CODEWITHSUNDEEP
c#pythonredis
scarpacci
scarpacci

Reputation: 9194

Utilizing two Redis instances - Similar to Mongos

I have been reading that the proper way to scale Redis is to add a Separate instance (Even on the same machine is ok because CPU intensive). What I am wondering is if there are any existing components out there that facilitate the round robin / write / read similar to Mongos so that I could just call into it and it would properly write / read to one of the underlying instances. I realize that it is more complicated that what I have represented above, but didn't want to re-invent the wheel by trying to write my own proxy, etc to handle this.

Any suggestions / tips, etc would be appreciated.

Thanks,

S

Upvotes: 0

Views: 347

Answers (1)

The Real Bill
The Real Bill

Reputation: 15773

The approach will work for scaling reads, but not writes as Redis is not yet released with redis-cluster.

For load balancing reads, any TCP load balancer should work fine such as Balance. I link that one because it is software based and pretty simple to set up and use. Of course, if you have a hardware load balancer you could do it there, or use any of several other software based load balancers.

Another option is to implement round robin in your client code, though I prefer to not do that myself. Once redis-cluster is released it won't really matter which server you connect to.

For balancing writes, you'll need to go the route of sharding your data, which is described rather well IMO at Craigslist's Redis usgae page. If you think you'll need to go this route, I'd recommend taking the line JZ takes and do the underlying setup in advance. Ideally once redis-cluster is ready there should be minimal, if any, code changes to move to the cluster handling it for you.

If you want a single IP to handle both reads and writes as well as multiple sharded write masters you would likely need to write that "proxy" yourself, or put the code in the client code you write. Alternatively, this proxy announcement may hold what you need, though I don't see anything about routing writes in it.

Ultimately, I think you'd need to test and validate you actually need that write scaling before implementing it. I've found that if I have all reads on one or more slaves, and have the slaves manage disk persistence, performance of writes is usually not an issue.

Upvotes: 1

Related Questions