Creating GlideClusterClient resulting in connection timeout

[sumanthns-commits]

Hi,

I am trying to create a GlideClusterClient in java to connect to a self-hosted valkey cluster on cloud. However, I am getting the below error
glide.api.models.exceptions.ClosingException: Connection error: Timeout

Also, I can connect to the cluster node using the same address and credentials with valkey-cli, to rule out any network problem. I can successfully create a standalone GlideClient connecting to a primary node in the same cluster too.

I tried giving a higher connection timeout (upto 10s) as recommended here, but it did not help

Here the code snipper I am using

    suspend fun flockGlideClusterClient(flockProperties: FlockProperties) {
        val glideClusterClient = GlideClusterClient.createClient(
            GlideClusterClientConfiguration.builder()
                .address(NodeAddress.builder().host("flockssl.d8pm.kitt-inf.net").port(6382).build())
                .credentials(
                    ServerCredentials.builder().username(flockProperties.username).password(flockProperties.password)
                        .build()
                )
                .useTLS(true)
                .requestTimeout(10000)
                .advancedConfiguration(
                    AdvancedGlideClusterClientConfiguration.builder().connectionTimeout(10000).build()
                )
                .build()
        ).await()
        val result = glideClusterClient.ping().await()
        println("Glide Cluster Client ping result: $result")
    }

[sumanthns-commits]

Doing the same with a standalone GlideClient works. Here is the code snippet

    suspend fun flockGlideClient(flockProperties: FlockProperties): GlideClient {
        val glideClient = GlideClient.createClient(
            GlideClientConfiguration.builder()
                .address(NodeAddress.builder().host("flockssl.d8pm.kitt-inf.net").port(6382).build())
                .credentials(
                    ServerCredentials.builder().username(flockProperties.username).password(flockProperties.password)
                        .build()
                )
                .useTLS(true)
                .advancedConfiguration(AdvancedGlideClientConfiguration.builder().connectionTimeout(10000).build())
                .build()
        ).get()
        val result = glideClient.ping().await()
        println("Glide Client ping result: $result") // Can see this getting printed as expected
    }
    ```

[sumanthns-commits]

Also, the user has permissions to run cluster nodes and cluster slots on all nodes in the cluster

[avifenesh]

@sumanthns-commits Hi, can you share how do you host it exactly?

[sumanthns-commits]

@avifenesh The cluster is hosted as a kubernetes statefulset in EKS. There is a NLB listening on multiple ports to route the traffic to the cluster through a reverse proxy layer, with each port bound to a particular valkey node.
Ex -> Load balancer port 6379 -> node-1, port 6380 -> node-2 etc

Having said that, the client app currently uses lettuce as a client to integrate with cluster. It is using a custom SocketAddressResolver to convert the valkey node hostnames to the load balancer host name and port.

We may need a custom address resolution in Glide too to get it working. Does Glide support it? If yes, could you link me to some examples ?

[avifenesh]

GLIDE does not support it currently, but I think this is an important feature to support.
The current architecture will ask the nodes about the topology map and will work based on it, so it will direct the request to the address the node is aware of.
The solution will probably be the NLB listening on the same addresses probably.

Could you start a new issue with a feature request for “resolver”?

[sumanthns-commits]

Thanks for your reply @avifenesh . Created #4396