fix: incorrect cross-cluster call warning when clusterName is not set

[daguimu]

Describe what this PR does / why we need it

In NacosLoadBalancer.getInstanceResponse(), the "A cross-cluster call occurs" warning log is incorrectly placed in the outer else block, which triggers when clusterName is blank (null or empty). Since PR #3771 removed the default value for clusterName, this warning fires on every load balancing request when users don't explicitly configure a cluster name.

The warning should only appear when clusterName IS configured but no same-cluster instances are found — that is the actual cross-cluster call scenario.

Does this pull request fix one issue?

Fixes #4020

Describe how you did it

Moved the else block with the warning log from the outer if (StringUtils.isNotBlank(clusterName)) into the inner if (!CollectionUtils.isEmpty(sameClusterInstances)).

Before (incorrect):

if (StringUtils.isNotBlank(clusterName)) {
    // filter same cluster instances
    if (!CollectionUtils.isEmpty(sameClusterInstances)) {
        instancesToChoose = sameClusterInstances;
    }
}
else {
    log.warn("A cross-cluster call occurs..."); // triggers when clusterName is BLANK
}

After (correct):

if (StringUtils.isNotBlank(clusterName)) {
    // filter same cluster instances
    if (!CollectionUtils.isEmpty(sameClusterInstances)) {
        instancesToChoose = sameClusterInstances;
    }
    else {
        log.warn("A cross-cluster call occurs..."); // triggers when clusterName is SET but no match
    }
}

Describe how to verify it

1. Start a Spring Cloud Alibaba application with Nacos discovery without setting spring.cloud.nacos.discovery.cluster-name
2. Make load-balanced requests
3. Verify no "A cross-cluster call occurs" warning in logs

Tests Added

Change Point	Test
Blank clusterName should not warn	blankClusterNameShouldNotWarnCrossCluster()
Empty clusterName should not warn	emptyClusterNameShouldNotWarnCrossCluster()
Same cluster instances selected correctly	sameClusterShouldSelectSameClusterInstance()
No same-cluster instances still returns response	noSameClusterInstancesShouldStillReturnResponse()
Empty instances returns EmptyResponse	emptyServiceInstancesShouldReturnEmptyResponse()

Special notes for reviews

PR #4021 proposed the same fix for the 2021.x branch. This PR applies it to the current 2025.1.x branch with comprehensive unit tests.

[uuuyuqi]

Thanks for the fix. One follow-up concern: this warning is still emitted from the request path in NacosLoadBalancer#getInstanceResponse(). That means if a service stays in a real cross-cluster fallback state for some time, the log can still be printed once per request and flood the logs under traffic.

Would you consider adding some form of throttling here, for example:

• log at most once per time window per (serviceId, clusterName), or
• log only on state transition (entering cross-cluster fallback), and suppress repeats until recovery

The current fix removes the false positives, which is great, but rate-limiting the warning would make the behavior much safer in sustained fallback scenarios.

[daguimu]

@uuuyuqi Good point! I've updated this PR to include log throttling — the cross-cluster warning now logs at most once per 5 minutes per (serviceId, clusterName) using a ConcurrentHashMap<String, Long> timestamp tracker. This prevents log flooding under sustained cross-cluster fallback traffic while still surfacing the warning when it first occurs.

Added test: crossClusterWarnShouldBeThrottled() — verifies the throttle state is tracked correctly after repeated cross-cluster calls.

[uuuyuqi]

LGTM