Moved all flag code from `staging/src/k8s.io/apiserver/pkg/util/[flag|globalflag]` to `component-base/cli/[flag|globalflag]` except for the term function because of unwanted dependencies.
The goal of this change is to remove the registration of signal
handling from pkg/kubelet. We now pass in a stop channel.
If you register a signal handler in `main()` to aid in a controlled
and deliberate exit then the handler registered in `pkg/kubelet` often
wins and the process exits immediately. This means all other signal
handler registrations are currently racy if `DockerServer.Start()` is
directly or indirectly invoked.
This change also removes another signal handler registration from
`NewAPIServerCommand()`; a stop channel is now passed to this
function.
This moves plugin/pkg/scheduler to pkg/scheduler and
plugin/cmd/kube-scheduler to cmd/kube-scheduler.
Bulk of the work was done with gomvpkg, except for kube-scheduler main
package.
Refactor the kube-scheduler configuration API, command setup, and server
setup according to the guidelines established in #32215 and using the
kube-proxy refactor (#34727) as a model of a well factored component
adhering to said guidelines.
* Config API: clarify meaning and use of algorithm source by replacing
modality derived from bools and string emptiness checks with an explicit
AlgorithmSource type hierarchy.
* Config API: consolidate client connection config with common structs.
* Config API: split and simplify healthz/metrics server configuration.
* Config API: clarify leader election configuration.
* Config API: improve defaulting.
* CLI: deprecate all flags except `--config`.
* CLI: port all flags to new config API.
* CLI: refactor to match kube-proxy Cobra command style.
* Server: refactor away configurator.go to clarify application wiring.
* Server: refactor to more clearly separate wiring/setup from running.
Fixes#52428.
Automatic merge from submit-queue (batch tested with PRs 47961, 46276)
Remove duplicate error message output in hyperkube.
**What this PR does / why we need it**:
Currently hyperkube binary will always print duplicate messages if fatal error occurs:
```
$ ./kubelet # run without permission
I0523 08:38:02.642638 25701 feature_gate.go:144] feature gates: map[]
W0523 08:38:02.642827 25701 server.go:472] No API client: no api servers specified
Error: failed to run Kubelet: error reading /var/run/kubernetes/kubelet.key, certificate and key must be supplied as a pair
Error: failed to run Kubelet: error reading /var/run/kubernetes/kubelet.key, certificate and key must be supplied as a pair
```
This is due to RunToExit will print out the error which Run has just printed. https://github.com/kubernetes/kubernetes/blob/8bee44b/cmd/hyperkube/hyperkube.go#L178-L189
This was introduced in following commit for adding GOMAXPROCS calls.
1e679f0069
The following commit removes GOMAXPROCS calls, but did not remove the corresponding error output line.
88ea80b572
**Which issue this PR fixes**
**Special notes for your reviewer**:
**Release note**:
Automatic merge from submit-queue (batch tested with PRs 45514, 45635)
hyperkube_test should not depend on number of spaces.
From #45524.
Apparently adding a long flag to kube-controller-manager breaks the hyperkube unit tests, because they depend on number of spaces :)
**Release note**:
```release-note
NONE
```
Darren Shepherd