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>.
unpack dynamic kubelet config payloads to files
This PR unpacks the downloaded ConfigMap to a set of files on the node.
This enables other config files to ride alongside the
KubeletConfiguration, and the KubeletConfiguration to refer to these
cohabitants with relative paths.
This PR also stops storing dynamic config metadata (e.g. current,
last-known-good config records) in the same directory as config
checkpoints. Instead, it splits the storage into `meta` and
`checkpoints` dirs.
The current store dir structure is as follows:
```
- dir named by --dynamic-config-dir (root for managing dynamic config)
| - meta (dir for metadata, e.g. which config source is currently assigned, last-known-good)
| - current (a serialized v1 NodeConfigSource object, indicating the assigned config)
| - last-known-good (a serialized v1 NodeConfigSource object, indicating the last-known-good config)
| - checkpoints (dir for config checkpoints)
| - uid1 (dir for unpacked config, identified by uid1)
| - file1
| - file2
| - ...
| - uid2
| - ...
```
There are some likely changes to the above structure before dynamic config goes beta, such as renaming "current" to "assigned" for clarity, and extending the checkpoint identifier to include a resource version, as part of resolving #61643.
```release-note
NONE
```
/cc @luxas @smarterclayton
This PR unpacks the downloaded ConfigMap to a set of files on the node.
This enables other config files to ride alongside the
KubeletConfiguration, and the KubeletConfiguration to refer to these
cohabitants with relative paths.
This PR also stops storing dynamic config metadata (e.g. current,
last-known-good config records) in the same directory as config
checkpoints. Instead, it splits the storage into `meta` and
`checkpoints` dirs.
Automatic merge from submit-queue (batch tested with PRs 62192, 61866, 62206, 62360). 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 rkt references in the codebase
```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>.
Fixes restartKubelet in test/e2e_node failure.
Looks like there is some recent change on how we start kubelet service
in test_e2e_node. Fixes restartKubelet() to get right kubelet service
name to cope with the change.
**What this PR does / why we need it**:
**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 kubelet-serial-gce-e2e test failure:
https://k8s-testgrid.appspot.com/wg-resource-management#kubelet-serial-gce-e2e
Thanks a lot to @mindprince for noticing this!
**Special notes for your reviewer**:
**Release note**:
```release-note
```
Looks like there is some recent change on how we start kubelet service
in test_e2e_node. Fixes restartKubelet() to get right kubelet service
name to cope with the change.
Automatic merge from submit-queue (batch tested with PRs 61498, 62030). 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>.
Delete in-tree support for NVIDIA GPUs.
This removes the alpha Accelerators feature gate which was deprecated in 1.10 (#57384).
The alternative feature DevicePlugins went beta in 1.10 (#60170).
Fixes#54012
```release-note
Support for "alpha.kubernetes.io/nvidia-gpu" resource which was deprecated in 1.10 is removed. Please use the resource exposed by DevicePlugins instead ("nvidia.com/gpu").
```
Automatic merge from submit-queue (batch tested with PRs 61482, 61740). 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>.
Make systemd service name for kubelet use a timestamp in e2e-node tests.
**What this PR does / why we need it**:
This makes it easier to figure out which execution was last when looking at the output of `systemd list-units kubelet-*.service`.
We try to find the name of the /tmp/node-e2e-* directory and use the same timestamp if we can. Otherwise, we just call Now() again, which isn't as nice (as the unit name and directory name will not match) but will still produce unit names that will be ordered when launching multiple subsequent executions on the same host.
**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
N/A
**Special notes for your reviewer**:
Tested using `make test-e2e-node REMOTE=true` and then checking `systemctl list-units kubelet-*.service` on the target host.
```
$ systemctl list-units kubelet-*.service
kubelet-20180326T142016.service loaded active exited /tmp/node-e2e-20180326T142016/kubelet --kubeconfig /tmp/node-e2e-20180326T142016/kubeconfig --root-dir /var/lib/kubelet ...
kubelet-20180326T143550.service loaded active exited /tmp/node-e2e-20180326T143550/kubelet --kubeconfig /tmp/node-e2e-20180326T143550/kubeconfig --root-dir /var/lib/kubelet ...
```
The units are sorted in the order they were launched.
**Release note**:
```release-note
NONE
```
The numbers will only be available when docker.service has its own
memory and cpu cgroups, which doesn't necessarily happen unless the unit
has Delegate=yes configured.
Let's work around that by checking the status of Delegate, in the case
where we are:
* running Docker
* running Systemd
* able to check the status through systemctl
* the status is explicitly Delegate=no (the default)
If all of those are true, let's make CPU and Memory expectations
optional.
Tested: make test-e2e-node REMOTE=true HOSTS=centos-e2e-node FOCUS="Summary API"
This is necessary to show any RootFs usage on systems where the backing
filesystem of overlay2 is xfs.
The current test only created directories (for mount points) in the
upper layer of the overlay. Outside of the mount namespace, only the
directories are visible. When running `du` on those, usually filesystems
will show some usage, but not xfs, which shows a disk usage of 0 for
directories.
Fix this by creating a file in the root directory, outside the volumes,
in order to trigger some disk usage that can be measured by `du`.
Tested: make test-e2e-node REMOTE=true HOSTS=centos-e2e-node FOCUS="Summary API"
Automatic merge from submit-queue (batch tested with PRs 61829, 61908, 61307, 61872, 60100). 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>.
node authorizer sets up access rules for dynamic config
This PR makes the node authorizer automatically set up access rules for
dynamic Kubelet config.
I also added some validation to the node strategy, which I discovered we
were missing while writing this.
This PR is based on another WIP from @liggitt.
```release-note
The node authorizer now automatically sets up rules for Node.Spec.ConfigSource when the DynamicKubeletConfig feature gate is enabled.
```
This makes it easier to figure out which execution was last when looking
at the output of `systemd list-units kubelet-*.service`.
We try to find the name of the /tmp/node-e2e-* directory and use the
same timestamp if we can. Otherwise, we just call Now() again, which
isn't as nice (as the unit name and directory name will not match) but
will still produce unit names that will be ordered when launching
multiple subsequent executions on the same host.
Curl is more ubiquitous than wget. For instance, the GCE centos-7 and
rhel-7 image families ship curl by default, but not wget.
Looking at the shell scripts under cluster/, they tend to use curl more
than wget. (The ones that use wget, such as get-kube.sh, try curl first
and only fallback to wget if it's not available.)
Tested: by running node-e2e-test on Ubuntu, COS and CentOS.
This PR makes the node authorizer automatically set up access rules for
dynamic Kubelet config.
I also added some validation to the node strategy, which I discovered we
were missing while writing this.
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>.
Replace package "golang.org/x/net/context" with "context"
**What this PR does / why we need it**:
Replace package "golang.org/x/net/context" with "context"
**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#60560
**Special notes for your reviewer**:
As of Go 1.7 this package(golang.org/x/net/context) is available in the standard library under the name context. see (https://godoc.org/golang.org/x/net/context)
It is almost machinery replace.
**Release note**:
```release-note
NONE
```
Automatic merge from submit-queue (batch tested with PRs 61378, 60915, 61499, 61507, 61478). 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>.
Capture pod startup phases as metrics
Learning from https://github.com/kubernetes/kubernetes/issues/60589, we should also start collecting and graphing sub-parts of pod-startup latency.
/sig scalability
/kind feature
/priority important-soon
/cc @wojtek-t
```release-note
NONE
```
Automatic merge from submit-queue (batch tested with PRs 59740, 59728, 60080, 60086, 58714). 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>.
more concise to merge the slice
**What this PR does / why we need it**:
more concise to merge the slice
**Special notes for your reviewer**:
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>.
Fixes the races around devicemanager Allocate() and endpoint deletion.
There is a race in predicateAdmitHandler Admit() that getNodeAnyWayFunc()
could get Node with non-zero deviceplugin resource allocatable for a
non-existing endpoint. That race can happen when a device plugin fails,
but is more likely when kubelet restarts as with the current registration
model, there is a time gap between kubelet restart and device plugin
re-registration. During this time window, even though devicemanager could
have removed the resource initially during GetCapacity() call, Kubelet
may overwrite the device plugin resource capacity/allocatable with the
old value when node update from the API server comes in later. This
could cause a pod to be started without proper device runtime config set.
To solve this problem, introduce endpointStopGracePeriod. When a device
plugin fails, don't immediately remove the endpoint but set stopTime in
its endpoint. During kubelet restart, create endpoints with stopTime set
for any checkpointed registered resource. The endpoint is considered to be
in stopGracePeriod if its stoptime is set. This allows us to track what
resources should be handled by devicemanager during the time gap.
When an endpoint's stopGracePeriod expires, we remove the endpoint and
its resource. This allows the resource to be exported through other channels
(e.g., by directly updating node status through API server) if there is such
use case. Currently endpointStopGracePeriod is set as 5 minutes.
Given that an endpoint is no longer immediately removed upon disconnection,
mark all its devices unhealthy so that we can signal the resource allocatable
change to the scheduler to avoid scheduling more pods to the node.
When a device plugin endpoint is in stopGracePeriod, pods requesting the
corresponding resource will fail admission handler.
Tested:
Ran GPUDevicePlugin e2e_node test 100 times and all passed now.
**What this PR does / why we need it**:
**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 https://github.com/kubernetes/kubernetes/issues/60176
**Special notes for your reviewer**:
**Release note**:
```release-note
Fixes the races around devicemanager Allocate() and endpoint deletion.
```
There is a race in predicateAdmitHandler Admit() that getNodeAnyWayFunc()
could get Node with non-zero deviceplugin resource allocatable for a
non-existing endpoint. That race can happen when a device plugin fails,
but is more likely when kubelet restarts as with the current registration
model, there is a time gap between kubelet restart and device plugin
re-registration. During this time window, even though devicemanager could
have removed the resource initially during GetCapacity() call, Kubelet
may overwrite the device plugin resource capacity/allocatable with the
old value when node update from the API server comes in later. This
could cause a pod to be started without proper device runtime config set.
To solve this problem, introduce endpointStopGracePeriod. When a device
plugin fails, don't immediately remove the endpoint but set stopTime in
its endpoint. During kubelet restart, create endpoints with stopTime set
for any checkpointed registered resource. The endpoint is considered to be
in stopGracePeriod if its stoptime is set. This allows us to track what
resources should be handled by devicemanager during the time gap.
When an endpoint's stopGracePeriod expires, we remove the endpoint and
its resource. This allows the resource to be exported through other channels
(e.g., by directly updating node status through API server) if there is such
use case. Currently endpointStopGracePeriod is set as 5 minutes.
Given that an endpoint is no longer immediately removed upon disconnection,
mark all its devices unhealthy so that we can signal the resource allocatable
change to the scheduler to avoid scheduling more pods to the node.
When a device plugin endpoint is in stopGracePeriod, pods requesting the
corresponding resource will fail admission handler.
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 node-e2e test for ShareProcessNamespace
**What this PR does / why we need it**: Adds a node-e2e test for kubernetes/features#495
**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#59554
**Special notes for your reviewer**: This requires a feature gate to be enabled in both the kubelet and API server. I'm not sure which jenkins configs need to be updated (or if these are even still used) so I just updated a pile of them.
opened kubernetes/test-infra#7030 for https://github.com/kubernetes/test-infra/blob/master/jobs/config.json
**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>.
fix todo:Move function readinessCheck to util
**What this PR does / why we need it**:
fix todo:Move function readinessCheck to util in test/e2e_node/services
**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 60376, 55584, 60358, 54631, 60291). 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 gcloud docker -- since it's deprecated
docker handles this now and it raises an error.
try 3
```release-note
NONE
```
Automatic merge from submit-queue (batch tested with PRs 60236, 60332, 57375, 60451, 57408). 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>.
Update device plugin e2e_node test to not changing Kubelet config
as DevicePlugins feature is enabled by default now.
**What this PR does / why we need it**:
**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
```
Automatic merge from submit-queue (batch tested with PRs 59159, 60318, 60079, 59371, 57415). 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>.
Made a couple API changes to deviceplugin/v1beta1 to avoid future
incompatible API changes:
- Add GetDevicePluginOptions rpc call. This is needed when we switch
from Registration service to probe-based plugin watcher.
- Change AllocateRequest and AllocateResponse to allow device requests
from multiple containers in a pod. Currently only made mechanical
change on the devicemanager and test code to cope with the API but
still issues an Allocate call per container. We can modify the
devicemanager in 1.11 to issue a single Allocate call per pod.
The change will also facilitate incremental API change to communicate
pod level information through Allocate rpc if there is such future
need.
**What this PR does / why we need it**:
Made a couple API changes to deviceplugin/v1beta1 to avoid future incompatible API changes.
**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 https://github.com/kubernetes/kubernetes/issues/59370
**Special notes for your reviewer**:
**Release note**:
```release-note
```
Automatic merge from submit-queue (batch tested with PRs 60324, 60269, 59771, 60314, 59941). 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>.
expunge the word 'manifest' from Kubelet's config API
The word 'manifest' technically refers to a container-group specification
that predated the Pod abstraction. We should avoid using this legacy
terminology where possible. Fortunately, the Kubelet's config API will
be beta in 1.10 for the first time, so we still had the chance to make
this change.
I left the flags alone, since they're deprecated anyway.
I changed a few var names in files I touched too, but this PR is the
just the first shot, not the whole campaign
(`git grep -i manifest | wc -l -> 1248`).
```release-note
Some field names in the Kubelet's now v1beta1 config API differ from the v1alpha1 API: PodManifestPath is renamed to PodPath, ManifestURL is renamed to PodURL, ManifestURLHeader is renamed to PodURLHeader.
```
