However, all other types *cannot* be deleted explicitly, and instead are cleaned up when all references are removed. This means we can send partial updates for non-root types, without deleting unsent resources. This effectively allows doing delta updates over SotW. This optimization is critical for our endpoints generator, ensuring that when a pod scales we only need to update the endpoints within that pod.

Ce découplage des parties, et le fait d'être un "micro-framework" qui puisse être étendu pour couvrir exactement ce
qui est nécessaire, était une caractéristique clé que je voulais conserver.

| `minio_cluster_audit_failed_messages`     | `counter` | Total number of messages that failed to send since start | `target_id` |
| `minio_cluster_audit_target_queue_length` | `gauge`   | Number of unsent messages in queue for target            | `target_id` |
| `minio_cluster_audit_total_messages`      | `counter` | Total number of messages sent since start                | `target_id` |

### `/cluster/usage/objects`

|:----------------------------------|:----------------------------------------------------------|
| `minio_audit_failed_messages`     | Total number of messages that failed to send since start. |
| `minio_audit_target_queue_length` | Number of unsent messages in queue for target.            |
| `minio_audit_total_messages`      | Total number of messages sent since start.                |

## Cluster Capacity Metrics

			metrics = append(metrics, MetricV2{
				Description: MetricDescription{
					Namespace: minioNamespace,
					Subsystem: auditSubsystem,
					Name:      "target_queue_length",
					Help:      "Number of unsent messages in queue for target",
					Type:      gaugeMetric,
				},
				VariableLabels: map[string]string{"target_id": id},
				Value:          float64(st.QueueLength),
			})