずつで以下構築と確認を行う。

ソース

構築

  • kubernetesでreplicasetとserviceでpodとloadbalancerを起動する
$ kubectl create configmap redis-conf --from-file=redis.conf 
$ kubectl create -f replicasets
$ kubectl create -f services
  • podをredis clusterに繋ぐ

redis cluster作成コマンド実行用podを起動する

$ kubectl run -i --tty ubuntu --image=ubuntu --restart=Never /bin/bash

必要なライブラリー類をインストールする

$ apt-get update
$ apt-get install ruby vim wget redis-tools
$ gem install redis
$ wget http://download.redis.io/redis-stable/src/redis-trib.rb
$ chmod 755 redis-trib.rb

redis clusterを作成する

$ ./redis-trib.rb create --replicas 1 \
  10.107.255.1:6379 \
  10.107.255.2:6379 \
  10.107.255.3:6379 \
  10.107.255.4:6379 \
  10.107.255.5:6379 \
  10.107.255.6:6379
>>> Creating cluster
>>> Performing hash slots allocation on 6 nodes...
Using 3 masters:
10.107.255.1:6379
10.107.255.2:6379
10.107.255.3:6379
Adding replica 10.107.255.4:6379 to 10.107.255.1:6379
Adding replica 10.107.255.5:6379 to 10.107.255.2:6379
Adding replica 10.107.255.6:6379 to 10.107.255.3:6379
M: d4dfb6366289f581e395428811af4bb6e1f07e2d 10.107.255.1:6379   10.104.1.34
   slots:0-5460 (5461 slots) master
M: 4b23c907ef408e5efd6225fffd11a6cdff5de337 10.107.255.2:6379   10.104.1.35
   slots:5461-10922 (5462 slots) master
M: 189ee8e87202326675adef4f507a80b5b5873369 10.107.255.3:6379   10.104.1.36
   slots:10923-16383 (5461 slots) master
S: 37b2b6b625d8bfba7c8997abfe894df099761937 10.107.255.4:6379   10.104.1.37
   replicates d4dfb6366289f581e395428811af4bb6e1f07e2d
S: f2229c73056a819f3cee6521d231e65ce1f4c180 10.107.255.5:6379   10.104.0.14
   replicates 4b23c907ef408e5efd6225fffd11a6cdff5de337
S: ad1aa4c73849622068c96976998f778673a2ce4c 10.107.255.6:6379   10.104.1.38
   replicates 189ee8e87202326675adef4f507a80b5b5873369
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 10.107.255.1:6379)
M: d4dfb6366289f581e395428811af4bb6e1f07e2d 10.107.255.1:6379
   slots:0-5460 (5461 slots) master
   1 additional replica(s)
M: 189ee8e87202326675adef4f507a80b5b5873369 10.104.1.36:6379
   slots:10923-16383 (5461 slots) master
   1 additional replica(s)
M: 4b23c907ef408e5efd6225fffd11a6cdff5de337 10.104.1.35:6379
   slots:5461-10922 (5462 slots) master
   1 additional replica(s)
S: ad1aa4c73849622068c96976998f778673a2ce4c 10.104.1.38:6379
   slots: (0 slots) slave
   replicates 189ee8e87202326675adef4f507a80b5b5873369
S: f2229c73056a819f3cee6521d231e65ce1f4c180 10.104.0.14:6379
   slots: (0 slots) slave
   replicates 4b23c907ef408e5efd6225fffd11a6cdff5de337
S: 37b2b6b625d8bfba7c8997abfe894df099761937 10.104.1.37:6379
   slots: (0 slots) slave
   replicates d4dfb6366289f581e395428811af4bb6e1f07e2d
[OK] All nodes agree about slots configuration.
>>> Check for open slots...
>>> Check slots coverage...
[OK] All 16384 slots covered.

確認

dockerを落としても他のredisノードから値が帰ってくる事を確認する

  • masterを1pod止めて、自動起動するか確認する

pod一覧

$ kubectl get pods --show-all
NAME            READY     STATUS    RESTARTS   AGE
redis-1-2pg4v   1/1       Running   0          56m
redis-2-fbzks   1/1       Running   0          56m
redis-3-4fs93   1/1       Running   0          56m
redis-4-pjxt1   1/1       Running   0          56m
redis-5-bt1wm   1/1       Running   0          56m
redis-6-qpvvl   1/1       Running   0          56m
ubuntu          1/1       Running   0          20m

redis-1(master)podを止める

$ kubectl stop pod redis-1-2pg4v
Command "stop" is deprecated, use "delete" instead.
pod "redis-1-2pg4v" deleted

pod一覧

$ kubectl get pods --show-all
NAME            READY     STATUS    RESTARTS   AGE
redis-1-b7blk   1/1       Running   0          3m  ← redis-1(master)のpodが新たに起動された
redis-2-fbzks   1/1       Running   0          1h
redis-3-4fs93   1/1       Running   0          1h
redis-4-pjxt1   1/1       Running   0          1h
redis-5-bt1wm   1/1       Running   0          1h
redis-6-qpvvl   1/1       Running   0          1h
ubuntu          1/1       Running   0          24m
  • dockerを止めても正常にデータ取得できるか
a12948@gke-satsuma-redis-default-pool-f99bcf1a-cft1 ~ $ docker ps -a | grep redis-server
4b1cac0124e9        redis:3.2.0-alpine                                                     "redis-server /etc/re"   8 minutes ago       Up 8 minutes                                   k8s_redis.638d2d73_redis-1-b7blk_default_2df04edf-e91c-11e6-999b-42010af00135_72d2a8a6
6fef91ca8ca5        redis:3.2.0-alpine                                                     "redis-server /etc/re"   5 hours ago         Exited (0) 8 minutes ago                       k8s_redis.638d2d73_redis-1-b7blk_default_2df04edf-e91c-11e6-999b-42010af00135_74b5a100
$ kubectl exec -it ubuntu sh
# redis-cli -h 10.107.244.37 -c
10.107.244.37:6379> get hoge
"fuga"
10.107.244.37:6379> get hogehoge
"fugafuga"

ノードを落としてもredisノードから値が帰ってくる事を確認する

!!container clusterのnodeが1つになってしまうと自動起動でいないpodができてしまうため、
少なくても2つのnodeが必要である。
nodeが1つ落ちてしまった時を考慮し、クラスタの設定を以下にする

  • node最小サイズ3
  • container cluster 自動スケーリング ON
$ kubectl get nodes
NAME                                           STATUS    AGE
gke-satsuma-redis-default-pool-f99bcf1a-cft1   Ready     2d
gke-satsuma-redis-default-pool-f99bcf1a-lj8d   Ready     9m

$ kubectl delete node gke-satsuma-redis-default-pool-f99bcf1a-cft1
node "gke-satsuma-redis-default-pool-f99bcf1a-cft1" deleted

$ kubectl get nodes
NAME                                           STATUS    AGE
gke-satsuma-redis-default-pool-f99bcf1a-lj8d   Ready     10m
$ kubectl exec -it ubuntu sh
# redis-cli -h 10.107.244.37 -c
10.107.244.37:6379> get hoge
-> Redirected to slot [1525] located at 10.104.0.20:6379
"fuga"
10.104.0.20:6379> get hogehoge
"fugafuga"
10.104.0.20:6379> get hoge
"fuga"
10.104.0.20:6379> get hogehoge
"fugafuga"

serviceに紐付けて全redisノードにランダムアクセスするも、値を正常に取得できる事を確認する

  • service redisredis-1redis-6endpointに向いているか確認
$ kubectl get services
NAME         CLUSTER-IP      EXTERNAL-IP   PORT(S)              AGE
kubernetes   10.107.240.1    <none>        443/TCP              2d
redis        10.107.244.37   <none>        6379/TCP,16379/TCP   52m
redis-1      10.107.255.1    <none>        6379/TCP,16379/TCP   52m
redis-2      10.107.255.2    <none>        6379/TCP,16379/TCP   52m
redis-3      10.107.255.3    <none>        6379/TCP,16379/TCP   52m
redis-4      10.107.255.4    <none>        6379/TCP,16379/TCP   52m
redis-5      10.107.255.5    <none>        6379/TCP,16379/TCP   52m
redis-6      10.107.255.6    <none>        6379/TCP,16379/TCP   52m
$ kubectl describe service redis
Name:     redis
Namespace:    default
Labels:     <none>
Selector:   app=redis
Type:     ClusterIP
IP:     10.107.244.37
Port:     redis 6379/TCP
Endpoints:    10.104.0.18:6379,10.104.0.19:6379,10.104.0.20:6379 + 3 more...
Port:     cluster 16379/TCP
Endpoints:    10.104.0.18:16379,10.104.0.19:16379,10.104.0.20:16379 + 3 more...
Session Affinity: None
  • masterでデータを挿入し、loadbalancerからデータが正しく取得できるか確認する

redis-1(master)にてデータを挿入する

$ kubectl exec -it redis-1-2pg4v -- sh 
/data # redis-cli -c
127.0.0.1:6379> set hoge fuga
OK

redis-3(master)にてデータを挿入する

$ kubectl exec -it redis-3-4fs93 -- sh
/data # redis-cli -c
127.0.0.1:6379> set hogehoge fugafuga
-> Redirected to slot [694] located at 10.104.1.44:6379
OK

redis(loadalancer)にてデータを取得してみる

$ kubectl exec -it ubuntu -- sh
# redis-cli -h 10.107.244.37 -c
10.107.244.37:6379> get hoge
-> Redirected to slot [1525] located at 10.104.1.44:6379
"fuga"
10.104.1.44:6379> get hogehoge
"fugafuga"

redis-2(master)にて取得してみる

$ kubectl exec -it redis-2-fbzks sh                                                                                                       (feature/redis✱)
/data # redis-cli -c
127.0.0.1:6379> get hoge
-> Redirected to slot [1525] located at 10.104.1.46:6379
"fuga"
10.104.1.46:6379> get hogehoge
"fugafuga"

redis-5(slave)にて取得してみる

$ kubectl exec -it redis-5-bt1wm sh
/data # redis-cli -c
127.0.0.1:6379> get hoge
-> Redirected to slot [1525] located at 10.104.1.46:6379
"fuga"
10.104.1.46:6379> get hogehoge
"fugafuga"

検証

- gcp redis上で1000000万件データ作成

$ redis-cli debug populate 1000000
10.104.0.6:6379> mget key:0
1) "value:0"
10.104.1.4:6379> mget key:99999
-> Redirected to slot [2036] located at 10.104.0.6:6379
1) "value:99999"

- 使用memory確認

$ redis-cli info
# Memory
used_memory:2278256
used_memory_human:2.17M
used_memory_rss:3805184
used_memory_peak:2278256
used_memory_peak_human:2.17M
used_memory_lua:36864
mem_fragmentation_ratio:1.67
mem_allocator:libc

保存するデータの文字列の長さで予想されるデータ件数を見積もり、設定したメモリで耐えるか検証が別途必要である