RoundRobinSocketAddressSupplier is not refreshing when RedisURI changes

[langchristian96]

Bug Report

RoundRobinSocketAddressSupplier is not refreshing nodes IPs if only IPs change while nodeIds remain the same. Because of this the ConnectionWatchdog keeps retrying to connect with an old IP. (seems to be similar to what was reported under #1343 )

Current Behavior

• Redis cluster with 6 nodes running under Kubernetes
• Remove redis pods 2 by 2 until all of them get new IPs
• After a while, the connection is remade and io.lettuce.core.cluster.RoundRobinSocketAddressSupplier#get is being called
• Partitions supplier gives us the correct partition with same node Ids but different IPs (which is correct)
• The following reset code inside the if block does not get executed
  if (!roundRobin.isConsistent(partitions)) { resetRoundRobin(partitions); }
  because the equals method from RedisClusterNode only compares the node ids
• We end up reconnecting using the old ip

Expected behavior/code

If the nodes IPs change, we should do the round robin rebuild.

Environment

• Lettuce version(s): 6.1.5.RELEASE
• Redis version: 6.2.6

Possible Solution

Modify the equals method from RedisClusterNode to check the ip/uri as well

Additional context

dynamicRefreshSources was set to false and the DNS resolver used was DirContextDnsResolver

[langchristian96]

On a second thought, comparing the URIs might not be the most correct approach since there can be aliases and it might affect other parts of the code. I think it might be better to go on another route and modify the reset conditions to check specifically if anything changed from an URI pov. Any thoughts on this?

[mp911de]

I'm happy for suggestions on how we can improve here.

[langchristian96]

@mp911de, I've made the changes I suggested. Mind taking a look when you have time please?

[langchristian96]

Managed to test the problematic scenario as well today and confirmed it works fine with the fix