Automatic merge from submit-queue (batch tested with PRs 64503, 64903, 64643, 64987). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Use unix.EpollWait to determine when memcg events are available to be Read
**What this PR does / why we need it**:
This fixes a file descriptor leak introduced in https://github.com/kubernetes/kubernetes/pull/60531 when the `--experimental-kernel-memcg-notification` kubelet flag is enabled. The root of the issue is that `unix.Read` blocks indefinitely when reading from an event file descriptor and there is nothing to read. Since we refresh the memcg notifications, these reads accumulate until the memcg threshold is crossed, at which time all reads complete. However, if the node never comes under memory pressure, the node can run out of file descriptors.
This PR changes the eviction manager to use `unix.EpollWait` to wait, with a 10 second timeout, for events to be available on the eventfd. We only read from the eventfd when there is an event available to be read, preventing an accumulation of `unix.Read` threads, and allowing the event file descriptors to be reclaimed by the kernel.
This PR also breaks the creation, and updating of the memcg threshold into separate portions, and performs creation before starting the periodic synchronize calls. It also moves the logic of configuring memory thresholds into memory_threshold_notifier into a separate file.
This also reverts https://github.com/kubernetes/kubernetes/pull/64582, as the underlying leak that caused us to disable it for testing is fixed here.
Fixes#62808
**Release note**:
```release-note
NONE
```
/sig node
/kind bug
/priority critical-urgent
Automatic merge from submit-queue (batch tested with PRs 64503, 64903, 64643, 64987). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Create system:cluster-autoscaler account & role and introduce it to C…
**What this PR does / why we need it**:
This PR adds cluster-autoscaler ClusterRole & binding, to be used by the Cluster Autoscaler (kubernetes/autoscaler repository).
It also updates GCE scripts to make CA use the cluster-autoscaler user account.
User account instead of Service account is chosen to be more in line with kube-scheduler.
**Which issue(s) this PR fixes**:
Fixes [issue 383](https://github.com/kubernetes/autoscaler/issues/383) from kubernetes/autoscaler.
**Special notes for your reviewer**:
This PR might be treated as a security fix since prior to it CA on GCE was using system:cluster-admin account, assumed due to default handling of unsecured & unauthenticated traffic over plain HTTP.
**Release note**:
```release-note
A cluster-autoscaler ClusterRole is added to cover only the functionality required by Cluster Autoscaler and avoid abusing system:cluster-admin role.
action required: Cloud providers other than GCE might want to update their deployments or sample yaml files to reuse the role created via add-on.
```
Automatic merge from submit-queue. If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
remove deprecated option '--enable-custom-metrics'
**Release note**:
```release-note
deprecated and inactive option '--enable-custom-metrics' is removed in 1.11
```
Automatic merge from submit-queue (batch tested with PRs 64276, 64094, 64719, 64766, 64750). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Use default seccomp profile for GCE manifests
**What this PR does / why we need it**:
This PR sets the default seccomp profile of unprivileged addons to 'docker/default' for GCE manifests. This PR is a followup of #62662. We are using 'docker/default' instead of 'runtime/default' in addons in order to handle node version skew. When seccomp profile is applied automatically by default later, we can remove those annotations.
This is PR is part of #39845.
**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
Fixes #
**Special notes for your reviewer**:
**Release note**:
```release-note
NONE
```
Automatic merge from submit-queue (batch tested with PRs 63453, 64592, 64482, 64618, 64661). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Revert "Remove rescheduler and corresponding tests from master"
Reverts kubernetes/kubernetes#64364
After discussing with @bsalamat on how DS controllers(ref: https://github.com/kubernetes/kubernetes/pull/63223#discussion_r192277527) cannot create pods if the cluster is at capacity and they have to rely on rescheduler for making some space, we thought it is better to
- Bring rescheduler back.
- Make rescheduler priority aware.
- If cluster is full and if **only** DS controller is not able to create pods, let rescheduler be run and let it evict some pods which have less priority.
- The DS controller pods will be scheduled now.
So, I am reverting this PR now. Step 2, 3 above are going to be in rescheduler.
/cc @bsalamat @aveshagarwal @k82cn
Please let me know your thoughts on this.
```release-note
Revert #64364 to resurrect rescheduler. More info https://github.com/kubernetes/kubernetes/issues/64725 :)
```
Automatic merge from submit-queue (batch tested with PRs 61610, 64591, 58143, 63929). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Add netd as an addon for GCP
**What this PR does / why we need it**:
Add netd as an addon for GKE.
The PR will add setup functions and var to help deploy netd daemon on GKE.
Please checkout more detail for netd at https://github.com/GoogleCloudPlatform/netd
**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
Fixes #
**Special notes for your reviewer**:
**Release note**:
```release-note
NONE
```
Automatic merge from submit-queue (batch tested with PRs 61610, 64591, 58143, 63929). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Create CoreDNS and kube-dns folders
**What this PR does / why we need it**:
Separate the CoreDNS and kube-dns manifests by creating their own folders (dns/coredns and dns/kube-dns)
**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
Fixes#61435
**Special notes for your reviewer**:
**Release note**:
```release-note
NONE
```
cc @MrHohn
Automatic merge from submit-queue. If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Add ipvs module loading logic to gce scripts
**What this PR does / why we need it**:
Add ipvs module loading logic to gce scripts.
Fixes a part of #59402.
/cc @Lion-Wei
/assign @roberthbailey @m1093782566
**Release note**:
```release-note
None
```
Automatic merge from submit-queue. If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Disable memcg for testing prior to 1.11 release
**What this PR does / why we need it**:
Turn off kubelet memory cgroup notifications on GCE to unblock scalability testing.
Related issue: #62808
```release-note
NONE
```
/sig node
/kind bug
/priority critical-urgent
/assign @shyamjvs @yujuhong
Automatic merge from submit-queue (batch tested with PRs 63417, 64249, 64242, 64128, 64275). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
graduate DynamicKubeletConfig feature to beta
Everything in the https://github.com/kubernetes/features/issues/281 beta target except a few documentation-related items is complete. Docs should be done very soon, so I'm sending this PR to graduate to beta before freeze.
```release-note
The dynamic Kubelet config feature is now beta, and the DynamicKubeletConfig feature gate is on by default. In order to use dynamic Kubelet config, ensure that the Kubelet's --dynamic-config-dir option is set.
```
/cc @luxas
Automatic merge from submit-queue (batch tested with PRs 64060, 63904, 64218, 64208, 64247). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Revert enable PodPreset admission and also enable settings.k8s.io/v1a…
…lpha1 api resource
**What this PR does / why we need it**:
Enable PodPreset admission for there are alpha feature test cases covering it. Simultaneously enable sttings.k8s.io/v1alpha1 api resource.
Fixes#63843
**Release note**:
```release-note
NONE
```
Automatic merge from submit-queue (batch tested with PRs 63434, 64172, 63975, 64180, 63755). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Dump Stack when docker fails on healthcheck
Save stack dump of docker daemon in order to be able to
investigate why docker daemon was unresposive to `docker ps`
See https://github.com/moby/moby/blob/master/daemon/daemon.go on
how docker sets up a trap for SIGUSR1 with `setupDumpStackTrap()`
**What this PR does / why we need it**:
This allows us to investigate why docker daemon was unresponsive to "docker ps" command.
**Special notes for your reviewer**:
Manually tested on Ubuntu and COS.
**Release note**:
```release-note
NONE
```
Send SIGUSR1 to dockerd to save stack dump of docker daemon
in order to be able to investigate why docker daemon was
unresposive to health check done by `docker ps`.
See https://github.com/moby/moby/blob/master/daemon/daemon.go on
how docker sets up a trap for SIGUSR1 with `setupDumpStackTrap()`
Automatic merge from submit-queue (batch tested with PRs 63969, 63902, 63689, 63973, 63978). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Reuse existing CA cert path for kubelet certs
**What this PR does / why we need it**: configure-helper.sh already knows the path to CA cert, re-use that to avoid typos.
**Special notes for your reviewer**:
**Release note**:
```release-note
NONE
```
Automatic merge from submit-queue (batch tested with PRs 63569, 63918, 63980, 63295, 63989). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
New event exporter config with support for new stackdriver resources
New event exporter, with support for use new and old stackdriver resource model.
This should also be cherry-picked to release-1.10 branch, as all fluentd-gcp components support new and stackdriver resource model.
```release-note
Update event-exporter to version v0.2.0 that supports old (gke_container/gce_instance) and new (k8s_container/k8s_node/k8s_pod) stackdriver resources.
```
Automatic merge from submit-queue. If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Use 'docker/default' as default seccomp profile for unprivileged PodSecurityPolicy
**What this PR does / why we need it**:
This PR sets the default seccomp profile for unprivileged PodSecurityPolicy to 'docker/default'. This PR is a followup of [#62662](https://github.com/kubernetes/kubernetes/pull/62662). We are using 'docker/default' instead of 'runtime/default' in addons in order to handle node version skew. When default seccomp profile is applied later, we can remove those annotations.
**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
Fixes#39845
**Special notes for your reviewer**:
**Release note**:
```release-note
NONE
```
Automatic merge from submit-queue. If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Deprecate InfluxDB cluster monitoring
InfluxDB cluster monitoring addon will no longer be supported and will be removed in k8s 1.12.
Default monitoring solution will be changed to `standalone`.
Heapster will still be deployed for backward compatibility of `kubectl top`
```release-note
Stop using InfluxDB as default cluster monitoring
InfluxDB cluster monitoring is deprecated and will be removed in v1.12
```
cc @piosz
Automatic merge from submit-queue. If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
gce: Prefer MASTER_ADVERTISE_ADDRESS in apiserver setup
MASTER_ADVERTISE_ADDRESS is used to set the --advertise-address flag
for the apiserver. It's useful for running the apiserver behind a load
balancer.
However, if PROJECT_ID, TOKEN_URL, TOKEN_BODY, and NODE_NETWORK are
all set, the GCE VM's external IP address will be fetched and used
instead and MASTER_ADVERTISE_ADDRESS will be ignored.
Change this behavior so that MASTER_ADVERTISE_ADDRESS takes precedence
because it's more specific. We still fall back to using the VM's
external IP address if the other variables are set.
Also: Move the setting of --ssh-user and --ssh-keyfile based on
PROXY_SSH_USER) to a top-level block because this is common to all
codepaths.
```release-note
NONE
```
Automatic merge from submit-queue (batch tested with PRs 63167, 63357). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Install and use crictl in gce kube-up.sh
Download and use crictl in gce kube-up.sh.
This PR:
1. Downloads crictl `v1.0.0-beta.0` onto the node, which supports CRI v1alpha2. We'll upgrade it to `v1.0.0-beta.1` soon after the release is cut.
2. Change `kube-docker-monitor` to `kube-container-runtime-monitor`, and let it use `crictl` to do health monitoring.
3. Change `e2e-image-puller` to use `crictl`. Because of https://github.com/kubernetes/kubernetes/issues/63355, it doesn't work now. But in `crictl v1.0.0-beta.1`, we are going to statically link it, and the `e2e-image-puller` should work again.
4. Use `systemctl kill --kill-who=main` instead of `pkill`, the reason is that:
a. `pkill docker` will send `SIGTERM` to all processes including `dockerd`, `docker-containerd`, `docker-containerd-shim`. This is not a problem for Docker 17.03 CE, because `containerd-shim` in containerd 0.2.x doesn't exit with SIGERM (see [code](https://github.com/containerd/containerd/blob/v0.2.x/containerd-shim/main.go#L123)). However, `containerd-shim` in containerd 1.0+ does exit with SIGTERM (see [code](https://github.com/containerd/containerd/blob/master/cmd/containerd-shim/main_unix.go#L200)). This means that `pkill docker` and `pkill containerd` will kill all shim processes for Docker 17.11+ and containerd 1.0+.
b. We can use `pkill -x` instead. However, docker systemd service name is `docker`, but daemon process name is `dockerd`. We have to introduce another environment variable to specify "daemon process name". Given so, it seems easier to just use `systemctl kill` which only requires systemd service name. `systemctl kill --kill-who=main` will make sure only main process receives SIGTERM.
Signed-off-by: Lantao Liu <lantaol@google.com>
/cc @filbranden @yujuhong @feiskyer @mrunalp @kubernetes/sig-node-pr-reviews @kubernetes/sig-cluster-lifecycle-pr-reviews
**Release note**:
```release-note
Kubernetes cluster on GCE have crictl installed now. Users can use it to help debug their node. The documentation of crictl can be found https://github.com/kubernetes-incubator/cri-tools/blob/master/docs/crictl.md.
```
