SE.Redis not discovering Replica node AWS Cluster Mode Disabled

[StroebelePeaksWare]

I have a Redis cache with a single replica Node (for failover) but we would like to use it for read commands.

My understanding is that using the ConnectionMultiplexer with the primary node should auto discover the replicas but it is not.

Using get GetEndPoints() I can see it only finds the primary node
Using endpoint Unspecified/redacted.com:6379. Role: Standalone, Connected: True. Replica: False. Cluster node count: (null)

so I expanded logging and can see where it get INFO sets server type to Standalone

redacted.com:6379: Auto-configured (CLIENT) connection-id: 9560805
redacted.com:6379: Auto-configured (INFO) role: primary
redacted.com:6379: Auto-configured (INFO) version: 7.1.0
redacted.com:6379: Auto-configured (INFO) server-type: Standalone

So I jumped out to Cloud Shell to run the info command and it does return the Replication info

# Server
redis_mode:standalone
...
# Replication
role:master
connected_slaves:1

It looks like you have a topology issue with the discovery. I think you are over prioritizing redis_mode in your classification, and need to check the Replication info with redis_mode:standalone with role:master.

By giving ConnectionMultiplexer both the primary and reader endpoints I was able to work around this issue, but I am concerned how this setup will function for people giving the connectionMultiplexer just the primary endpoint with a second node for failover if SE.Redis doesn't know the replica exists will it still failover?

[mgravell]

Right now, the answer to that is "no" - without "cluster", the library does not attempt to discover much topology outside of what you tell it.

In part, this is because of problems with identification - short name Vs fully qualified name Vs internal name Vs public IP Vs private IP: knowing whether two different endpoints are actually the same server is challenging.

In part, this is because of problems of geography and georedundancy - a replication chain can span continents.
The current state is probably also strongly because when SE.Redis was originated, SE used georeplication but didn't use "cluster", so simply listing local endpoints was very pragmatic.

The good news is that on the roadmap (see "discussions"), one of the themes is "active active" - now; I know this scenario isn't that, but: it requires solving many of the same problems... or at least there's some crossover. So hopefully: some improvements there soon.

[StroebelePeaksWare]

Thanks for the context!

Is there a place in the documentation where we can make it clearer that auto discovery only works with clustered caches in the current version? I spent longer than I would care to admit debugging why the replica was never hit.

[mgravell]

I'll add something to configuration docs.