continue
				}
				unLocker := bmg.LockAndSetBuckets(buckets)
				defer unLocker()
			}
			gatherers = append(gatherers, gatherer)
		}
	}

	if len(gatherers) == 0 {
		return notFoundHandler
	}

	return promhttp.HandlerFor(prometheus.Gatherers(gatherers), h.opts)
}

// ServeHTTP - implements http.Handler interface.
//

	gatherers := prometheus.Gatherers{
		registry,
	}

	return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
		tc, ok := r.Context().Value(mcontext.ContextTraceKey).(*mcontext.TraceCtxt)
		if ok {
			tc.FuncName = funcName
			tc.ResponseRecorder.LogErrBody = true
		}

		mfs, err := gatherers.Gather()
		if err != nil && len(mfs) == 0 {

	madmin.PeerSite
	self         bool
	DeploymentID string
	Replicated   bool // true if already participating in site replication
	Empty        bool // true if cluster has no buckets
}

// getSiteStatuses gathers more info on the sites being added
func (c *SiteReplicationSys) getSiteStatuses(ctx context.Context, sites ...madmin.PeerSite) (psi []PeerSiteInfo, err error) {
	psi = make([]PeerSiteInfo, 0, len(sites))
	for _, v := range sites {

*   TPU embedding support
    *   Added `profile_data_directory` to `EmbeddingConfigSpec` in
        `_tpu_estimator_embedding.py`. This allows embedding lookup statistics
        gathered at runtime to be used in embedding layer partitioning
        decisions.
*   PluggableDevice
    *   Third-party devices can now connect to TensorFlow as plug-ins through