Clarification regarding startup nodes about redisc HOT 5 CLOSED

Hello Alan,

Thanks for raising this, I wasn't familiar with how AWS handles Redis cluster. I guess the source of the confusion is this comment that mentions that only master nodes should be specified:

// StartupNodes is the list of initial nodes that make up
// the cluster. The values are expected as "address:port"
// (e.g.: "127.0.0.1:6379"). Only master nodes should be
// specified.

That strikes me as odd and I can't remember why I added that. I don't think it's reasonable to expect the caller to know what are the current master nodes, and I can't see why specifying replicas wouldn't work. I'll give it a closer look and come back at you.

I took a look at AWS doc, and it seems like each node in the cluster would get its own endpoint? That's from that page: https://docs.aws.amazon.com/AmazonElastiCache/latest/UserGuide/Endpoints.html#Endpoints.Find.RedisCluster

That would seem more reliable to me than resolving the IP beforehand, if that's for a long-running process (e.g. in case AWS takes a node down then back up with a different IP).

Martin

from redisc.

That would seem more reliable to me than resolving the IP beforehand, if that's for a long-running process (e.g. in case AWS takes a node down then back up with a different IP).

Hmm scratch that as Redis will return the IP addresses when requesting the CLUSTER SLOTS anyway, so yeah what you proposed would make sense.

from redisc.

I added a test of a Cluster started with a replica address, it works fine as expected. I also removed the erroneous mention in StartupNodes documentation.

from redisc.

Neat. Now that I understand the lib better, I think that should be sufficient. We're just passing in the primary endpoint with the list of 12 nodes in our cluster (4shard x 3way replication) and it's doing fine.

I think one issue I see is if it fails to dial the first result in the DNS response, but we can solve that with a custom dialer if it becomes a problem (I think).

Thanks!

from redisc.

Cool, I'll close this issue but let me know if you think of anything that could be done to improve the support for the AWS scenario.

from redisc.