/**
   * Invoked when OkHttp decides whether to retry after a connectivity failure.
   *
   * OkHttp won't retry when it is configured not to:
   *
   *  * If retries are forbidden with [OkHttpClient.retryOnConnectionFailure]. (OkHttp's defaults
   *    permit retries.)
   *  * If OkHttp already attempted to transmit the request body, and [RequestBody.isOneShot] is
   *    true.
   *

* Flex drivers by default do not produce metrics. Flex plugins can enable metrics collection by setting  the capability 'supportsMetrics' to true. Make sure the file system can support fs stat to produce metrics in this case. ([#67508](https://github.com/kubernetes/kubernetes/pull/67508), [@brahmaroutu](https://github.com/brahmaroutu))

```
~ mc admin config set alias/ scanner
KEY:
scanner  manage namespace scanning for usage calculation, lifecycle, healing and more

ARGS:

func (s *erasureSets) LocalStorageInfo(ctx context.Context, metrics bool) StorageInfo {
	var storageInfo StorageInfo

	storageInfos := make([]StorageInfo, len(s.sets))

	g := errgroup.WithNErrs(len(s.sets))
	for index := range s.sets {
		g.Go(func() error {
			storageInfos[index] = s.sets[index].LocalStorageInfo(ctx, metrics)
			return nil
		}, index)
	}

	// Wait for the go routines.

		sm.ReplicaSize += peer.ReplicaSize
		sm.Proxied.Add(madmin.ReplProxyMetric(peer.Proxied))
		for dID, v := range peer.Metrics {
			v2, ok := sm.Metrics[dID]
			if !ok {
				v2 = madmin.SRMetric{}
				v2.Failed.ErrCounts = make(map[string]int)
			}

			// use target endpoint metrics from node which has been up the longest
			if v2.LastOnline.After(v.LastOnline) || v2.LastOnline.IsZero() {

number of workers used for replication defaults to 100. Based on network bandwidth and system load, the number of workers used in replication can be configured using `mc admin config set alias api` to set the `replication_workers`. The prometheus metrics exposed by MinIO can be used to plan resource allocation and bandwidth management to optimize replication speed.

If synchronous replication is configured above, replication is attempted right away prior to returning the PUT object response....

pkg runtime/metrics, const KindBad = 0
pkg runtime/metrics, const KindBad ValueKind
pkg runtime/metrics, const KindFloat64 = 2
pkg runtime/metrics, const KindFloat64 ValueKind
pkg runtime/metrics, const KindFloat64Histogram = 3
pkg runtime/metrics, const KindFloat64Histogram ValueKind
pkg runtime/metrics, const KindUint64 = 1
pkg runtime/metrics, const KindUint64 ValueKind

你还可能有**其他原因**，这将使你更容易拥有一个带有**多个进程**的**单个容器**，而不是拥有每个容器中都有**单个进程**的**多个容器**。

例如（取决于你的设置）你可以在同一个容器中拥有一些工具，例如 Prometheus exporter，该工具应该有权访问**每个请求**。

在这种情况下，如果你有**多个容器**，默认情况下，当 Prometheus 来**读取metrics**时，它每次都会获取**单个容器**的metrics（对于处理该特定请求的容器），而不是获取所有复制容器的**累积metrics**。

在这种情况， 这种做法会更加简单：让**一个容器**具有**多个进程**，并在同一个容器上使用本地工具（例如 Prometheus exporter）收集所有内部进程的 Prometheus 指标并公开单个容器上的这些指标。

---

        }
        
        // Fall back to direct server query
        return directory.performDirectEnumeration();
    }
}
```

## 10. Monitoring and Metrics

### 10.1 Directory Cache Statistics
```java
public class DirectoryCacheStatistics {
    private final AtomicLong cacheHits = new AtomicLong();
    private final AtomicLong cacheMisses = new AtomicLong();

* The implicit registration of Prometheus metrics for request count and latency have been removed, and a plug-able interface was added. If you were using our client libraries in your own binaries and want these metrics, add the following to your imports in the main package: "k8s.io/pkg/client/metrics/prometheus".  ([#30638](https://github.com/kubernetes/kubernetes/pull/30638), [@krousey](https://github.com/krousey))