// says its local for a given host - it is true for all endpoints
	// for the same host. Following code ensures that this assumption
	// is true and it works in all scenarios and it is safe to assume
	// for a given host.
	endpointLocalMap := make(map[string]bool)
	for _, ep := range endpoints {
		if ep.IsLocal {
			endpointLocalMap[ep.Host] = ep.IsLocal
		}
	}

### Suporte do OpenAPI

O OpenAPI não suporta uma maneira de declarar um *parâmetro de rota* que contenha uma *rota* dentro, dado que isso poderia levar a cenários que são difíceis de testar e definir.

No entanto, você pode fazer isso no **FastAPI**, usando uma das ferramentas internas do Starlette.

		// Add IP/port combo to set. Note that we set Replace to false here - we _did_ previously
		// set it to true, but in theory that could mask weird scenarios where K8S triggers events out of order ->
		// an add(sameIPreused) then delete(originalIP).
		// Which will result in the new pod starting to fail healthchecks.
		//

	// for this instance of unlocker, we should let the
	// caller know that lock is not successfully released
	// yet.
	if len(locks)-tolerance == tolerance {
		// In case of split brain scenarios where
		// tolerance is exactly half of the len(*locks)
		// then we need to make sure we have unlocked
		// upto tolerance+1 - especially for RUnlock
		// to ensure that we don't end up with active

- The scheduling queue didn't notice any extenders' failures, it could miss some cluster events,
  and it could end up Pods rejected by Extenders stuck in unschedulable pod pool in 5min in the worst-case scenario.
  Now, the scheduling queue notices extenders' failures and requeue Pods rejected by Extenders appropriately. ([#122043](https://github.com/kubernetes/kubernetes/pull/122043), [@sanposhiho](https://github.com/sanposhiho)) [SIG Scheduling]

- Cleanup subnet in frontend IP configs to prevent huge subnet request bodies in some scenarios. ([#98290](https://github.com/kubernetes/kubernetes/pull/98290), [@nilo19](https://github.com/nilo19)) [SIG Cloud Provider]

// as a hex encoded string
func (p *PutObjReader) MD5CurrentHexString() string {
	md5sumCurr := p.rawReader.MD5Current()
	var appendHyphen bool
	// md5sumcurr is not empty in two scenarios
	// - server is running in strict compatibility mode
	// - client set Content-Md5 during PUT operation
	if len(md5sumCurr) == 0 {
		// md5sumCurr is only empty when we are running
		// in non-compatibility mode.

		// Optimization for certain applications like
		// - Cohesity
		// - Actifio, Splunk etc.
		// which send ListObjects requests where the actual object
		// itself is the prefix and max-keys=1 in such scenarios
		// we can simply verify locally if such an object exists
		// to avoid the need for ListObjects().
		objInfo, err := z.GetObjectInfo(ctx, bucket, prefix, ObjectOptions{NoLock: true})
		if err == nil {

### Um Processo por Contêiner

Nesse tipo de cenário, provavelmente você desejará ter **um único processo (Uvicorn) por contêiner**, pois já estaria lidando com a replicação no nível do cluster.

### Ein Prozess pro Container

In einem solchen Szenario möchten Sie wahrscheinlich **einen einzelnen (Uvicorn-)Prozess pro Container** haben, da Sie die Replikation bereits auf Cluster ebene durchführen würden.