virtualization-set-up

Description

A terraform repository for provisioning a virtual kubernetes cluster on VirtualBox.

Components

packer virtual machine images

Packer is used to create a Ubuntu server virtual machine image for vagrant (.box file) with a basic stack of packages.

Kubernetes cluster

Provisions a Kubernetes cluster in VirtualBox with the desired configuration. The virtual machines are provisioned with vagrant. The Kubernetes cluster is configured with Ansible. Terraform manages these resources.

Logging in via SSH can be done by ssh vagrant@localhost -p 2200 -i ~/.ssh/private-key.

At this point, vagrant cannot be used to manage the virtual machines unless the environment variables for the VMI name and the SSH private key are explicitly set.

• i.e., if you want to ssh in a machine, you need to export virtual_machine="" and export private_key_path="" first and then vagrant ssh vmi-name.
• check this issue for more information.

MetalLB

MetalLB is needed so that an external IP of the kubernetes cluster can be used; otherwise NodePorts should be used, which is inconvenient. Installed via Helm.

Linkerd

Linkerd is used as a service mesh provider. Injections can be done on a namespace level by annotating with linkerd.io/inject: enabled.

Kubernetes dashboard

The cluster comes with the Kubernetes dashboard installed. It can be accessed via a LoadBalancer service through IP or through a NodePort on port 30002 and with a token that is fetched using kubectl -n kubernetes-dashboard create token admin-user. Alternatively, a host like (kubernetes-dashboard.my-cluster.local) can be specified in /etc/hosts and used in browser.

Prometheus and Grafana

The cluster includes Prometheus and Grafana, as per the Prometheus community. Custom configurations are applied so that Prometheus works with Linkerd.

ArgoCD

ArgoCD is available and accessible via a LoadBalancer service.

Usage

You need to have Packer, Vagrant, Terraform, Ansible, Helm and VirtualBox installed. Run terraform init && terraform plan to get an idea of what will be executed. It is recommended to run terraform apply -target=module.[...] in the order the components are listed above. terraform destroy will destroy all resources, cleaning up VirtualBox machines as well.

Ideas

• move packer variables to somewhere else; they are hardcoded and essentially duplicated;
• explore Terragrunt;
• check ways for moving the loop outside of the Vagrantfile. Currently it is needed there because machines (apparently) need to be created sequentially, and Terraform doesn't have a good way to sequentialize resources with count or for_each. Check this SO answer for a suggested (and not especially neat) approach.
• check ways for alternative version specification; the current approach relies too much on the format of CHANGES.md.
• add removing of virtual boxes on destroying the k8s cluster; otherwise vagrant uses the old virtualboxes despite packer creating new ones.
• limit usage of depends_on, if possible.