Automatic merge from submit-queue (batch tested with PRs 50953, 51082)
Fix mergekey of initializers; Repair invalid update of initializers
Fix https://github.com/kubernetes/kubernetes/issues/51131
The PR did two things to make parallel patching `metadata.initializers.pending` possible:
* Add mergekey to initializers.pending
* Let the initializer admission plugin set the `metadata.intializers` to nil if an update makes the `pending` and the `result` both nil, instead of returning a validation error. Otherwise if multiple initializer controllers sending the patch removing themselves from `pending` at the same time, one of them will get a validation error.
```release-note
The patch to remove the last initializer from metadata.initializer.pending will result in metadata.initializer to be set to nil (assuming metadata.initializer.result is also nil), instead of resulting in an validation error.
```
Automatic merge from submit-queue
Let the quota evaluator handle mutating specs of pod & pvc
### Background
The final goal is to address https://github.com/kubernetes/kubernetes/issues/47837, which aims to allow more mutation for uninitialized objects.
To do that, we [decided](https://github.com/kubernetes/kubernetes/issues/47837#issuecomment-321462433) to let the admission controllers to handle mutation of uninitialized objects.
### Issue
#50399 attempted to fix all admission controllers so that can handle mutating uninitialized objects. It was incomplete. I didn't realize although the resourcequota admission plugin handles the update operation, the underlying evaluator didn't. This PR updated the evaluators to handle updates of uninitialized pods/pvc.
### TODO
We still miss another piece. The [quota replenish controller](https://github.com/kubernetes/kubernetes/blob/master/pkg/controller/resourcequota/replenishment_controller.go) uses the sharedinformer, which doesn't observe the deletion of uninitialized pods at the moment. So there is a quota leak if a pod is deleted before it's initialized. It will be addressed with https://github.com/kubernetes/kubernetes/issues/48893.
Automatic merge from submit-queue
Fixed gke auth update wait condition.
Lookup whoami on gke using gcloud auth list.
Make sure we do not run the test on any cluster older than 1.7.
**What this PR does / why we need it**: Fixes issue with aggregator e2e test on GKE
**Which issue this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close that issue when PR gets merged)*: fixes#50945
**Special notes for your reviewer**: There is a TODO, follow up will be provided when the immediate problem is resolved.
**Release note**: ```release-note
NONE
```
Automatic merge from submit-queue (batch tested with PRs 51134, 51122, 50562, 50971, 51327)
Made the tests ensure that Cluster Autoscaler is on before running.
**What this PR does / why we need it**:
**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**:
**Release note**:
```release-note
NONE
```
Automatic merge from submit-queue
AllowedNotReadyNodes allowed to be not ready for absolutely *any* reason
It's as good as we allow those many nodes to be not part of the cluster at all, ever.
Btw - currently our 5k-node correctness test fails if "kubelet stopped posting node status" or "route not created", etc (ref: https://storage.googleapis.com/kubernetes-jenkins/logs/ci-kubernetes-e2e-gce-scale-correctness/3/build-log.txt)
cc @kubernetes/sig-scalability-misc
Automatic merge from submit-queue (batch tested with PRs 51244, 50559, 49770, 51194, 50901)
Distribute pods efficiently in CA scalability tests
**What this PR does / why we need it**:
Instead of using runReplicatedPodOnEachNode method
which is suited to a small number of nodes,
distribute pods on the nodes with desired load
using RCs that eat up all the space we want to be
empty after distribution.
**Release note**:
```
NONE
```
Automatic merge from submit-queue (batch tested with PRs 50213, 50707, 49502, 51230, 50848)
StatefulSet: Deflake e2e `kubectl exec` commands.
This may help with another source of flakiness found while investigating #48031.
We seem to get a lot of flakes due to "connection refused" while running `kubectl exec`. I can't find any reason this would be caused by the test flow, so I'm adding retries to see if that helps.
Automatic merge from submit-queue (batch tested with PRs 51224, 51191, 51158, 50669, 51222)
StatefulSet: Deflake e2e "restart" phase.
This addresses another source of flakiness found while investigating #48031.
The test used to scale the StatefulSet down to 0, wait for ListPods to return 0 matching Pods, and then scale the StatefulSet back up.
This was prone to a race in which StatefulSet was told to scale back up before it had observed its own deletion of the last Pod, as evidenced by logs showing the creation of Pod ss-1 prior to the creation of the replacement Pod ss-0.
Instead, we now wait for the controller to observe all deletions before scaling it back up. This should fix flakes of the form:
```
Too many pods scheduled, expected 1 got 2
```
We seem to get a lot of flakes due to "connection refused" while running
`kubectl exec`. I can't find any reason this would be caused by the test
flow, so I'm adding retries to see if that helps.
Instead of using runReplicatedPodOnEachNode method
which is suited to a small number of nodes,
distribute pods on the nodes with desired load
using RCs that eat up all the space we want to be
empty after distribution.
Automatic merge from submit-queue (batch tested with PRs 51193, 51154, 42689, 51189, 51200)
Re-enable OIR e2e tests.
Re-enabling test skeleton for opaque integer resources originally submitted as part of #41870. The e2e was disabled since it was flaky. This is the first step toward re-enabling them. Currently all cases are skipped, so this exercises only the BeforeEach behavior and the deferred removal of OIRs from a node.
cc @timothysc
Automatic merge from submit-queue (batch tested with PRs 51108, 51035, 50539, 51160, 50947)
Auto-calculate CLUSTER_IP_RANGE based on cluster size
In preparation for eliminating CLUSTER_IP_RANGE env var from job configs, making it less error prone while folks try to start their own large cluster tests (https://github.com/kubernetes/kubernetes/issues/50907).
/cc @kubernetes/sig-scalability-misc @wojtek-t @gmarek
Automatic merge from submit-queue (batch tested with PRs 51113, 46597, 50397, 51052, 51166)
Add statefulset upgrade tests to cluster_upgrade
**What this PR does / why we need it**:
Adds already created statefulset upgrade tests to cluster_upgrade.go. With further test infra changes, this will allow them to be continuously run, giving better signals.
Detect and prevent issues like https://github.com/kubernetes/kubernetes/issues/48327
**Release note**:
```release-note
NONE
```
Automatic merge from submit-queue (batch tested with PRs 51113, 46597, 50397, 51052, 51166)
implement proposal 34058: hostPath volume type
**What this PR does / why we need it**:
implement proposal #34058
**Which issue this PR fixes** : fixes#46549
**Special notes for your reviewer**:
cc @thockin @luxas @euank PTAL
Automatic merge from submit-queue
Skip "Simple pod should support exec through kubectl proxy" test
As reported in https://github.com/kubernetes/kubernetes/issues/50466,
this test doesn't work in GKE because it uses a bearer token and the feature only works with client certs.
As the feature that is broken in GKE is new and didn't work before, it
is safe to juste ignore the test and consider the feature as "still not
working" in GKE.
**What this PR does / why we need it**: Fixes the broken test in https://k8s-testgrid.appspot.com/release-master-blocking#gke
**Which issue this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close that issue when PR gets merged)*: works-around #50466
**Special notes for your reviewer**:
**Release note**:
```release-note
NONE
```
The test used to scale the StatefulSet down to 0, wait for ListPods to
return 0 matching Pods, and then scale the StatefulSet back up.
This was prone to a race in which StatefulSet was told to scale back up
before it had observed its own deletion of the last Pod, as evidenced by
logs showing the creation of Pod ss-1 prior to the creation of the
replacement Pod ss-0.
We now wait for the controller to observe all deletions before
scaling it back up. This should fix flakes of the form:
```
Too many pods scheduled, expected 1 got 2
```
Automatic merge from submit-queue (batch tested with PRs 50257, 50247, 50665, 50554, 51077)
Replace hard-code "cpu" and "memory" to consts
**What this PR does / why we need it**:
There are many places using hard coded "cpu" and "memory" as resource name. This PR replace them to consts.
**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**:
/kind cleanup
**Release note**:
```release-note
NONE
```
Automatic merge from submit-queue
StatefulSet: Deflake e2e "Saturate" phase.
This should reduce one source of flakiness found while investigating #48031.
The "Saturate" phase of StatefulSet e2e tests verifies orderly startup by controlling when each Pod is allowed to report Ready. If a Pod unexepectedly goes down during the test, the replacement Pod
created by the controller will forget if it was already allowed to report Ready.
After this change, the signal that allows each Pod to report Ready is persisted in the Pod's PVC. Thus, the replacement Pod will remember that it was already told to proceed to a Ready state.
Automatic merge from submit-queue (batch tested with PRs 51102, 50712, 51037, 51044, 51059)
[sig-network-e2e] Remove redundant sig prefix from tests
**What this PR does / why we need it**:
Remove redundant sig prefix from:
```
[sig-network] Networking [sig-network] Granular Checks: Services [Slow] should function for endpoint-Service: http
[sig-network] Networking [sig-network] Granular Checks: Services [Slow] should function for endpoint-Service: udp
[sig-network] Networking [sig-network] Granular Checks: Services [Slow] should function for node-Service: http
[sig-network] Networking [sig-network] Granular Checks: Services [Slow] should function for node-Service: udp
[sig-network] Networking [sig-network] Granular Checks: Services [Slow] should function for pod-Service: http
[sig-network] Networking [sig-network] Granular Checks: Services [Slow] should function for pod-Service: udp
[sig-network] Networking [sig-network] Granular Checks: Services [Slow] should update endpoints: http
[sig-network] Networking [sig-network] Granular Checks: Services [Slow] should update endpoints: udp
[sig-network] Networking [sig-network] Granular Checks: Services [Slow] should update nodePort: http [Slow]
[sig-network] Networking [sig-network] Granular Checks: Services [Slow] should update nodePort: udp [Slow]
[sig-network] Loadbalancing: L7 [sig-network] GCE [Slow] [Feature:Ingress] should conform to Ingress spec
[sig-network] Loadbalancing: L7 [sig-network] GCE [Slow] [Feature:Ingress] should create ingress with given static-ip
```
Umbrella issue #49161
**Special notes for your reviewer**:
cc @xiangpengzhao
**Release note**:
```release-note
NONE
```
As reported in https://github.com/kubernetes/kubernetes/issues/50466,
this test doesn't work in GKE because the transport layer doesn't work
with dialing.
As the feature that is broken in GKE is new and didn't work before, it
is safe to juste ignore the test and consider the feature as "still not
working" in GKE.
Automatic merge from submit-queue (batch tested with PRs 50893, 50913, 50963, 50629, 50640)
Increase latency threshold for list api calls
This is only a short-term solution to make our density test green. In the long-term, we should measure as per our new SLIs.
From @wojtek-t's [doc](https://docs.google.com/document/d/1Q5qxdeBPgTTIXZxdsFILg7kgqWhvOwY8uROEf0j5YBw) on the new SLIs/SLOs, we have the following SLO for list calls:
```
SLO1: In default Kubernetes installation, 99th percentile of SLI2 per cluster-day:
<= 1s if total number of objects of the same type as resource in the system <= X
<= 5s if total number of objects of the same type as resource in the system <= Y
<= 30s if total number of objects of the same types as resource in the system <= Z
```
I would guess that 170,000 pods would fall into the 2nd bracket (at least) and hence the new value of 5s. WDYT?
cc @kubernetes/sig-scalability-misc @wojtek-t @gmarek
The "Saturate" phase of StatefulSet e2e tests verifies orderly startup
by controlling when each Pod is allowed to report Ready.
If a Pod unexepectedly goes down during the test, the replacement Pod
created by the controller will forget if it was already allowed to
report Ready.
After this change, the signal that allows each Pod to report Ready is
persisted in the Pod's PVC. Thus, the replacement Pod will remember that
it was already told to proceed to a Ready state.
Automatic merge from submit-queue (batch tested with PRs 47896, 50678, 50620, 50631, 51005)
Made the difference between scale-up timeout and cluster set-up timeout explicit.
**What this PR does / why we need it**:
**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**:
**Release note**:
```release-note
NONE
```
Automatic merge from submit-queue (batch tested with PRs 46512, 50146)
Make metav1.(Micro)?Time functions take pointers
Is there any reason for those functions not to be on pointers?
