Redis high availability and load balancing

Question

I'm using Redis for some time and now we want to move to a redundant cluster which would be fault tolerant, highly available and load balanced.

I principally use Redis for some PHP web application which stores some process status information (I'm using PHPRedis for this) and I would like to use it also for PHP session store in a load balanced cluster.

I already have a Load Balancing that I use for Nginx and MySQL (based on pfSense).

I saw that Redis Cluster exists for Redis 3, but I also read some comments about the fact that it is not ready for production, others that tells that Redis cannot do a complete master/master cluster configuration.

What is your advise?

• Try Redis Cluster
• Move to another key=value engine (CouchDB? Memcached? Others?)
• Don't mind about load balancing and use only one node of the cluster per time, and having a manual failover, as I don't use persistence
• Other ideas?

Thank you very much for your help!

Solution 1

Redis Sentinel will allow you to run multiple instances of redis and handles failover for you. http://redis.io/topics/sentinel

Solution 2

An obvious solution to loadbalance redis is to use sentinel.

As far as I can see predis does not support sentinel as of now. See:

https://github.com/nrk/predis/issues/131

A possible solution can be Redis sentinel + HAProxy failover if you want to keep using predis.

Comments

[Nathan Basanese]

// , Mat, did it work?

[Kim]

@Pierre.Vriens maybe I am not following you correctly but as far as I see the question is how to loadbalance redis. I am pointing out that an obvious solution like sentinel will not work when using predis. Possibly that was not clear from my answer.

[Pierre.Vriens]

ok, that helps a bit to better understand your answer. I think you should edit it to somehow integrate that in it also (and then delete your comment).

[Adam Lassek]

The question specifically mentions load balancing, which Sentinel doesn't do. You should at least address that in your comment.