Direction looks right to me. One suggestion: this PR is really doing two independent things, and it'd be worth either calling them out explicitly or splitting into two PRs.

1. Bug fix — copy(). Today the RedisClientConfig copy constructor reuses the same FailedNodeDetector instance across every per-node RedisClient. Since detectors carry runtime state, one node's failures count toward another node's window, so isNodeFailed() is evaluated against cross-node-polluted state. copy() giving each node independent state is a genuine correctness fix and stands on its own.

2. Observability — setNodeAddress(InetSocketAddress). This is purely additive: no behavior change for the built-in detectors, it just lets a custom detector wrapper know which node it's observing so it can attribute/log the failed address.

These are separable — the copy() fix is valuable regardless, and the address tagging only becomes meaningful once detectors are per-node, so it reads more like a follow-up built on top of (1). Splitting (or at least separating the commits) would make the bug fix easier to review and backport independently of the instrumentation. Either way, the description should make the distinction explicit.

Java CI failed

thanks @mrniko fixed the checstyle failure for CI. can you help me approve another java CI workflow?

@mrniko it's weird. I retried exact same CI job locally and it ran successfully.

JAVA_HOME=$(/usr/libexec/java_home -v 25) PATH="$(/usr/libexec/java_home -v 25)/bin:$PATH" mvn -T 1C -B clean verify -DskipTests -Dmaven.test.skip=false --file pom.xml

[INFO] ------------------------------------------------------------------------
[INFO] BUILD SUCCESS
[INFO] ------------------------------------------------------------------------
[INFO] Total time:  45.291 s (Wall Clock)
[INFO] Finished at: 2026-07-01T18:19:27-07:00
[INFO] ------------------------------------------------------------------------

Seems like that version got published later in maven repository and also this whole failure doesn't look like related to above PR.

The dependabot PR (#7237) ran its Java CI at 17:57–17:58 UTC on 2026-07-01, hit the exact same quarkus-extension-maven-plugin with the exact same unpinned-version warning,
  resolved to the exact same 3.37.1, and it succeeded — because by then, ~7 hours after 3.37.1 was published, Maven Central's mirrors had fully caught up and
  quarkus-devtools-common:3.37.1 was resolvable everywhere.

I'm not able to rerun it but can you help take a look or re-run? thanks!