Kubernetes Redis Cluster issue

3/19/2017

I'm trying to create redis cluster using kubernetes on centos. I have my kubernetes master running on one host and kubernetes slaves on 2 different hosts.

etcdctl get /kube-centos/network/config

{ "Network": "172.30.0.0/16", "SubnetLen": 24, "Backend": { "Type": "vxlan" } }

Here is my replication controller

apiVersion: v1
kind: ReplicationController
metadata:
  name: redis-master
  labels:
  app: redis
  role: master
  tier: backend
spec:
  replicas: 6
  template:
    metadata:
      labels:
        app: redis
        role: master
        tier: backend
    spec:
      containers:
      - name: master
        image: redis
        command: 
          - "redis-server"
        args:
          - "/redis-master/redis.conf"
        ports:
        - containerPort: 6379
        volumeMounts:
        - mountPath: /redis-master
          name: config
        - mountPath: /redis-master-data 
          name: data
        volumes:
        - name: data
          emptyDir: {}
        - name: config
          configMap:
            name: redis-config
            items:
            - key: redis-config
              path: redis.conf    

kubectl create -f rc.yaml

NAME                           READY     STATUS    RESTARTS   AGE       IP            NODE
redis-master-149tt             1/1       Running   0          8s        172.30.96.4   centos-minion-1
redis-master-14j0k             1/1       Running   0          8s        172.30.79.3   centos-minion-2
redis-master-3wgdt             1/1       Running   0          8s        172.30.96.3   centos-minion-1
redis-master-84jtv             1/1       Running   0          8s        172.30.96.2   centos-minion-1
redis-master-fw3rs             1/1       Running   0          8s        172.30.79.4   centos-minion-2
redis-master-llg9n             1/1       Running   0          8s        172.30.79.2   centos-minion-2

Redis-config file used

appendonly yes
cluster-enabled yes
cluster-config-file /redis-master/nodes.conf
cluster-node-timeout 5000
dir /redis-master
port 6379

I used the following command to create the kubernetes service.

kubectl expose rc redis-master --name=redis-service --port=6379 --target-port=6379 --type=NodePort

Name:           redis-service
Namespace:      default
Labels:         app=redis
                role=master
                tier=backend
Selector:       app=redis,role=master,tier=backend
Type:           NodePort
IP:             10.254.229.114
Port:           <unset> 6379/TCP
NodePort:       <unset> 30894/TCP
Endpoints:      172.30.79.2:6379,172.30.79.3:6379,172.30.79.4:6379 + 3     more...
Session Affinity:   None
No events.

Now I have all the pods and service up and running. I'm using redis-trib pod to create redis cluster.

kubectl exec -it redis-trib bash

./redis-trib.rb create --replicas 1 172.30.79.2:6379 172.30.79.3:6379 172.30.79.4:6379 172.30.96.2:6379 172.30.96.3:6379 172.30.96.4:6379

Redis Cluster created as expected with the below message.

[OK] All 16384 slots covered.

Now I should be able to access my redis-cluster on kubernetes node IP(192.168.240.116) and nodePort(30894) from any host within my network. Everything works as expected when I execute the below command from one of the kubernetes node.

redis-cli -p 30894 -h 192.168.240.116 -c

192.168.240.116:30894> set foo bar
-> Redirected to slot [12182] located at 172.30.79.4:6379
OK
172.30.79.4:6379> 

When I run the same command from different (non-kubernetes) node within the same network, I see the connected timed out error.

redis-cli -c -p 30894 -h 192.168.240.116

192.168.240.116:30894> set foo bar
-> Redirected to slot [12182] located at 172.30.79.4:6379
Could not connect to Redis at 172.30.79.4:6379: Connection timed out

Is it not possible to access the redis-cluster outside the kubernetes cluster network when exposed using NodePort service type?

Also I cannot use LoadBalancer service type as I'm not hosting it on cloud.

I have been stuck with this issue for quite a while. Can someone suggest on what approach I should use to access my redis-cluster outside my network ?

Thanks

-- user3610007
flannel
kubernetes
redis

1 Answer

3/20/2017

Running ./redis-trib.rb create --replicas 1 172.30.79.2:6379 172.30.79.3:6379 172.30.79.4:6379 172.30.96.2:6379 172.30.96.3:6379 172.30.96.4:6379 doesn't make sense with this setup.

The port 6379 is only accessible through the service which you brough up, but never directly as you try. That's why you run into issues when you try to use your setup.

What you can do is to expose each POD with it's own service and have one additional cluster services to loadbalance external requests. As shown in the example repository from Kelsey Hightower. This way the PODs can communicate though the internally exposed ports and (external) clients can use the loadbalanced cluster port. The implication then is also that each POD requires it's own ReplicaSet (or Deployment). There's a long talk available on YouTube from Kelsey explaining the setup - YouTube / Slideshare.

An alternative would be to use a single redis master as shown in other examples.

-- pagid
Source: StackOverflow