Problems connecting to AWS MemoryDB cluster endpoint

Hi,

We have an application that is using AWS MemoryDB. We have setup a cluster with one shard with two nodes. One of the nodes (named 0001-001) is a primary read/write while the other one is a read replica (named 0001-002).

After deploying the application, connecting to MemoryDB sometimes fails when we use the cluster endpoint connection string to connect. If we restart the application a few times it suddenly starts working. It seems to be random when it succeeds or not. The error we get is the following:

Endpoint Unspecified/ourapp-memorydb-cluster-0001-001.ourapp-memorydb-cluster.xxxxx.memorydb.eu-west-1.amazonaws.com:6379 serving hashslot 6024 is not reachable at this point of time. Please check connectTimeout value. If it is low, try increasing it to give the ConnectionMultiplexer a chance to recover from the network disconnect. IOCP: (Busy=0,Free=1000,Min=2,Max=1000), WORKER: (Busy=0,Free=32767,Min=2,Max=32767), Local-CPU: n/a

• If we connect directly to the primary read/write node we get no such errors.
• If we connect directly to the read replica it always fails. It even gets the error above, compaining about the "0001-001" node.
• We use StackExchange.Redis 2.2.4

Configuration:

new ConfigurationOptions()
{
    EndPoints =
    {
        { "clustercfg.ourapp-memorydb-cluster.xxxxx.memorydb.eu-west-1.amazonaws.com", 6379 } // Cluster endpoint
    },
    User = "username",
    Password = "password",
    Ssl = true,
    AbortOnConnectFail = false,
    ConnectTimeout = 60000
};

We could "solve" it by directly connecting to the primary node, but if a failover occurs and 0001-002 becomes the primary node we would have to manually change our connectionstring.

Are we missing any configuration to be able to use the cluster endpoint connection string? Is this a known issue for MemoryDB?

Thanks // Rickard

Start with the reported ConfigurationOptions and StackExchange.Redis 2.2.4 cluster-endpoint behavior, then reproduce the connection using AWS MemoryDB with a primary and read replica. Trace how the client discovers and routes cluster nodes, including the reported hash-slot error; done means the endpoint behavior is understood and a verified fix or supported-configuration conclusion is documented.