Context: Load-balancing in Exoscale is performed by associating many
workers with the same EIP. This works, however, the workers cannot access
themselves via the EIP, which is needed at least for cert-managers
"self-test".
Problem: The old iptables based workaround felt fragile and disappointed
me at least once.
New solution: Add the EIP to a loopback interface on each worker.
* Remove contrib/vault
This is marked as broken since 2018 / 3dcb914607
This still reference apiserver.pem, not used since ddffdb63bf
Signed-off-by: Etienne Champetier <e.champetier@ateme.com>
* Finish nuking vault from the codebase
Signed-off-by: Etienne Champetier <e.champetier@ateme.com>
This replaces kube-master with kube_control_plane because of [1]:
The Kubernetes project is moving away from wording that is
considered offensive. A new working group WG Naming was created
to track this work, and the word "master" was declared as offensive.
A proposal was formalized for replacing the word "master" with
"control plane". This means it should be removed from source code,
documentation, and user-facing configuration from Kubernetes and
its sub-projects.
NOTE: The reason why this changes it to kube_control_plane not
kube-control-plane is for valid group names on ansible.
[1]: https://github.com/kubernetes/enhancements/blob/master/keps/sig-cluster-lifecycle/kubeadm/2067-rename-master-label-taint/README.md#motivation
* terraform support for UpCloud
* terraform support for UpCloud
* terraform support for UpCloud
* terraform support for UpCloud
* terraform support for UpCloud
* terraform support for UpCloud
* terraform support for UpCloud
* Updates to README.md and main.tf files
* formatting and updating readme
* added a .terraform_validate CI job
* fixed format issue
* added sample inventory
* added symbolic link to group_vars
* added missing tf variables and minor fixes
* added text formatting
* minor formatting fixes
* Add terraform scripts for vSphere
* Fixup: Add terraform scripts for vSphere
* Add inventory generation
* Use machines var to provide IPs
* Add README file
* Add default.tfvars file
* Fix newlines at the end of files
* Remove master.count and worker.count variables
* Fixup cloud-init formatting
* Fixes after initial review
* Add warning about disabled DHCP
* Fixes after second review
* Add sample-inventory
This replaces KUBE_MASTERS with KUBE_CONTROL_HOSTS because of [1]:
```
The Kubernetes project is moving away from wording that is
considered offensive. A new working group WG Naming was created
to track this work, and the word "master" was declared as offensive.
A proposal was formalized for replacing the word "master" with
"control plane". This means it should be removed from source code,
documentation, and user-facing configuration from Kubernetes and
its sub-projects.
```
[1]: https://github.com/kubernetes/enhancements/blob/master/keps/sig-cluster-lifecycle/kubeadm/2067-rename-master-label-taint/README.md#motivation
* contrib/terraform/exoscale: Rework SSH public keys
Exoscale has a few limitations with `exoscale_ssh_keypair` resources.
Creating several clusters with these scripts may lead to an error like:
```
Error: API error ParamError 431 (InvalidParameterValueException 4350): The key pair "lj-sc-ssh-key" already has this fingerprint
```
This patch reworks handling of SSH public keys. Specifically, we rely on
the more cloud-agnostic way of configuring SSH public keys via
`cloud-init`.
* contrib/terraform/exoscale: terraform fmt
* contrib/terraform/exoscale: Add terraform validate
* contrib/terraform/exoscale: Inline public SSH keys
The Terraform scripts need to install some SSH key, so that Kubespray
(i.e., the "Ansible part") can take over. Initially, we pointed the
Terraform scripts to `~/.ssh/id_rsa.pub`. This proved to be suboptimal:
Operators sharing responbility for a cluster risk unnecessarily replacing resources.
Therefore, it has been determined that it's best to inline the public
SSH keys. The chosen variable `ssh_public_keys` provides some uniformity
with `contrib/azurerm`.
* Fix Terraform Exoscale test
* Fix Terraform 0.14 test
This variable was added as KUBE_MASTERS_MASTERS. That's probably a typo.
Remove the redundant `_MASTERS` suffix. Also, document the variable in the
help message.
This fixes the following failures:
./contrib/offline/README.md:14:1 MD014/commands-show-output Dollar signs used before commands without showing output [Context: "$ ./manage-offline-container-i..."]
./contrib/offline/README.md:20:1 MD014/commands-show-output Dollar signs used before commands without showing output [Context: "$ ./manage-offline-container-i..."]
* [terraform/aws] Fix Terraform >=0.13 warnings
Terraform >=0.13 gives the following warning:
```
Warning: Interpolation-only expressions are deprecated
```
The fix was tested as follows:
```
rm -rf .terraform && terraform0.12.26 init && terraform0.12.26 validate
rm -rf .terraform && terraform0.13.5 init && terraform0.13.5 validate
rm -rf .terraform && terraform0.14.3 init && terraform0.14.3 validate
```
which gave no errors nor warnings.
* [terraform/openstack] Fixes for Terraform >=0.13
Terraform >=0.13 gives the following error:
```
Error: Failed to install providers
Could not find required providers, but found possible alternatives:
hashicorp/openstack -> terraform-provider-openstack/openstack
```
This patch fixes these errors.
This fix was tested as follows:
```
rm -rf .terraform && terraform0.12.26 init && terraform0.12.26 validate
rm -rf .terraform && terraform0.13.5 init && terraform0.13.5 validate
rm -rf .terraform && terraform0.14.3 init && terraform0.14.3 validate
```
which gave no errors nor warnings for Terraform 0.13.5 and Terraform
0.14.3. Unfortunately, 0.12.x gives a harmless warning, but
with 0.14.3 out the door, I guess we need to move on.
* [terraform/packet] Fixes for Terraform >=0.13
This fix was tested as follows:
```
export PACKET_AUTH_TOKEN=blah-blah
rm -rf .terraform && terraform0.12.26 init && terraform0.12.26 validate
rm -rf .terraform && terraform0.13.5 init && terraform0.13.5 validate
rm -rf .terraform && terraform0.14.3 init && terraform0.14.3 validate
```
Errors are gone, but warnings still remain. It is impossible to please
all three versions of Terraform.
* Add tests for Terraform >=0.13
Now markdownlint covers ./README.md and md files under ./docs only.
However we have a lot of md files under different directories also.
This enables markdownlint for other md files also.
* fix flake8 errors in Kubespray CI - tox-inventory-builder
* Invalidate CRI-O kubic repo's cache
Signed-off-by: Victor Morales <v.morales@samsung.com>
* add support to configure pkg install retries
and use in CI job tf-ovh_ubuntu18-calico (due to it failing often)
* Switch Calico, Cilium and MetalLB image repos to Quay.io
Co-authored-by: Victor Morales <v.morales@samsung.com>
Co-authored-by: Barry Melbourne <9964974+bmelbourne@users.noreply.github.com>
* Add note about changing private IP in admin.conf.
When I run kubespray, a load balancer is created which should be used instead of the ip of the controller node.
* Procedure to find load balancer and update admin.conf
When I run kubespray, a load balancer is used instead of the private ip of the controller.
I kept seeing `TLS handshake error from 10.250.250.158:63770: EOF` from two IP addresses that correlate to my ELB. Changing the health check from TCP to HTTPS stopped the errors from being generated.
It was documented as if it were an Ansible variable, but it is a Terraform variable.
This also means the colon syntax was incorrect. TF variables are assigned with an equals sign.
Co-authored-by: rptaylor <rptaylor@uvic.ca>
This changes MetalLB contrib to one of addons for deploying MetalLB with
Kubernetes cluster deployment. By the default, Kubespray doesn't deploy
MetalLB addon.
inventory_builder creates hosts.yaml file with hostnames like "node1",
"node2", etc. Even if specifying override_system_hostname=false, the
output of "kubectl get nodes" shows those hostnames ("node1", etc.)
without using actual hostnames.
To solve this issue, this adds an option USE_REAL_HOSTNAME to get
actual hostnames when creating hosts.yaml file instead of "node1", etc.
Since MetalLB v0.8[1], metallb:speaker has started publishing an event
nodeAssigned on k8s resource.
To support MetalLB v0.8+, this allows metallb:speaker to create events.
[1]: 5cc6e23776 (diff-60053ad6fecb5a3cfabb6f3d9e720899R246)
If running MetalLB v0.7.3 on k8s v1.18.2, metallb pods output the
following parsing error of v1.ServiceList:
$ kubectl logs controller-dbb46cf84-fw8h8 -n metallb-system
{
"caller":"reflector.go:205",
"level":"error",
"msg":"go.universe.tf/metallb/internal/k8s/k8s.go:231:
Failed to list *v1.Service: v1.ServiceList:
Items: []v1.Service: v1.Service: ObjectMeta:
v1.ObjectMeta: readObjectFieldAsBytes:
expect : after object field, parsing 1605
Then an external IP address is never allocated to the Service of
LoadBalancer type.
By updating MetalLB version to the latest v0.9[1] today, this issue
can be solved.
[1]: https://hub.docker.com/r/metallb/controller/tags
