When joining a control plane node and "upgrading" the cluster setup (for
example, to update etcd addresses after adding a new etcd) in the same
playbook run, the node can take a bit of time to become ready after
joining.
This triggers a kubeadm preflight check (ControlPlaneNodesReady) in
kubeadm upgrade, which is run directly after the join tasks.
Add a configurable wait for the control plane node to become Ready to
fix this race condition.
Co-authored-by: Max Gautier <mg@max.gautier.name>
Fixes a bug where `kube-apiserver` fails to start if the PodSecurity
configuration file doesn't have the `apiVersion` and `kind` keys.
Signed-off-by: Alejandro Macedo <alex.macedopereira@gmail.com>
Co-authored-by: Alejandro Macedo <alex.macedopereira@gmail.com>
* Refactor control plane upgrades with reconfiguration support
Adds revised support for:
- The previously removed `--config` argument for `kubeadm upgrade apply`
- Changes to `ClusterConfiguration` as part of the `upgrade-cluster.yml` playbook lifecycle
- kubeadm-config `v1beta4` `UpgradeConfiguration` for the `kubeadm upgrade apply` command: [UpgradeConfiguration v1beta4](https://kubernetes.io/docs/reference/config-api/kubeadm-config.v1beta4/#kubeadm-k8s-io-v1beta4-UpgradeConfiguration).
* Add kubeadm upgrade node support
Per discussion:
- Use `kubeadm upgrade node` on secondary control plane upgrades
- Add support for UpgradeConfiguration.node in kubeadm-config.v1beta4
- Remove redundant `allowRCUpgrades` config
- Revert from `block` for first and secondary control plane back to unblocked tasks since they no longer share much code and it's more readable this way
* Add kubelet and kube-proxy reconfiguration to upgrades
* Fix task to use `kubeadm init phase etcd local`
* Rebase with changes from "Adapt checksums and versions to new hashes updater" PR
* Add `imagePullPolicy` and `imagePullSerial` to kubeadm-config v1beta4 `InitConfiguration.nodeRegistration`
(cherry picked from commit b551fe083d)
* kubeadm: do not ignore preflight errors blindly
The "ignoring all errors" seems to date back to the inception of the
kubeadm support (it was --skip-preflight-check before).
This can mask real errors and prevent users from seeing them.
Do not ignore any errors by default and make the set of ignored errors
configurable.
* download/kubeadm: remove redundant task
The mode is already set by the previous `copy` task.
* Validate kubeadm configs
This should help to fail early when we have invalid kubeadm configs (from
a kubespray bug or a misconfiguration).
* kubeadm-upgrade: remove unnecessary bool cast
* Convert kubeadm join discovery timeout to v1beta4 config
* CI: Ignore kubeadm:Mem errors on some setup.
Currently there is not much difference between the files, if there are more changes in the future,
please use different files to distinguish them (you can use the kubeadm_config_api_version variable)
Signed-off-by: ChengHao Yang <17496418+tico88612@users.noreply.github.com>
I added the kubeadm_config_api_version variable in the previous commit,
and remove kubeadm api version condition.
Signed-off-by: ChengHao Yang <17496418+tico88612@users.noreply.github.com>
v1beta4 has changed a lot in this file (e.g. ExtraArgs etc.), so it was implemented in separate files.
Signed-off-by: ChengHao Yang <17496418+tico88612@users.noreply.github.com>
The fallback_ips tasks are essentially serializing the gathering of one
fact on all the hosts, which can have dramatic performance implications
on large clusters (several minutes).
This is essentially a reversal of 35f248dff0
Being able to run without refreshing the cache facts is not worth it.
We keep fallback_ip for now, simply changing the access to a normal
hostvars variable instead of a custom dictionnary.
Nodes to api-server relies by default certificates, and bootstrap
tokens, and there should be no need to generate tokens for every nodes,
even when enabling static token auth.
Testing for group membership with group names makes Kubespray more
tolerant towards the structure of the inventory.
Where 'inventory_hostname in groups["some_group"] would fail if
"some_group" is not defined, '"some_group" in group_names' would not.
Specifying one directory for kubeadm patches is not ideal:
1. It does not allow working with multiples inventories easily
2. No ansible templating of the patch
3. Ansible path searching can sometimes be confusing
Instead, provide the patch directly in a variable, and add some quality
of life to handle components targeting and patch ordering more
explicitly (`target` and `type` which are translated to the kubeadm
scheme which is based on the file name)
kubernetes/control-plane and kubernetes/kubeadm roles both push kubeadm
patches in the same way.
Extract that code and make it a dependency of both.
This is safe because it's only configuration for kubeadm, which only
takes effect when kubeadm is run.
* feat: add user facing variable with default
* feat: remove rolebinding to anonymous users after init and upgrade
* feat: use file discovery for secondary control plane nodes
* feat: use file discovery for nodes
* fix: do not fail if rolebinding does not exist
* docs: add warning about kube_api_anonymous_auth
* style: improve readability of delegate_to parameter
* refactor: rename discovery kubeconfig file
* test: enable new variable in hardening and upgrade test cases
* docs: add option to config parameters
* test: multiple instances and upgrade
