blockSize   int
	streamWg    sync.WaitGroup
	reuseBlocks bool
}

// newMetacacheWriter will create a serializer that will write objects in given order to the output.
// Provide a block size that affects latency. If 0 a default of 128KiB will be used.
// Block size can be up to 4MiB.
func newMetacacheWriter(out io.Writer, blockSize int) *metacacheWriter {
	if blockSize < 8<<10 {
		blockSize = 128 << 10
	}
	w := metacacheWriter{

 *       anything about how the property is tested. To find out, read the code.
 *   <li>These tests are "conformance tests", and do not attempt to test throughput, latency,
 *       scalability or other performance factors (see the separate "jtreg" tests for a set intended
 *       to check these for the most central aspects of functionality.) So, most tests use the

			runtime.Gosched()
			acc = e.cached.Load()
		}
	}
	atomic.AddInt64(&acc.N, 1)
	atomic.AddInt64(&acc.Total, int64(value))
	atomic.AddInt64(&acc.Size, sz)
}

// total returns the total call count and latency for the last minute.
func (e *lockedLastMinuteLatency) total() AccElem {
	e.mu.Lock()
	defer e.mu.Unlock()
	return e.lastMinuteLatency.getTotal()
}

 *       anything about how the property is tested. To find out, read the code.
 *   <li>These tests are "conformance tests", and do not attempt to test throughput, latency,
 *       scalability or other performance factors (see the separate "jtreg" tests for a set intended
 *       to check these for the most central aspects of functionality.) So, most tests use the

   *             cancelled the {@code Future}.
   *       </ol>
   *       Given all these possibilities, it is frequently possible for listeners to execute in UI
   *       threads, RPC network threads, or other latency-sensitive threads. In those cases, slow
   *       listeners can harm responsiveness, slow the system as a whole, or worse. (See also the
   *       note about locking below.)

          "format": "time_series",
          "instant": false,
          "interval": "",
          "intervalFactor": 1,
          "legendFormat": "{{bucket,range,operation}}",
          "refId": "A",
          "step": 300
        }
      ],
      "title": "Replication Latency (millis)",
      "type": "bargauge"
    },

- Added apiserver_envelope_encryption_kms_operations_latency_seconds metric to measure the KMSv2 grpc calls latency. ([#115649](https://github.com/kubernetes/kubernetes/pull/115649), [@aramase](https://github.com/aramase)) [SIG API Machinery, Auth and Testing]

  // Timed Get
  // There are a few design constraints to consider
  // * We want to be responsive to small timeouts, unpark() has non trivial latency overheads (I
  //   have observed 12 micros on 64-bit linux systems to wake up a parked thread). So if the
  //   timeout is small we shouldn't park(). This needs to be traded off with the cpu overhead of

- `kubelet` now emits a metric for end-to-end pod startup latency, including image pull. ([#121041](https://github.com/kubernetes/kubernetes/pull/121041), [@ruiwen-zhao](https://github.com/ruiwen-zhao))
- `kubelet` now exposes latency metrics of different stages of the node startup. ([#118568](https://github.com/kubernetes/kubernetes/pull/118568), [@qiutongs](https://github.com/qiutongs))

identified in large-scale clusters when deleting thousands of pods simultaneously across hundreds of nodes. Kubelets in this scenario can encounter temporarily increased latency of `delete pod` API calls -- above the target service level objective of 1 second. If you run clusters with this usage pattern and if pod deletion latency could be an issue for you, you might want to wait until the issue is resolved before you upgrade.

For more information and for updates on resolution of this...