- Kube-proxy, when using a Service with External or LoadBalancer IPs on UDP services , was consuming a large amount of CPU because it was not filtering by the Service destination port and trying to delete all the UDP entries associated to the service. ([#130505](https://github.com/kubernetes/kubernetes/pull/130505), [@aojea](https://github.com/aojea)) [SIG...

- Improved performance of `Endpoint` and `EndpointSlice` controllers when there are a large number of services in a single namespace by making pod-to-service lookup asynchronous. ([#134739](https://github.com/kubernetes/kubernetes/pull/134739), [@shyamjvs](https://github.com/shyamjvs)) [SIG Apps and Network]

- Kube-apiserver: gzip compression switched from level 4 to level 1 to improve large list call latencies in exchange for higher network bandwidth usage (10-50% higher). This increases the headroom before very large unpaged list calls exceed request timeout limits. ([#112399](https://github.com/kubernetes/kubernetes/pull/112399), [@shyamjvs](https://github.com/shyamjvs)) [SIG API Machinery]

## Changes by Kind

### Bug or Regression

- Kube-apiserver: gzip compression switched from level 4 to level 1 to improve large list call latencies in exchange for higher network bandwidth usage (10-50% higher). This increases the headroom before very large unpaged list calls exceed request timeout limits. ([#112400](https://github.com/kubernetes/kubernetes/pull/112400), [@shyamjvs](https://github.com/shyamjvs)) [SIG API Machinery]

### Bug or Regression

- An inefficient lock in EndpointSlice controller metrics cache has been reworked. Network programming latency may be significantly reduced in certain scenarios, especially in clusters with a large number of Services. ([#107169](https://github.com/kubernetes/kubernetes/pull/107169), [@robscott](https://github.com/robscott)) [SIG Apps and Network]

all types in its type set must have the same underlying slice type <code>[]E</code>.
</p>

<p>
If the capacity of <code>s</code> is not large enough to fit the additional
values, <code>append</code> <a href="#Allocation">allocates</a> a new, sufficiently large underlying
array that fits both the existing slice elements and the additional values.
Otherwise, <code>append</code> re-uses the underlying array.
</p>

    }
    source.close()
  }

  /** Confirm that an unacknowledged write times out.  */
  @Test
  fun writeTimeouts() {
    val server = MockWebServer()
    // Sockets on some platforms can have large buffers that mean writes do not block when
    // required. These socket factories explicitly set the buffer sizes on sockets created.
    val socketBufferSize = 4 * 1024
    server.serverSocketFactory =

- Fixed bug in reflector that couldn't recover from "Too large resource version" errors with API servers 1.17.0-1.18.5 ([#94316](https://github.com/kubernetes/kubernetes/pull/94316), [@janeczku](https://github.com/janeczku)) [SIG API Machinery]

- KCCM: fixed transient node addition and removal caused by #121090 while syncing load balancers on large clusters with a lot of churn. ([#121091](https://github.com/kubernetes/kubernetes/pull/121091), [@alexanderConstantinescu](https://github.com/alexanderConstantinescu))

* Never prevent deletion of resources as part of namespace lifecycle ([#48733](https://github.com/kubernetes/kubernetes/pull/48733), [@liggitt](https://github.com/liggitt))
* AWS cloudprovider plugin: Fix for large clusters (200+ nodes). Also fix bug with volumes not getting detached from a node after reboot. ([#48312](https://github.com/kubernetes/kubernetes/pull/48312), [@gnufied](https://github.com/gnufied))