- If a pod has a running instance, the stats of its previously terminated instances will not show up in the kubelet summary stats any more for CRI runtimes such as containerd and cri-o. This keeps the behavior consistent with Docker integration, and fixes an issue that some container Prometheus metrics don't work when there are summary stats for multiple instances of the same pod. ([#77426](https://github.com/kubernetes/kubernetes/pull/77426),...

* GCE: Fixes ILB sync on legacy networks and auto networks with unique subnet names ([#53410](https://github.com/kubernetes/kubernetes/pull/53410), [@nicksardo](https://github.com/nicksardo))
* Fix the bug that query Kubelet's stats summary with CRI stats enabled results in error. ([#53107](https://github.com/kubernetes/kubernetes/pull/53107), [@Random-Liu](https://github.com/Random-Liu))

* - Disable thin_ls due to excessive iops ([#43113](https://github.com/kubernetes/kubernetes/pull/43113), [@dashpole](https://github.com/dashpole))
    * - Ignore .mount cgroups, fixing dissappearing stats
    * - Fix wc goroutine leak
    * - Update aws-sdk-go dependency to 1.6.10

- Changed kubelet logs from `error` to `info` for uncached partitions when using CRI stats provider. ([#100448](https://github.com/kubernetes/kubernetes/pull/100448), [@saschagrunert](https://github.com/saschagrunert))

- Potentially breaking change - Updating the polling interval for Windows stats collection from 1 second to 10 seconds ([#116546](https://github.com/kubernetes/kubernetes/pull/116546), [@marosset](https://github.com/marosset)) [SIG Node and Windows]

- Added support for `PodAndContainerStatsFromCRI` feature gate, which allows a user to specify their pod stats must also come from the CRI, not `cAdvisor`. ([#103095](https://github.com/kubernetes/kubernetes/pull/103095), [@haircommander](https://github.com/haircommander))

### Windows

This release continues to enable more existing features on Windows, including container CPU resources, image filesystem stats, and flexvolumes. It also adds Windows service control manager support and experimental support for Hyper-V isolation of single-container pods.

### OpenStack

- Added support for pod `hostNetwork` field selector ([#110477](https://github.com/kubernetes/kubernetes/pull/110477), [@halfcrazy](https://github.com/halfcrazy)) [SIG Apps and Node]
- Added swap to stats to Summary API and Prometheus endpoints (stats/summary and /metrics/resource). ([#118865](https://github.com/kubernetes/kubernetes/pull/118865), [@iholder101](https://github.com/iholder101))

* Bumped Heapster version to 1.4.1: ([#50642](https://github.com/kubernetes/kubernetes/pull/50642), [@piosz](https://github.com/piosz))
    * handle gracefully problem when kubelet reports duplicated stats for the same container (see [#47853](https://github.com/kubernetes/kubernetes/pull/47853)) on Heapster side
    * fixed bugs and improved performance in Stackdriver Sink

  E0303 17:25:03.436781 63839 summary_sys_containers.go:47] Failed to get system container stats for "/docker": failed to get cgroup stats for "/docker": failed to get container info for "/docker": unknown container "/docker"