- client-go discovery clients constructed using `NewDiscoveryClientForConfig` or `NewDiscoveryClientForConfigOrDie` default to rate limits that cause normal discovery request patterns to take several seconds. This is fixed in https://issue.k8s.io/86168 and will be resolved in v1.17.1. As a workaround, the `Burst` value can be adjusted higher in the rest.Config passed into `NewDiscoveryClientForConfig` or `NewDiscoveryClientForConfigOrDie`....

*   Pod preemption has been improved.
    *   Now takes PodDisruptionBudget into account. ([#56178](https://github.com/kubernetes/kubernetes/pull/56178),[ @bsalamat](https://github.com/bsalamat))

consequence, all volume mounts in containers are now `rslave` on Linux by default. To make this default work in all Linux environments the entire mount tree should be marked as shareable, e.g. via `mount --make-rshared /`. All Linux distributions that use systemd already have the root directory mounted as rshared and hence they need not do anything. In Linux environments without systemd we recommend running `mount --make-rshared /` during boot before docker is started, ([@jsafrane](https://github.com/jsafrane))...

Call             Type T            Result

make(T, n)       slice             slice of type T with length n and capacity n
make(T, n, m)    slice             slice of type T with length n and capacity m

make(T)          map               map of type T
make(T, n)       map               map of type T with initial space for approximately n elements

make(T)          channel           unbuffered channel of type T

- QueueingHint implementation for NodeUnschedulable is reverted because we found potential scenarios where events that make Pods schedulable could be missed. ([#122326](https://github.com/kubernetes/kubernetes/pull/122326), [...

* GCE: Fix to make the built-in `kubernetes` service properly point to the master's load balancer address in clusters that use multiple master VMs. ([#63696](https://github.com/kubernetes/kubernetes/pull/63696), [@grosskur](https://github.com/grosskur))

- Fix the regression with the slow pods termination. Before this fix pods may take an additional time to terminate - up to one minute. Reversing the change that ensured that CNI resources cleaned up when the pod is removed on API server. ([#97980](https://github.com/kubernetes/kubernetes/pull/97980), [@SergeyKanzhelev](https://g...

- Make STS available replicas optional again. ([#109241](https://github.com/kubernetes/kubernetes/pull/109241), [@ravisantoshgudimetla](https://github.com/ravisantoshgudimetla))

job controller back-off logic is now decoupled from workqueue. In case of parallelism > 1, if there are multiple new failures in a reconciliation cycle, all the failures are taken into account to compute the back-off. Previously, the back-off kicked in for all types of failures; with this change, only pod failures are taken into account. If the back-off limits exceeds, the job is marked as failed immediately; before this change, the job is marked as failed in the next back-off. ([#114768](https:...

- Decrease the amount of time it takes to modify kubeconfig files with large amounts of contexts ([#67093](https://github.com/kubernetes/kubernetes/pull/67093), [@juanvallejo](https://github.com/juanvallejo))
- Make EBS volume expansion faster ([#66728](https://github.com/kubernetes/kubernetes/pull/66728), [@gnufied](https://github.com/gnufied))