When doing the very first upgrade from a cluster that contains the
source of truth in the ClusterStatus struct, the new kubeadm logic
will try to retrieve this information from annotations.
This changeset adds to both etcd and apiserver endpoint retrieval the
special case in which they won't retry if we are in such cases. The
logic will retry if we find any unknown error, but will not retry in
the following cases:
- etcd annotations do not contain etcd endpoints, but the overall list
of etcd pods is greater than 0. This means that we listed at least
one etcd pod, but they are missing the annotation.
- API server annotation is not found on the api server pod for a given
node name, but no errors aside from that one were found. This means
that the API server pod is present, but is missing the annotation.
In both cases there is no point in retrying, and so, this speeds up the
upgrade path when coming from a previous existing cluster.
While `ClusterStatus` will be maintained and uploaded, it won't be
used by the internal `kubeadm` logic in order to determine the etcd
endpoints anymore.
The only exception is during the first upgrade cycle (`kubeadm upgrade
apply`, `kubeadm upgrade node`), in which we will fallback to the
ClusterStatus to let the upgrade path add the required annotations to
the newly created static pods.
While kubeadm does not support CA rotation,
the users might still attempt to perform this manually.
For kubeconfig files, updating to a new CA is not reflected
and users need to embed new CA PEM manually.
On kubeconfig cert renewal, always keep the embedded CA
in sync with the one on disk.
Includes a couple of typo fixes.
- Add handlers for service account issuer metadata.
- Add option to manually override JWKS URI.
- Add unit and integration tests.
- Add a separate ServiceAccountIssuerDiscovery feature gate.
Additional notes:
- If not explicitly overridden, the JWKS URI will be based on
the API server's external address and port.
- The metadata server is configured with the validating key set rather
than the signing key set. This allows for key rotation because tokens
can still be validated by the keys exposed in the JWKs URL, even if the
signing key has been rotated (note this may still be a short window if
tokens have short lifetimes).
- The trust model of OIDC discovery requires that the relying party
fetch the issuer metadata via HTTPS; the trust of the issuer metadata
comes from the server presenting a TLS certificate with a trust chain
back to the from the relying party's root(s) of trust. For tests, we use
a local issuer (https://kubernetes.default.svc) for the certificate
so that workloads within the cluster can authenticate it when fetching
OIDC metadata. An API server cannot validly claim https://kubernetes.io,
but within the cluster, it is the authority for kubernetes.default.svc,
according to the in-cluster config.
Co-authored-by: Michael Taufen <mtaufen@google.com>
It turns out that the dual-stack feature enabled doesn't mean that
the cluster MUST be dual-stack, it only indicates that it MAY be
dual-stack but CAN be single-stack.
We should relax the validation to allow single-stack clusters
with dual-stack enabled.
If a Node name in the cluster is already taken and this Node is Ready,
prevent TLS bootsrap on "kubeadm join" and exit early.
This change requires that a new ClusterRole is granted to the
"system:bootstrappers:kubeadm:default-node-token" group to be
able get Nodes in the cluster. The same group already has access
to obtain objects such as the KubeletConfiguration and kubeadm's
ClusterConfiguration.
The motivation of this change is to prevent undefined behavior
and the potential control-plane breakdown if such a cluster
is racing to have two nodes with the same name for long periods
of time.
The following values are validated in the following precedence
from lower to higher:
- actual hostname
- NodeRegistration.Name (or "--node-name") from JoinConfiguration
- "--hostname-override" passed via kubeletExtraArgs
If the user decides to not let kubeadm know about a custom node name
and to instead override the hostname from a kubelet systemd unit file,
kubeadm will not be able to detect the problem.
- Extend the exponential backoff for add/remove/... retry to
11 steps ~=106 seconds. From experiments for 3 and more members
the race can take more that ~=26 seconds.
- Increase the dialTimeout for client creation to 40 seconds.
20 seconds seems racy for 3 and more members.
For the etcd client, amend AddMember() to handle a very
rare bug when multiple members can end up with the same
name. Match the member peer address and assign it the name of
the member we are adding. For the rest of the members with missing
names use their member IDs as name. The etcd node is not disrupted
by the unknown names.
The important aspects are:
- The number of members of the initial cluster must match
the members in the cluster.
- The member we are current adding is present in the initial cluster.
As part of #68522, Switching off the cAdvisor v1 Json API that we expose
directly. These include /stats/, /stats/container, /stats/{podName}/{containerName},
and /stats/{namespace}/{podName}/{uid}/{containerName}
The CoreDNS GA feature-gate in kubeadm was deprecated since 1.13.
The k8s policy is to remove the gate 2 releases after it transitions
to GA:
https://kubernetes.io/docs/reference/using-api/deprecation-policy/#deprecation
We kept it around for longer to prevent existing setups from breaking
as it caused minimal maintenance overhead.
This creates a new EndpointSliceProxying feature gate to cover EndpointSlice
consumption (kube-proxy) and allow the existing EndpointSlice feature gate to
focus on EndpointSlice production only. Along with that addition, this enables
the EndpointSlice feature gate by default, now only affecting the controller.
The rationale here is that it's really difficult to guarantee all EndpointSlices
are created in a cluster upgrade process before kube-proxy attempts to consume
them. Although masters are generally upgraded before nodes, and in most cases,
the controller would have enough time to create EndpointSlices before a new node
with kube-proxy spun up, there are plenty of edge cases where that might not be
the case. The primary limitation on EndpointSlice creation is the API rate limit
of 20QPS. In clusters with a lot of endpoints and/or with a lot of other API
requests, it could be difficult to create all the EndpointSlices before a new
node with kube-proxy targeting EndpointSlices spun up.
Separating this into 2 feature gates allows for a more gradual rollout with the
EndpointSlice controller being enabled by default in 1.18, and EndpointSlices
for kube-proxy being enabled by default in the next release.
The warning message
```
[config] WARNING: Ignored YAML document with GroupVersionKind ...
```
is printed for all GVKs that are not part of the kubeadm core types.
This is wrong as the component config types are supported and successfully
parsed and used despite the fact that the warning is printed for them too.
Hence this simple fix first checks if the group of the GVK is a supported
component config group and the warning is printed only if it's not.
Signed-off-by: Rostislav M. Georgiev <rostislavg@vmware.com>
kubeadm deploys the apiserver, controller-manager and the scheduler
using liveness probes.
The bind-address option is used to configure the probe address, in
case this is configured with an unspecified address, the probe
will fail. When using an unspecified address the probe host field is
left empty, otherwise the bind-address is used.
The function validateKubeConfig() can end up comparing
a user generated kubeconfig to a kubeconfig generated by kubeadm.
If a user kubeconfig has a CA that is base64 encoded with whitespace,
if said kubeconfig is loaded using clientcmd.LoadFromFile()
the CertificateAuthorityData bytes will be decoded from base64
and placed in the v1.Config raw. On the other hand a kubeconfig
generated by kubeadm will have the ca.crt parsed to a Certificate
object with whitespace ignored in the PEM input.
Make sure that validateKubeConfig() tolerates whitespace differences
when comparing CertificateAuthorityData.
kubeadm removed the deprecated "--address" flag for controller-manager
and scheduler in favor of "--bind-address"
We should use bind-address to configure the manifest probe addresses.
If the user has modified the kubelet.conf post TLS bootstrap
to become invalid, the function getNodeNameFromKubeletConfig() can
panic. This was observed to trigger in "kubeadm reset" use cases.
Add basic validation and unit tests around parsing the kubelet.conf
with the aforementioned function.
Currently if the controlplane fails to init, we print out a message
with some example commands that only show docker CLI.
This tries to improve that by printing the example commands for
docker, cri-o and containerd by checking the socket looking for
the default docker socket.
switch api helper functions to v1 CRD api
switch v1 CRD for apiserver internal
switch to v1 CRD for internal controllers
api storage/validation related changes
move local-defaulting utils private to prevent spreading
boilerplate
keep the subresource status/scale spec nil unless it's enabled
clean up empty space
CreateOrMutateConfigMap was not resilient when it was trying to Create
the ConfigMap. If this operation returned an unknown error the whole
operation would fail, because it was strict in what error it was
expecting right afterwards: if the error returned by the Create call
was a IsAlreadyExists error, it would work fine. However, if an
unexpected error (such as an EOF) happened, this call would fail.
We are seeing this error specially when running control plane node
joins in an automated fashion, where things happen at a relatively
high speed pace.
It was specially easy to reproduce with kind, with several control
plane instances. E.g.:
```
[upload-config] Storing the configuration used in ConfigMap "kubeadm-config" in the "kube-system" Namespace
I1130 11:43:42.788952 887 round_trippers.go:443] POST https://172.17.0.2:6443/api/v1/namespaces/kube-system/configmaps?timeout=10s in 1013 milliseconds
Post https://172.17.0.2:6443/api/v1/namespaces/kube-system/configmaps?timeout=10s: unexpected EOF
unable to create ConfigMap
k8s.io/kubernetes/cmd/kubeadm/app/util/apiclient.CreateOrMutateConfigMap
/go/src/k8s.io/kubernetes/_output/dockerized/go/src/k8s.io/kubernetes/cmd/kubeadm/app/util/apiclient/idempotency.go:65
```
This change makes this logic more resilient to unknown errors. It will
retry on the light of unknown errors until some of the expected error
happens: either `IsAlreadyExists`, in which case we will mutate the
ConfigMap, or no error, in which case the ConfigMap has been created.
kubeadm always use the IPv4 localhost address by defaultA for etcd
The probe hostname is obtained before the generation of the etcd
parameters, so it can't detect the right IP familiy for the
host of the probe.
This causes that with IPv6 clusters doesn't work because the probe
uses the IPv4 localhost address.
This patchs configures the right localhost address based on the used
AdvertiseAddress IP family.
- Use `schema.TypeMeta` instead of custom `struct` for VK
- More strict check on GVK after `Interpret` in `SplitYAMLDocuments`
- Adjust `Interpret` comment to include JSON
- Add retrieveValidatedConfigInfo to be able to better unit
test the function.
- Break some of the logic in RetrieveValidatedConfigInfo into
helper functions.
- Pass JoinConfiguration.Discovery to RetrieveValidatedConfigInfo
instead of JoinConfiguration.
- Use the discovery timeout per API call to fetch cluster-info
(optionally the user value can be slit in 2).
- Add detailed unit tests for retrieveValidatedConfigInfo.
kubeadm's current implementation of component config support is "kind" centric.
This has its downsides. Namely:
- Kind names and numbers can change between config versions.
Newer kinds can be ignored. Therefore, detection of a version change is
considerably harder.
- A component config can have only one kind that is managed by kubeadm.
Thus a more appropriate way to identify component configs is required.
Probably the best solution identified so far is a config group.
A group name is unlikely to change between versions, while the kind names and
structure can.
Tracking component configs by group name allows us to:
- Spot more easily config version changes and manage alternate versions.
- Support more than one kind in a config group/version.
- Abstract component configs by hiding their exact structure.
Hence, this change rips off the old kind based support for component configs
and replaces it with a group name based one. This also has the following
extra benefits:
- More tests were added.
- kubeadm now errors out if an unsupported version of a known component group
is used.
Signed-off-by: Rostislav M. Georgiev <rostislavg@vmware.com>
- Add a new preflight check for upgrade that runs the pause container
with -v in a Job.
- Wait for the Job to complete and return an error after N seconds.
- Manually clean the Job because we don't have the TTL controller
enabled in kubeadm yet (it's still alpha).
Currently this uses the combined kubelet output (stdout + stderr), but this
causes parsing issues if the kubelet logs something on stderr.
Thus we ignore the entire stderr and use stdout only.
We do disable a couple of tests here. That is because the fakeexecer only
supports combined output and return a "not supported" error if `.Output()`
gets invoked thus permanently failing those.
Signed-off-by: Rostislav M. Georgiev <rostislavg@vmware.com>
This patch removes pkg/util/mount completely, and replaces it with the
mount package now located at k8s.io/utils/mount. The code found at
k8s.io/utils/mount was moved there from pkg/util/mount, so the code is
identical, just no longer in-tree to k/k.
update tests
add comment
amend var name
update comment
add check for empty slice
fix tests
fix mask size in test
review feedback
add ipv4 and ipv6 flag for mask sizes
add to violation exception list
remove import alias
run update-openapi-spec
review feedback
run update-bazel
review feedback
review feedback
