Reliable, Scalable Redis on OpenShift

The following document describes the deployment of a reliable, multi-node Redis on OpenShift.

It deploys a master with replicated slaves, as well as replicated redis sentinels which are use for health checking and failover.

Build configuration

1. Create image stream and build config

       oc process -f build/redis-build.yml \
           -p REDIS_IMAGE_NAME=redis-ha \
           -p GIT_REPO=https://github.com/openlab-red/redis-ha.git \
           | oc create -f -
   

2. Start the build

       oc start-build redis-ha-build
   

Deployment Configuration

Persistent

Create a persistent storage, the storage must be RWX.

Create a new persistent storage (RWX)

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: backend-redis-storage
spec:
  accessModes:
  - ReadWriteMany
  resources:
    requests:
      storage: 1Gi

    oc create -f /tmp/storage.yml

Turning up an initial master/sentinel pod

Create a bootstrap master and sentinels with relative service

    export REDIS_PREFIX=backend
    export REDIS_NAME=${REDIS_PREFIX}-redis
    
    oc process -f templates/redis-master.yml \
        -p REDIS_SERVICE_PREFIX=${REDIS_PREFIX} \
        -p REDIS_IMAGE=redis-ha:latest \
        -p REDIS_PV=${REDIS_NAME}-storage \
        | oc create -f -

Turning up replicated redis slave servers

Create a deployment config for redis slave servers

    oc process -f templates/redis-slave.yml \
        -p REDIS_SERVICE_PREFIX=${REDIS_PREFIX} \
        -p REDIS_IMAGE=redis-ha:latest \
        -p REDIS_PV=${REDIS_NAME}-storage \
        | oc create -f -

Create a deployment config for redis sentinels

    oc process -f templates/redis-sentinel.yml \
        -p REDIS_SERVICE_PREFIX=${REDIS_PREFIX} \
        -p REDIS_IMAGE=redis-ha:latest \
        | oc create -f -

Scale down the bootstrap master

Scale down the original master pod

    oc scale --replicas=0 dc ${REDIS_NAME}-master

Failover

Recommended setup

For a recommended setup that can resist more failures, set the replicas to 5 (default) for Redis and Sentinel.

With 5 or 6 sentinels, a maximum of 2 can go down for a failover begin.

With 7 sentinels, a maximum of 3 nodes can go down.

Custom settings

For a fast failover under 1 minute

• REDIS_DOWN_AFTER_MILLIS=20000
• REDIS_FAILOVER_TIMEOUT=30000

Migration from existing single Redis instance

Migration from existing single redis instance

Backup and Recovery

Backup and Recovery

References

• https://redis.io/topics/sentinel
• https://github.com/kubernetes/examples/blob/master/staging/storage/redis/README.md
• https://github.com/sclorg/redis-container/blob/master/README.md
• https://github.com/mjudeikis/redis-openshift/blob/master/README.md