Kubeadm requires manual version updates of its current supported k8s
control plane version and minimally supported k8s control plane and
kubelet versions every release cycle.
To avoid that, in constants.go:
- Add the helper function getSkewedKubernetesVersion() that can be
used to retrieve a MAJOR.(MINOR+n).0 version of k8s. It currently
uses the kubeadm version populated in "component-base/version" during
the kubeadm build process.
- Use the function to set existing version constants (variables).
Update util/config/common.go#NormalizeKubernetesVersion() to
tolerate the case where a k8s version in the ClusterConfiguration
is too old for the kubeadm binary to use during code freeze.
Include unit tests for the new utilities.
This change optimizes the kubeadm/etcd `AddMember` client-side function
by stopping early in the backoff loop when a peer conflict is found
(indicating the member has already been added to the etcd cluster). In
this situation, the function will stop early and relay a call to
`ListMembers` to fetch the current list of members to return. With this
optimization, front-loading a `ListMembers` call is no longer necessary,
as this functionally returns the equivalent response.
This helps reduce the amount of time taken in situational cases where an
initial client request to add a member is accepted by the server, but
fails client-side.
This situation is possible situationally, such as if network latency
causes the request to timeout after it was sent and accepted by the
cluster. In this situation, the following loop would occur and fail with
an `ErrPeerURLExist` response, and would be stuck until the backoff
timeout was met (roughly ~2min30sec currently).
Testing Done:
* Manual testing with an etcd cluster. Initial "AddMember` call was
successful, and the etcd manifest file was identical to prior version
of these files. Subsequent calls to add the same member succeeded
immediately (retaining idempotency), and the resulting manifest file
remains identical to previous version as well. The difference, this
time, is the call finished ~2min25sec faster in an identical test in
the environment tested with.
During operations such as "upgrade", kubeadm fetches the
ClusterConfiguration object from the kubeadm ConfigMap.
However, due to requiring node specifics it wraps it in an
InitConfiguration object. The function responsible for that is:
app/util/config#FetchInitConfigurationFromCluster().
A problem with this function (and sub-calls) is that it ignores
the static defaults applied from versioned types
(e.g. v1beta3/defaults.go) and only applies dynamic defaults for:
- API endpoints
- node registration
- etc...
The introduction of Init|JoinConfiguration.ImagePullPolicy now
has static defaulting of the NodeRegistration object with a default
policy of "PullIfNotPresent". Respect this defaulting by constructing
a defaulted internal InitConfiguration from
FetchInitConfigurationFromCluster() and only then apply the dynamic
defaults over it.
This fixes a bug where "kubeadm upgrade ..." fails when pulling images
due to an empty ("") ImagePullPolicy. We could assume that empty
string means default policy on runtime in:
cmd/kubeadm/app/preflight/checks.go#ImagePullCheck()
but that might actually not be the user intent during "init" and "join",
due to e.g. a typo. Similarly, we don't allow empty tokens
on runtime and error out.
- Make v1beta3 use bootstraptoken/v1 instead of local copies
- Make the internal API use bootstraptoken/v1
- Update validation, /cmd, /util and other packages
- Update v1beta2 conversion
In the Alpha stage of the feature in kubeadm to support
a rootless control plane, the allocation and assignment of
UID/GIDs to containers in the static pods will be automated.
This automation will require management of users and groups
in /etc/passwd and /etc/group.
The tools on Linux for user/group management are inconsistent
and non-standardized. It also requires us to include a number of
more dependencies in the DEB/RPMs, while complicating the UX for
non-package manager users.
The format of /etc/passwd and /etc/group is standardized.
Add code for managing (adding and deleting) a set of managed
users and groups in these files.
Ideally this should be part of dockershim/CRI and not on the
side of kubeadm.
Remove the detection during:
- During preflight
- During kubelet config defaulting
Update dependencies and the test images to use pause 3.5. We also
provide a changelog entry for the new container image version.
Signed-off-by: Sascha Grunert <sgrunert@redhat.com>
- Remove the object form v1beta3 and internal type
- Deprecate a couple of phases that were specifically designed / named to
modify the ClusterStatus object
- Adapt logic around annotation vs ClusterStatus retrieval
- Update unit tests
- Run generators
Running "go test ./cmd/kubeadm/app/..." results in these 3 files
being generated, since we have more callers to the functions
for generating unique private keys during pkiutil tests.
Add the files to ensure they are not generated locally all the time.
- Pin the ClusterConfiguration when fuzzing
the internal InitConfiguration that embeds it. Kubeadm includes
separate constructs for this embedding in the internal type
and this round trip is not viable.
- Remove the artificial calls to SetDefaults_ClusterConfiguration()
in v1beta{2|3}'s converters from public to internal InitConfiguration.
- Make sure the internal InitConfiguration.ClusterConfiguration is
defaulted in initconfiguration.go instead.
- scheme: switch to:
utilruntime.Must(scheme.SetVersionPriority(v1beta3.SchemeGroupVersion))
- change all imports in the code base from v1beta2 to v1beta3
- rename all import aliases for kubeadmapiv1beta2 to "kubeadmapiv".
this allows smaller diffs when changing the default public API.
Updates kubeadm version resolution to use kubernetes community infra
bucket to fetch appropriate k8s ci versions. The images are already
being pulled from the kubernetes community infra bucket meaning that a
mismatch can occur when the ci version is fetched from the google infra
bucket and the image is not yet present on k8s infra.
Follow-up to kubernetes/kubernetes#97087
Signed-off-by: hasheddan <georgedanielmangum@gmail.com>
Add DefaultedStaticInitConfiguration() which can be
used instead of DefaultedInitConfiguration() during unit tests.
The later can be slow since it performs dynamic defaulting.
Fixes an issue where some kubeadm phases fail if a certificate file
contains a certificate chain with one or more intermediate CA
certificates. The validation algorithm has been changed from requiring
that a certificate was signed directly by the root CA to requiring that
there is a valid certificate chain back to the root CA.
The error messages when the user feeds an invalid discovery token CA
hash are vague. Make sure to:
- Print the list of supported hash formats (currently only "sha256").
- Wrap the error from pubKeyPins.Allow() with a descriptive message.
- Mark the "node-role.kubernetes.io/master" key for labels
and taints as deprecated.
- During "kubeadm init/join" apply the label
"node-role.kubernetes.io/control-plane" to new control-plane nodes,
next to the existing "node-role.kubernetes.io/master" label.
- During "kubeadm upgrade apply", find all Nodes with the "master"
label and also apply the "control-plane" label to them
(if they don't have it).
- During upgrade health-checks collect Nodes labeled both "master"
and "control-plane".
- Rename the constants.ControlPlane{Taint|Toleraton} to
constants.OldControlPlane{Taint|Toleraton} to manage the transition.
- Mark constants.OldControlPlane{{Taint|Toleraton} as deprecated.
- Use constants.OldControlPlane{{Taint|Toleraton} instead of
constants.ControlPlane{Taint|Toleraton} everywhere.
- Introduce constants.ControlPlane{Taint|Toleraton}.
- Add constants.ControlPlaneToleraton to the kube-dns / CoreDNS
Deployments to make them anticipate the introduction
of the "node-role.kubernetes.io/control-plane:NoSchedule"
taint (constants.ControlPlaneTaint) on kubeadm control-plane Nodes.
Client side period validation of certificates should not be
fatal, as local clock skews are not so uncommon. The validation
should be left to the running servers.
- Remove this validation from TryLoadCertFromDisk().
- Add a new function ValidateCertPeriod(), that can be used for this
purpose on demand.
- In phases/certs add a new function CheckCertificatePeriodValidity()
that will print warnings if a certificate does not pass period
validation, and caches certificates that were already checked.
- Use the function in a number of places where certificates
are loaded from disk.
- Ensure the directory is created with 0700 via a new function
called CreateDataDirectory().
- Call this function in the init phases instead of the manual call
to MkdirAll.
- Call this function when joining control-plane nodes with local etcd.
If the directory creation is left to the kubelet via the
static Pod hostPath mounts, it will end up with 0755
which is not desired.
Pinning the kube-controller-manager and kube-scheduler kubeconfig files
to point to the control-plane-endpoint can be problematic during
immutable upgrades if one of these components ends up contacting an N-1
kube-apiserver:
https://kubernetes.io/docs/setup/release/version-skew-policy/#kube-controller-manager-kube-scheduler-and-cloud-controller-manager
For example, the components can send a request for a non-existing API
version.
Instead of using the CPE for these components, use the LocalAPIEndpoint.
This guarantees that the components would talk to the local
kube-apiserver, which should be the same version, unless the user
explicitly patched manifests.
Back in the v1alpha2 days the fuzzer test needed to be disabled. To ensure that
there were no config breaks and everything worked correctly extensive replacement
tests were put in place that functioned as unit tests for the kubeadm config utils
as well.
The fuzzer test has been reenabled for a long time now and there's no need for
these replacements. Hence, over time most of these were disabled, deleted and
refactored. The last remnants are part of the LoadJoinConfigurationFromFile test.
The test data for those old tests remains largely unused today, but it still receives
updates as it contains kubelet's and kube-proxy's component configs. Updates to these
configs are usually done because the maintainers of those need to add a new field.
Hence, to cleanup old code and reduce maintenance burden, the last test that depends
on this test data is finally refactored and cleaned up to represent a simple unit test
of `LoadJoinConfigurationFromFile`.
Signed-off-by: Rostislav M. Georgiev <rostislavg@vmware.com>
Add PatchStaticPod() in staticpod/utils.go
Apply patches to static Pods in:
- phases/controlplane/CreateStaticPodFiles()
- phases/etcd/CreateLocalEtcdStaticPodManifestFile() and
CreateStackedEtcdStaticPodManifestFile()
Add unit tests and update Bazel.
Component configs are used by kubeadm upgrade plan at the moment. However, they
can prevent kubeadm upgrade plan from functioning if loading of an unsupported
version of a component config is attempted. For that matter it's best to just
stop loading component configs as part of the kubeadm config load process.
Signed-off-by: Rostislav M. Georgiev <rostislavg@vmware.com>
In slower setups it can take more time for the existing cluster
to be in a healthy state, so the existing backoff of ~50 seconds
is apparently not sufficient.
The client dial can also fail for similar reasons.
Improve kubeadm's join toleration of adding new etcd members.
Wrap both the client dial and member add in a longer backoff
(up to ~200 seconds).
This particular change should be backported to the support skew.
In a future change for master, all etcd client operations should be
make consistent so that the etcd logic is in a sane state.
kubeadm uses image tags (such as `v3.4.3-0`) to specify the version of
etcd. However, the upgrade code in kubeadm uses the etcd client API to
fetch the currently deployed version. The result contains only the etcd
version without the additional information (such as image revision) that
is normally found in the tag. As a result it would refuse an upgrade
where the etcd versions match and the only difference is the image
revision number (`v3.4.3-0` to `v3.4.3-1`).
To fix the above issue, the following changes are done:
- Replace the existing etcd version querying code, that uses the etcd
client library, with code that returns the etcd image tag from the
local static pod manifest file.
- If an etcd `imageTag` is specified in the ClusterConfiguration during
upgrade, use that tag instead. This is done regardless if the tag was
specified in the configuration stored in the cluster or with a new
configuration supplied by the `--config` command line parameter.
If no custom tag is specified, kubeadm will select one depending on
the desired Kubernetes version.
- `kubeadm upgrade plan` no longer prints upgrade information about
external etcd. It's the user's responsibility to manage it in that
case.
Signed-off-by: Rostislav M. Georgiev <rostislavg@vmware.com>
Most of these could have been refactored automatically but it wouldn't
have been uglier. The unsophisticated tooling left lots of unnecessary
struct -> pointer -> struct transitions.
The selected key type is defined by kubeadm's --feature-gates option:
if it contains PublicKeysECDSA=true then ECDSA keys will be generated
and used.
By default RSA keys are used still.
Signed-off-by: Dmitry Rozhkov <dmitry.rozhkov@linux.intel.com>
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.
- 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.
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.
