Automatic merge from submit-queue (batch tested with PRs 41667, 41820, 40910, 41645, 41361)
Switch admission to use shared informers
Originally part of #40097
cc @smarterclayton @derekwaynecarr @deads2k @liggitt @sttts @gmarek @wojtek-t @timothysc @lavalamp @kubernetes/sig-scalability-pr-reviews @kubernetes/sig-api-machinery-pr-reviews
node cache don't need to get all the information about every
candidate node. For huge clusters, sending full node information
of all nodes in the cluster on the wire every time a pod is scheduled
is expensive. If the scheduler is capable of caching node information
along with its capabilities, sending node name alone is sufficient.
These changes provide that optimization in a backward compatible way
- removed the inadvertent signature change of Prioritize() function
- added defensive checks as suggested
- added annotation specific test case
- updated the comments in the scheduler types
- got rid of apiVersion thats unused
- using *v1.NodeList as suggested
- backing out pod annotation update related changes made in the
1st commit
- Adjusted the comments in types.go and v1/types.go as suggested
in the code review
Where possible, switch the scheduler to use generated listers and
informers. There are still some places where it probably makes more
sense to use one-off reflectors/informers (listing/watching just a
single node, listing/watching scheduled & unscheduled pods using a field
selector).
Automatic merge from submit-queue (batch tested with PRs 41421, 41440, 36765, 41722)
Use watch param instead of deprecated /watch/ prefix
Switches clients to use watch param instead of /watch/ prefix
```release-note
Clients now use the `?watch=true` parameter to make watch API calls, instead of the `/watch/` path prefix
```
Automatic merge from submit-queue (batch tested with PRs 41401, 41195, 41664, 41521, 41651)
Remove default failure domains from anti-affinity feature
Removing it is necessary to make performance of this feature acceptable at some point.
With default failure domains (or in general when multiple topology keys are possible), we don't have transitivity between node belonging to a topology. And without this, it's pretty much impossible to solve this effectively.
@timothysc
Automatic merge from submit-queue (batch tested with PRs 41604, 41273, 41547)
Switch pv controller to shared informer
This is WIP because I still need to do something with bazel? and add 'get storageclasses' to the controller-manager rbac role
@jsafrane PTAL and make sure I did not break anything in the PV controller. Do we need to clone the volumes/claims we get from the shared informer before we use them? I could not find a place where we modify them but you would know for certain.
cc @ncdc because I copied what you did in your other PRs.
Automatic merge from submit-queue (batch tested with PRs 41517, 41494, 41163)
Make scheduler_perf warn rather then fail if scheduling rate is betwe…
**What this PR does / why we need it**
Fix scheduler per bench tests so they pass even if there is a cold interval .
**Special notes for your reviewer**:
This wont effect CI, as this bench test is run manually by devs.
`/var/run` is not world-writable on my OSX 10.11.x setup, so tests that
standup a secure apiserver fail with the default cert dir. Use a
tempdir instead.
Automatic merge from submit-queue
Switch serviceaccounts controller to generated shared informers
Originally part of #40097
cc @deads2k @sttts @liggitt @smarterclayton @gmarek @wojtek-t @timothysc @kubernetes/sig-scalability-pr-reviews
Automatic merge from submit-queue
Switch resourcequota controller to shared informers
Originally part of #40097
I have had some issues with this change in the past, when I updated `pkg/quota` to use the new informers while `pkg/controller/resourcequota` remained on the old informers. In this PR, both are switched to using the new informers. The issues in the past were lots of flakey test failures in the ResourceQuota e2es, where it would randomly fail to see deletions and handle replenishment. I am hoping that now that everything here is consistently using the new informers, there won't be any more of these flakes, but it's something to keep an eye out for.
I also think `pkg/controller/resourcequota` could be cleaned up. I don't think there's really any need for `replenishment_controller.go` any more since it's no longer running individual controllers per kind to replenish. It instead just uses the shared informer and adds event handlers to it. But maybe we do that in a follow up.
cc @derekwaynecarr @smarterclayton @wojtek-t @deads2k @sttts @liggitt @timothysc @kubernetes/sig-scalability-pr-reviews
Automatic merge from submit-queue (batch tested with PRs 41248, 41214)
Switch hpa controller to shared informer
**What this PR does / why we need it**: switch the hpa controller to use a shared informer
**Which issue this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close that issue when PR gets merged)*: fixes #
**Special notes for your reviewer**: Only the last commit is relevant. The others are from #40759, #41114, #41148
**Release note**:
```release-note
```
cc @smarterclayton @deads2k @sttts @liggitt @DirectXMan12 @timothysc @kubernetes/sig-scalability-pr-reviews @jszczepkowski @mwielgus @piosz
Automatic merge from submit-queue (batch tested with PRs 40796, 40878, 36033, 40838, 41210)
Implement TTL controller and use the ttl annotation attached to node in secret manager
For every secret attached to a pod as volume, Kubelet is trying to refresh it every sync period. Currently Kubelet has a ttl-cache of secrets of its pods and the ttl is set to 1 minute. That means that in large clusters we are targetting (5k nodes, 30pods/node), given that each pod has a secret associated with ServiceAccount from its namespaces, and with large enough number of namespaces (where on each node (almost) every pod is from a different namespace), that resource in ~30 GETs to refresh all secrets every minute from one node, which gives ~2500QPS for GET secrets to apiserver.
Apiserver cannot keep up with it very easily.
Desired solution would be to watch for secret changes, but because of security we don't want a node watching for all secrets, and it is not possible for now to watch only for secrets attached to pods from my node.
So as a temporary solution, we are introducing an annotation that would be a suggestion for kubelet for the TTL of secrets in the cache and a very simple controller that would be setting this annotation based on the cluster size (the large cluster is, the bigger ttl is).
That workaround mean that only very local changes are needed in Kubelet, we are creating a well separated very simple controller, and once watching "my secrets" will be possible it will be easy to remove it and switch to that. And it will allow us to reach scalability goals.
@dchen1107 @thockin @liggitt
Automatic merge from submit-queue (batch tested with PRs 40917, 41181, 41123, 36592, 41183)
fix scheduler performance test script
**What this PR does / why we need it**:
`test-performance.sh` is in dir `kubernetes/test/integration/scheduler_perf`
the dir `kubernetes/test/component/scheduler/perf` does not exist
Thanks.
**Special notes for your reviewer**:
**Release note**:
```release-note
```
Automatic merge from submit-queue (batch tested with PRs 41023, 41031, 40947)
scrub aggregator names to eliminate discovery
Cleanup old uses of `discovery`. Also removes the legacy functionality.
@kubernetes/sig-api-machinery-misc @sttts
Automatic merge from submit-queue
Replace hand-written informers with generated ones
Replace existing uses of hand-written informers with generated ones.
Follow-up commits will switch the use of one-off informers to shared
informers.
This is a precursor to #40097. That PR will switch one-off informers to shared informers for the majority of the code base (but not quite all of it...).
NOTE: this does create a second set of shared informers in the kube-controller-manager. This will be resolved back down to a single factory once #40097 is reviewed and merged.
There are a couple of places where I expanded the # of caches we wait for in the calls to `WaitForCacheSync` - please pay attention to those. I also added in a commented-out wait in the attach/detach controller. If @kubernetes/sig-storage-pr-reviews is ok with enabling the waiting, I'll do it (I'll just need to tweak an integration test slightly).
@deads2k @sttts @smarterclayton @liggitt @soltysh @timothysc @lavalamp @wojtek-t @gmarek @sjenning @derekwaynecarr @kubernetes/sig-scalability-pr-reviews
Automatic merge from submit-queue (batch tested with PRs 35782, 35831, 39279, 40853, 40867)
genericapiserver: cut off more dependencies – episode 7
Follow-up of https://github.com/kubernetes/kubernetes/pull/40822
approved based on #40363
Automatic merge from submit-queue
Removed HPA objects from extensions api group
fix#29778
``` release-note
HorizontalPodAutoscaler is no longer supported in extensions/v1beta1 version. Use autoscaling/v1 instead.
```
cc @kubernetes/autoscaling
Automatic merge from submit-queue
Use full package path for definition name in OpenAPI spec
We were using short package name (last part of package name) plus type name for OpenAPI spec definition name. That can result in duplicate names and make the spec invalid. To be sure we will always have unique names, we are going to use full package name as definition name. Also "x-kubernetes-tag" custom field is added to definitions to list Group/Version/Kind for the definitions that has it. This will help clients to discover definitions easier.
Lastly, we've added a reference from old definition names to the new ones to keep backward compatibilities. The list of old definitions will not be updated.
**Release note**:
- Rename OpenAPI definition names to type's full package names to prevent duplicates
- Create OpenAPI extension "x-kubernetes-group-version-kind" for definitions to store Group/Version/Kind
- Deprecate old definition names and create a reference to the new definitions. Old definitions will be removed in the next release.
Automatic merge from submit-queue (batch tested with PRs 40239, 40397, 40449, 40448, 40360)
move the discovery and dynamic clients
Moved the dynamic client, discovery client, testing/core, and testing/cache to `client-go`. Dependencies on api groups we don't have generated clients for have dropped out, so federation, kubeadm, and imagepolicy.
@caesarxuchao @sttts
approved based on https://github.com/kubernetes/kubernetes/issues/40363
Automatic merge from submit-queue (batch tested with PRs 39538, 40188, 40357, 38214, 40195)
Decoupling scheduler creation from creation of scheduler.Config struc…
**What this PR does / why we need it**:
Adds functionality to the scheduler to initialize from an Configurator interface, rather then via a Config struct.
**Which issue this PR fixes**
Reduces coupling to `scheduler.Config` data structure format so that we can proliferate more interface driven composition of scheduler components.
Automatic merge from submit-queue (batch tested with PRs 40428, 40176)
Cleaup Affinity post conversion from annotations to fields
**What this PR does / why we need it**:
Cleans up leftover work from the conversion of affinity from annotations to fields.
fixes#40016
related #25319
**Special notes for your reviewer**:
There are some TODO items left for @luxas or @errordeveloper b/c they were trying to use affinity in a way that is not possible.
**Release note**:
```release-note
NONE
```
/cc @kubernetes/sig-scheduling-misc @rrati
Automatic merge from submit-queue
move pkg/fields to apimachinery
Purely mechanical move of `pkg/fields` to apimachinery.
Discussed with @lavalamp on slack. Moving this an `labels` to apimachinery.
@liggitt any concerns? I think the idea of field selection should become generic and this ends up shared between client and server, so this is a more logical location.
Automatic merge from submit-queue
make client-go more authoritative
Builds on https://github.com/kubernetes/kubernetes/pull/40103
This moves a few more support package to client-go for origination.
1. restclient/watch - nodep
1. util/flowcontrol - used interface
1. util/integer, util/clock - used in controllers and in support of util/flowcontrol
Automatic merge from submit-queue (batch tested with PRs 39898, 39904)
[scheduler] interface for config
**What this PR fixes**
This PR converts the Scheduler configuration factory into an interface, so that
- the scheduler_perf and scheduler integration tests dont rely on the struct for their implementation
- the exported functionality of the factory (i.e. what it needs to provide to create a scheduler configuration) is completely explicit, rather then completely coupled to a struct.
- makes some parts of the factory immutable, again to minimize possible coupling.
This makes it easier to make a custom factory in instances where we might specifically want to import scheduler logic without actually reusing the entire scheduler codebase.
Automatic merge from submit-queue (batch tested with PRs 39625, 39842)
Add RBAC v1beta1
Add `rbac.authorization.k8s.io/v1beta1`. This scrubs `v1alpha1` to remove cruft, then add `v1beta1`. We'll update other bits of infrastructure to code to `v1beta1` as a separate step.
```release-note
The `attributeRestrictions` field has been removed from the PolicyRule type in the rbac.authorization.k8s.io/v1alpha1 API. The field was not used by the RBAC authorizer.
```
@kubernetes/sig-auth-misc @liggitt @erictune
Automatic merge from submit-queue
replace global registry in apimachinery with global registry in k8s.io/kubernetes
We'd like to remove all globals, but our immediate problem is that a shared registry between k8s.io/kubernetes and k8s.io/client-go doesn't work. Since client-go makes a copy, we can actually keep a global registry with other globals in pkg/api for now.
@kubernetes/sig-api-machinery-misc @lavalamp @smarterclayton @sttts
Mayank Kumar