Best Practices for Sentinel with Multiple Databases

[jkratz55]

I'm using Redis to store cached product data but currently caching it by different floor plans, IE: US, vs International. Because they have the same keys I am keeping them in different Redis databases 0 and 1 respectively. I was using NewFailoverClusterClient because I want to route reads to replicas for max throughput and performance. However, everything was written to DB 0. After looking at the code it made sense since NewFailoverClusterClient is just an adapter for ClusterClient which doesn't support multiple DBs. So I suppose my question is what are the best practices if I want to use Sentinel and multiple databases? Ideally, I'd like reads to be distributed but it isn't clear to me that NewFailureClient doesn't just send all reads/writes to whatever node is currently the master.

[jkratz55]

Not sure if the same team maintains this documentation, but it would probably be helpful to call out explicitly that using NewFailoverClusterClient has the same behavior as using ClusterClient in that you can only have a single DB.

https://redis.uptrace.dev/guide/go-redis-sentinel.html#redis-server-client

[monkey92t]

FailoverClient support read replica nodes?
OR.... FailoverClusterClient supports DB-1?

[jkratz55]

Sorry, let me try to clarify what I'm asking.

If I use the NewFailoverClient function, it returns a *redis.Client. When I use that client, is it only communicating with the node that is currently the master? That is to say; if doing read and write operations, do they all go to the current master?

When using NewFailoverClusterClient it returns a ClusterClient, and looking at the code, it appears to ignore the DB field in the FailoverOptions struct, which makes sense as cluster mode doesn't support multiple databases.

Ultimately I want to run Redis deployed in Sentinel mode, but I'd like the client to send read-only operations to all replicas but also support using multiple databases.