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.
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>
- Don't always print to stdout that the kubelet is starting.
instead delegate this to the callers of TryStartKubelet.
- Add a new root kubeadm init phase called "kubelet-finalize"
- Add a sub-phase to "kubelet-finalize"
called "experimental-cert-rotation"
- "cert-rotation" performs the following actions:
- tries to guess if kubelet client cert rotation is enabled
- update the kubelet.conf to use the rotatable cert/key
The PR introducing 5bb8069 got merged accidentally (the CI robot not
respecting a hold). Hence, the feedback to that PR is merged separately.
Signed-off-by: Rostislav M. Georgiev <rostislavg@vmware.com>
As the hyperkube image is itself deprecated and moved out of tree, its use with
kubeadm gets deprecated too. Hence, deprecation messages will be printed when
it is used.
Signed-off-by: Rostislav M. Georgiev <rostislavg@vmware.com>
ToClientSet() in kubeconfig.go creates a clientset from
the passed Config object (kubeconfig). For IP addresses
that are not reachable e.g. Get() calls for ConfigMaps
can block for a few minutes with the default timeout.
Modify the timeout to a shorter value by passing an override.
There are two writes yet only one read on a non-buffered channel that is
created locally and not passed anywhere else.
Therefore, it could leak one of its two spawned Goroutines if either:
* The provided `f` takes longer than an erroneous result from
`waiter.WaitForHealthyKubelet`, or;
* The provided `f` completes before an erroneous result from
`waiter.WaitForHealthyKubelet`.
The fix is to add a one-element buffer so that the channel write happens
for the second Goroutine in these cases, allowing it to finish and freeing
references to the now-buffered channel, letting it to be GC'd.
3993c42431 introduced the propagation of *_PROXY
host env. variables to the kube-proxy container.
To allow The NODE_NAME variable to be properly updated by the downward
API make, sure we preserve the existing variables when adding *_PROXY.
This change removes dependencies on the internal types of the kubelet and
kube-proxy component configs. Along with that defaulting and validation is
removed as well. kubeadm will display a warning, that it did not verify the
component config upon load.
Signed-off-by: Rostislav M. Georgiev <rostislavg@vmware.com>
Checking if the path exists before creating the volume is
problematic because the path will be created regardless
after the initial call to "kubeadm init" and once the CM Pod
is running.
Then on subsequent calls to "kubeadm init" or the "control-plane"
phase the manifest for the CM will be different.
Always mount this path, but also consider the user provided
flag override from ClusterConfiguration.
Used strings instead of bytes in the TestTokenOutput test cases as
expected output is a plain text.
This should also simplify the data representation and the test code
a bit.
The flag has been problematic and abused by users.
While perhaps its original purpose was to be able to feed
a new version of the control-plane it also made it possible
to apply modifications to the ClusterConfiguration object
in the cluster. The lack of a feature in kubeadm for reconfiguration
of running clusters resulted in users using this flag for
the same purpose.
While it works for certain scenarios like updating
a static Pod for this control-plane only, it can result in
unexpected behavior if the user has for example fed a node name
different than the host name, when originally they created this node.
kubeadm 1.16 introduced the "kustomize" feature that
is a potential replacement for this user demand.
Add warning that this flag should not be used.
This field is not used in the kubeadm code. It was brought from
cli-runtime where it's used to support complex relationship between
command line parameters, which is not present in kubeadm.
This integration test allows us to detect if a given feature gate will
panic kubeadm. This builds on the assumption that a golang panic makes
the process exit with the code 2.
These tests are not trying to check if the init process succeeds or
not, their only purpose is to ensure that the exit code of the
`kubeadm init` invocation is not 2, thus, reflecting a golang panic.
Some refactors had to be made to the test code, so we return the exit
code along with stdout and stderr.
of service subnets.
Update DNS, Cert, dry-run logic to support list of Service CIDRs.
Added unit tests for GetKubernetesServiceCIDR and updated
GetDNSIP() unit test to inclue dual-sack cases.
The firewalld monitoring code was not well tested (and not easily
testable), would never be triggered on most platforms, and was only
being taken advantage of from one place (kube-proxy), which didn't
need it anyway since it already has its own resync loop.
Since the firewalld monitoring was the only consumer of pkg/util/dbus,
we can also now delete that.
Whenever kubeadm needs to fetch its configuration from the cluster, it gets
the component configuration of all supported components (currently only kubelet
and kube-proxy). However, kube-proxy is deemed an optional component and its
installation may be skipped (by skipping the addon/kube-proxy phase on init).
When kube-proxy's installation is skipped, its config map is not created and
all kubeadm operations, that fetch the config from the cluster, are bound to
fail with "not found" or "forbidden" (because of missing RBAC rules) errors.
To fix this issue, we have to ignore the 403 and 404 errors, returned on an
attempt to fetch kube-proxy's component config from the cluster.
The `GetFromKubeProxyConfigMap` function now supports returning nil for both
error and object to indicate just such a case.
Signed-off-by: Rostislav M. Georgiev <rostislavg@vmware.com>
The new etcd balancer (>3.3.14, 3.4.0) uses an asynchronous resolver for
endpoints. Without "WithBlock", the client may return before the
connection is up.
Signed-off-by: Gyuho Lee <leegyuho@amazon.com>
A recent commit added warnings for KubeletConfiguration and
KubeProxyConfiguration fields that kubeadm cares about and
does not recommend the user modifying them. Kubelet's
"rotateCertificates" cannot be handled using this function
as there is not way to figure out if the user has set it explicitly to
"false". Hardcode the value to "true" and add a comment about that.
Also apply the following changes to warnDefaultComponentConfigValue()
calls:
- use a local "kind" variable that defines the Kind we are warning about.
- fix wrong paths to fields.
- replace all stray calls of os.Exit() to util.CheckError() instead
- CheckError() now checks if the klog verbosity level is >=5
and shows a stack trace of the error
- don't call klog.Fatal in version.go
It seems undesirable that Kubernetes as a system should be
blocking a node if it's Linux kernel is way too new.
If such a problem even occurs we should exclude versions from
the list of supported versions instead of blocking users
from trying e.g. the latest 7.0.0-beta kernel because our
validators are not aware of this new version.
Usage of github.com/blang/semver is not needed and
k8s.io/apimachinery/pkg/util/version should be used instead
for semantic version parsing and version comparison.
Etcd v3.3.0 added the --listen-metrics-urls flag which allows specifying
addition URLs to the already present /health and /metrics endpoints.
While /health and /metrics are enabled for URLS defined with
--listen-client-urls (v3+ ?) they do require HTTPS.
Replace the present etcdctl based liveness probe with a standard HTTP
GET v1.Probe that connects to http://127.0.0.1:2381/health.
These endpoints are not reachable from the outside and only available
for localhost connections.
For file discovery, in case the user feeds a file for the CA
from the kubeconfig, make sure it's preloaded and embedded using
the new function EnsureCertificateAuthorityIsEmbedded().
This commit also applies cleanup:
- unroll validateKubeConfig() into ValidateConfigInfo() as this way
the default cluster can be re-used.
- in ValidateConfigInfo() reuse the variable config instead of creating
a new variable kubeconfig.
- make the Ensure* functions return descriptive errors instead of
wrapping the errors on the side of the callers.
Secure serving was already enabled for kube-controller-manager.
Do the same for kube-scheduler, by passing the flags
"authentication-kubeconfig" and "authorization-kubeconfig"
to the binary in the static Pod.
This change allows the scheduler to perform reviews on incoming
requests, such as:
- authentication.k8s.io/v1beta1 TokenReview
- authorization.k8s.io/v1 SubjectAccessReview
The authentication and authorization checks for "system:kube-scheduler"
users were previously enabled by PR 72491.
- common_test.go: use constants.CurrentKubernetesVersion
- diff_test.go: write temporary files instead of using testdata.
this allows us to not have to bump kubernetesVersions in the
testdata files (now removed)
- policy_test.go: apply fixes to tests that were previously passing,
but a bump in constants.go breaks them. these tests now work
for any version.
progagates the kubeadm FG to the individual k8scomponents
on the control-plane node.
* Note: Users who want to join worker nodes to the cluster
will have to specify the dual-stack FG to kubelet using the
nodeRegistration.kubeletExtraArgs option as part of their
join config. Alternatively, they can use KUBELET_EXTRA_ARGS.
kubeadm FG: kubernetes/kubeadm#1612
If empty "--kubernetes-version" is given (as it's not configurable now)
k8s.io/kubernetes/cmd/kubeadm/app/util/version.go.KubernetesReleaseVersion
will fetch the version from the internet.
But, this can fail:
% kubeadm init phase certs ca --cert-dir ...
unable to fetch file. URL: "https://dl.k8s.io/release/stable-1.txt", status: 502 Bad Gateway
failed to run commands: exit status 1
Can happen to other commands:
% kubeadm init phase kubeconfig controller-manager ...
% kubeadm init phase kubeconfig scheduler ...
This make "--kubernetes-version" configurable, so users can enable offline mode.
Signed-off-by: Gyuho Lee <leegyuho@amazon.com>
When adding a new etcd member the etcd cluster can enter a state
of vote, where any new members added at the exact same time will
fail with an error right away.
Implement exponential backoff retry around the MemberAdd call.
This solves a kubeadm problem when concurrently joining
control-plane nodes with stacked etcd members.
From experiment, a few retries with milliseconds apart are
sufficient to achieve the concurrent join of a 3xCP cluster.
Apply the same backoff to MemberRemove in case the concurrent
removal of members fails for similar reasons.
Nit: remove capitalization of preferred
Remove line from kubelet and add to separate PR for easier merge
nit: dependency added to separate PR
Add check to ensure strict policy cannot be set without feature gate enabled
Topology Manager runs "none" policy by default.
Added constants for policies and updated documentation.
libcni 0.7.0 caches ADD operation results and allows the runtime to
retrieve these from the cache. In case the user wants a different
cache directory than the defaul, plumb that through like we do
for --cni-bin-dir and --cni-conf-dir.
If the cluster has a PSP that blocks Pods from running as root
the DS that handles upgrade prepull will fail to create its Pods.
Workaround that by adding a PodSecurityContext with RunAsUser=999.
Instead of creating a Docker client and fetching an Info object
from the docker enpoint, call the "docker info" command
and populate a local dockerInfo struct from JSON output.
Also
- add unit tests.
- update import boss and bazel.
This change affects "test/e2e_node/e2e_node_suite_test.go"
as it consumes this Docker validator by calling
"system.ValidateSpec()".
Since we never use the cobras "SilenceErrors" or "SilenceUsage",
a command executed with "cmd.Execute()" will never return an error
without printing it.
The current behavior results in all error messages being printed twice:
Example:
$ kubectl abc
Error: unknown command "abc" for "kubectl"
Run 'kubectl --help' for usage.
unknown command "abc" for "kubectl"
This applies to all cli commands using Cobra. To verify, follow the code
path of the Execute function:
https://github.com/spf13/cobra/blob/c439c4fa0937/command.go#L793
Signed-off-by: Odin Ugedal <odin@ugedal.com>
- comment out Liz and Chuck until further notice.
Feel free to come back to kubeadm!!
- Add SataQiu as reviewer. Welcome.
- Add ereslibre as approver. Congrats!
