Cluster broadcast & health

Ask every pod at once and get one consolidated answer back — the pattern behind the built-in cluster health check, available for any function you annotate.

The pattern

Some questions are about the cluster, not about one pod behind the load balancer: is every pod healthy, what does each pod hold in its near cache, which config did each pod actually load? The broadcast bus answers them with scatter-gather over the backbone: one request is published on {root}:channel:broadcast, Redis reports how many pods received it, every pod (the caller included) runs the function and pushes its reply onto a short-lived list, and the caller consolidates the replies into one result. No pod list, no service discovery — the subscriber count is the pod count.

The consolidated health check

The built-in health function is registered on every pod automatically. Expose the endpoint (management.endpoints.web.exposure.include: …,dachealth) and one HTTP call to any pod checks all of them:

{
  "status": "UP",                    // UP · DOWN (a pod reported a problem) · UNKNOWN (replies missing)
  "expectedPods": 3,
  "respondedPods": 3,
  "nodes": [
    { "node": "5f2c…", "ok": true, "details": { "application": "indc-core", "status": "UP" } },
    { "node": "a91b…", "ok": true, "details": { "application": "indc-core", "status": "UP" } },
    { "node": "c07d…", "ok": true, "details": { "application": "indc-core", "status": "UP" } }
  ]
}

Each pod's status comes from its actuator HealthEndpoint when the application has one — so a pod whose database check fails reports DOWN here too — and is reported as UP otherwise (a pod that answers a broadcast is at least alive).

Broadcasting your own functions

Any bean method becomes cluster-callable with one annotation. The method takes no parameter or one String payload; it returns a String or any JSON-serialisable object:

@DacBroadcast("l1-stats")
public Map<String, Long> l1Stats() {          // runs on EVERY pod when broadcast
    return cacheManager.regions().stream()
            .collect(toMap(CacheRegion::name, r -> r.stats().l1Size()));
}

@Autowired Broadcaster broadcaster;

BroadcastResult result = broadcaster.broadcast("l1-stats", "", Duration.ofSeconds(5));
result.replies().forEach(reply ->
        log.info("pod {} -> {}", reply.nodeId(), reply.body()));
// result.complete()  — did every pod that received the request answer in time?
// result.allOk()     — did every reply succeed?

Rules and guarantees