This allows us to interrupt/kill the executed command if it exceeds the
timeout (not implemented by this commit).
Set timeout in Exec probes. HTTPGet and TCPSocket probes respect the
timeout, while Exec probes used to ignore it.
Add e2e test for exec probe with timeout. However, the test is skipped
while the default exec handler doesn't support timeouts.
Automatic merge from submit-queue
CRI: Add Status into CRI.
For https://github.com/kubernetes/kubernetes/issues/35701.
Fixes https://github.com/kubernetes/kubernetes/issues/35701.
This PR added a `Status` call in CRI, and the `RuntimeStatus` is defined as following:
``` protobuf
message RuntimeCondition {
// Type of runtime condition.
optional string type = 1;
// Status of the condition, one of true/false.
optional bool status = 2;
// Brief reason for the condition's last transition.
optional string reason = 3;
// Human readable message indicating details about last transition.
optional string message = 4;
}
message RuntimeStatus {
// Conditions is an array of current observed runtime conditions.
repeated RuntimeCondition conditions = 1;
}
```
Currently, only `conditions` is included in `RuntimeStatus`, and the definition is almost the same with `NodeCondition` and `PodCondition` in K8s api.
@yujuhong @feiskyer @bprashanth If this makes sense, I'll send a follow up PR to let dockershim return `RuntimeStatus` and let kubelet make use of it.
@yifan-gu @euank Does this make sense to rkt?
/cc @kubernetes/sig-node
Automatic merge from submit-queue
Initial work on running windows containers on Kubernetes
<!-- Thanks for sending a pull request! Here are some tips for you:
1. If this is your first time, read our contributor guidelines https://github.com/kubernetes/kubernetes/blob/master/CONTRIBUTING.md and developer guide https://github.com/kubernetes/kubernetes/blob/master/docs/devel/development.md
2. If you want *faster* PR reviews, read how: https://github.com/kubernetes/kubernetes/blob/master/docs/devel/faster_reviews.md
3. Follow the instructions for writing a release note: https://github.com/kubernetes/kubernetes/blob/master/docs/devel/pull-requests.md#release-notes
-->
This is the first stab at getting the Kubelet running on Windows (fixes#30279), and getting it to deploy network-accessible pods that consist of Windows containers. Thanks @csrwng, @jbhurat for helping out.
The main challenge with Windows containers at this point is that container networking is not supported. In other words, each container in the pod will get it's own IP address. For this reason, we had to make a couple of changes to the kubelet when it comes to setting the pod's IP in the Pod Status. Instead of using the infra-container's IP, we use the IP address of the first container.
Other approaches we investigated involved "disabling" the infra container, either conditionally on `runtime.GOOS` or having a separate windows-docker container runtime that re-implemented some of the methods (would require some refactoring to avoid maintainability nightmare).
Other changes:
- The default docker endpoint was removed. This results in the docker client using the default for the specific underlying OS.
More detailed documentation on how to setup the Windows kubelet can be found at https://docs.google.com/document/d/1IjwqpwuRdwcuWXuPSxP-uIz0eoJNfAJ9MWwfY20uH3Q.
cc: @ikester @brendandburns @jstarks
Automatic merge from submit-queue
[AppArmor] Hold bad AppArmor pods in pending rather than rejecting
Fixes https://github.com/kubernetes/kubernetes/issues/32837
Overview of the fix:
If the Kubelet needs to reject a Pod for a reason that the control plane doesn't understand (e.g. which AppArmor profiles are installed on the node), then it might contiinuously try to run the pod on the same rejecting node. This change adds a concept of "soft rejection", in which the Pod is admitted, but not allowed to run (and therefore held in a pending state). This prevents the pod from being retried on other nodes, but also prevents the high churn. This is consistent with how other missing local resources (e.g. volumes) is handled.
A side effect of the change is that Pods which are not initially runnable will be retried. This is desired behavior since it avoids a race condition when a new node is brought up but the AppArmor profiles have not yet been loaded on it.
``` release-note
Pods with invalid AppArmor configurations will be held in a Pending state, rather than rejected (failed). Check the pod status message to find out why it is not running.
```
@kubernetes/sig-node @timothysc @rrati @davidopp
Automatic merge from submit-queue
Separate Direct and Indirect streaming paths, implement indirect path for CRI
This PR refactors the `pkg/kubelet/container.Runtime` interface to remove the `ExecInContainer`, `PortForward` and `AttachContainer` methods. Instead, those methods are part of the `DirectStreamingRuntime` interface which all "legacy" runtimes implement. I also added an `IndirectStreamingRuntime` which handles the redirect path and is implemented by CRI runtimes. To control the size of this PR, I did not fully setup the indirect streaming path for the dockershim, so I left legacy path behind.
Most of this PR is moving & renaming associated with the refactoring. To understand the functional changes, I suggest tracing the code from `getExec` in `pkg/kubelet/server/server.go`, which calls `GetExec` in `pkg/kubelet/kubelet_pods.go` to determine whether to follow the direct or indirect path.
For https://github.com/kubernetes/kubernetes/issues/29579
/cc @kubernetes/sig-node
Automatic merge from submit-queue
CRI: Add devices to ContainerConfig
This PR adds devices to ContainerConfig and adds experimental GPU support.
cc/ @yujuhong @Hui-Zhi @vishh @kubernetes/sig-node
Automatic merge from submit-queue
Only set sysctls for infra containers
We did set the sysctls for each container in a pod. This opens up a way to set un-whitelisted sysctls during upgrade from v1.3:
- set annotation in v1.3 with an un-whitelisted sysctl. Set restartPolicy=Always
- upgrade cluster to v1.4
- kill container process
- un-whitelisted sysctl is set on restart of the killed container.
Automatic merge from submit-queue
SELinux Overhaul
Overhauls handling of SELinux in Kubernetes. TLDR: Kubelet dir no longer has to be labeled `svirt_sandbox_file_t`.
Fixes#33351 and #33510. Implements #33951.
If the infra container has already terminated, `docker inspect` will report
pid 0. The path constructed using the pid to check the network namespace of
the process will be invalid. This commit changes docker to report an empty
path to stop kubenet from erroring out whenever TearDown is called on an
exited infra container.
This is not a fix for all the plugins, as some plugins may require the actual
network namespace to tear down properly.
Previously, we used the docker config digest (also called "image ID"
by Docker) for the value of the `ImageID` field in the container status.
This was not particularly useful, since the config manifest is not
what's used to identify the image in a registry, which uses the manifest
digest instead. Docker 1.12+ always populates the RepoDigests field
with the manifest digests, and Docker 1.10 and 1.11 populate it when
images are pulled by digest.
This commit changes `ImageID` to point to the the manifest digest when
available, using the prefix `docker-pullable://` (instead of
`docker://`)
Previously, the `InspectImage` method of the Docker interface expected a
"pullable" image ref (name, tag, or manifest digest). If you tried to
inspect an image by its ID (config digest), the inspect would fail to
validate the image against the input identifier. This commit changes
the original method to be named `InspectImageByRef`, and introduces a
new method called `InspectImageByID` which validates that the input
identifier was an image ID.
Previously, we used the docker config digest (also called "image ID"
by Docker) for the value of the `ImageID` field in the container status.
This was not particularly useful, since the config manifest is not
what's used to identify the image in a registry, which uses the manifest
digest instead. Docker 1.12+ always populates the RepoDigests field
with the manifest digests, and Docker 1.10 and 1.11 populate it when
images are pulled by digest.
This commit changes `ImageID` to point to the the manifest digest when
available, using the prefix `docker-pullable://` (instead of
`docker://`)
Previously, the `InspectImage` method of the Docker interface expected a
"pullable" image ref (name, tag, or manifest digest). If you tried to
inspect an image by its ID (config digest), the inspect would fail to
validate the image against the input identifier. This commit changes
the original method to be named `InspectImageByRef`, and introduces a
new method called `InspectImageByID` which validates that the input
identifier was an image ID.
Automatic merge from submit-queue
Kubelet: fix port forward for dockershim
This PR fixes port forward for dockershim and also adds a `kubecontainer.FormatPod`.
Locally cluster has passed `--ginkgo.focus=Port\sforwarding'` tests.
cc/ @Random-Liu @yujuhong
Automatic merge from submit-queue
Add seccomp and apparmor support.
This PR adds seccomp and apparmor support in new CRI.
This a WIP because I'm still adding unit test for some of the functions. Sent this PR here for design discussion.
This PR is similar with https://github.com/kubernetes/kubernetes/pull/33450.
The differences are:
* This PR passes seccomp and apparmor configuration via annotations;
* This PR keeps the seccomp handling logic in docker shim because current seccomp implementation is very docker specific, and @timstclair told me that even the json seccomp profile file is defined by docker.
Notice that this PR almost passes related annotations in `api.Pod` to the runtime directly instead of introducing new CRI annotation.
@yujuhong @feiskyer @timstclair
This allows runtimes in different packages (dockertools, rkt, kuberuntime) to
share the same logic. Before this change, only dockertools support this
feature. Now all three packages support image pull throttling.
Automatic merge from submit-queue
dockertools: Don't use network plugin if net=host
I'm pretty sure this was just an oversight the first time around.
Before: `E0815 18:06:17.627468 976 docker_manager.go:350] NetworkPlugin kubenet failed on the status hook for pod 'sleep' - Unexpected command output Device "eth0" does not exist.`
After: No such logline is printed
The pod IP reported in `describe` is the same either way
cc @kubernetes/sig-node
This was already handled in most places. I think this is the only
remaining instance of it in the docker package.
This could lead to confusing results. E.g. if `networkPlugin` was cni,
it could lead to error logs about not getting network status for host
pods if eth0 didn't exist on the host.
Automatic merge from submit-queue
Set pod state as "unknown" when CNI plugin fails
Before this change, CNI plugin failure didn't change anything in the pod status, so pods having containers without requested network were "running".
Fixes#29148
Before this change, CNI plugin failure didn't change anything in
the pod status, so pods having containers without requested
network were "running".
Fixes#29148
Add a new docker integration with kubelet using the new runtime API.
This change adds the package with some skeletons, and implements some
of the basic operations.
Automatic merge from submit-queue
Allow PVs to specify supplemental GIDs
Retry of https://github.com/kubernetes/kubernetes/pull/28691 . Adds a Kubelet helper function for getting extra supplemental groups
Automatic merge from submit-queue
Removing images with multiple tags
If an image has multiple tags, we need to remove all the tags in order to make docker image removing successful.
#28491
Automatic merge from submit-queue
ImagePuller refactoring
A plain refactoring
- Moving image pullers to a new pkg/kubelet/images directory
- Hiding image pullers inside the new ImageManager
The next step is to consolidate the logic of the serialized and the parallel image pullers inside ImageManager
xref: #25577
Automatic merge from submit-queue
Kubelet: Set PruneChildren when removing image.
This is a bug introduced during switching to engine-api. https://github.com/kubernetes/kubernetes/issues/23563.
When removing image, there is an option `noprune`:
```
If prune is true, ancestor images will each attempt to be deleted quietly.
```
In go-dockerclient, the default value of the option is ["noprune=false"](https://github.com/fsouza/go-dockerclient/blob/master/image.go#L171), which means that ancestor images should be also removed. This is the expected behaviour.
However in engine-api, the option is changed to `PruneChildren`, and the default value is `PruneChildren=false`, which means that ancestor images won't be removed.
This makes `ImageRemove` only remove the first layer of the image, which causes the image garbage collection not working as expected.
This should be fixed in 1.3.
And thanks to @ronnielai for finding the bug! :)
/cc @kubernetes/sig-node
Automatic merge from submit-queue
docker_manager: Correct determineContainerIP args
This could result in the network plugin not retrieving the pod ip in a
call to SyncPod when using the `exec` network plugin.
The CNI and kubenet network plugins ignore the name/namespace arguments,
so they are not impacted by this bug.
I verified the second included test failed prior to correcting the
argument order.
Fixes#29161
cc @yujuhong
Automatic merge from submit-queue
Delete redundant if condition
The case `containerStatus == nil` has already been checked just above. It's redundant here.
This could result in the network plugin not retrieving the pod ip in a
call to SyncPod when using the `exec` network plugin.
The CNI and kubenet network plugins ignore the name/namespace arguments,
so they are not impacted by this bug.
I verified the second included test failed prior to correcting the
argument order.
Fixes#29161
Automatic merge from submit-queue
Support terminal resizing for exec/attach/run
```release-note
Add support for terminal resizing for exec, attach, and run. Note that for Docker, exec sessions
inherit the environment from the primary process, so if the container was created with tty=false,
that means the exec session's TERM variable will default to "dumb". Users can override this by
setting TERM=xterm (or whatever is appropriate) to get the correct "smart" terminal behavior.
```
Fixes#13585
Add support for terminal resizing for exec, attach, and run. Note that for Docker, exec sessions
inherit the environment from the primary process, so if the container was created with tty=false,
that means the exec session's TERM variable will default to "dumb". Users can override this by
setting TERM=xterm (or whatever is appropriate) to get the correct "smart" terminal behavior.
Automatic merge from submit-queue
Remove unnecessary calls to api.GetReference
These calls are unnecessary, can be removed. `Eventf` and others just call `GetReference` on the object they are passed.
cc @kubernetes/sig-node
