OKD with ansible

This repository holds ansible files for installing OKD using Ansible on hardware.

Installing a user-provisioned cluster on bare metal

It assumes you have the following hardware:

1. A debian server (e.g. Raspberry PI) for infrastructure (load balancing and serving the ignition files)
2. An intel machine that will first serve as bootstrap (can be the same as the machine that will later serve as worker1).
3. 3 master intel machines.

If you have more hardware, adjust the inventory file accordingly.

NOTE: If you want to run a different version of OKD, extract the installer under ./openshift-install which will stop retrieving the installer. The exact command to extract the installer is available per beta version at OKD Nightly Releases.

There are optional components to install which are controlled by ansible variables. These are defined in the inventory group vars.

variable	description
argocd	whether to install the argocd operator
sealed_secrets	Whether to pre-install a secret for sealed_secret to make sure git sealed secrets can be decrypted
set_etc_hostname_in_ignition_file	Whether to set the hostname in /etc/hostname
use_control_plane_nodes_for_compute	Whether to allow masters to be used for regular pods

Load Balancer

The load balancer can be installed on a debian server (e.g. Raspberry PI) and is used for infrastructure (load balancing and serving the ignition files).

Nginx is used for static hosting (ignition files) and HAProxy is used for load balancing. The HAProxy status interface is found at infra1.okd4.example.com:1936/stats.

The load balancer installation can be controlled with the following parameters:

Variable	Description	Default
lbs_loadbalancer_haproxy_stats_auth	Enable HAProxy status page authentication.	false
lbs_loadbalancer_haproxy_stats_username	HAProxy status page username.	admin
lbs_loadbalancer_haproxy_stats_password	HAProxy status page password.	password
configure_ufw	Configure UFW port openings.	false

Disconnected registry

In disconnected environments you normally have a docker registry which will supply all the OKD images.

To use a disconnected registry, set the following parameters:

variable	description
use_disconnected_registry	Boolean. Indicating a registry should be used. Example: true
disconnected_registry_trust_bundle_file	String. Filename of the root CA for the registry. Example: ./openshift-ca/example.crt
disconnected_registries	Array of objects. Mirror registries.
disconnected_registries.source	String. URL that will be replaced with mirror. Example: quay.io/openshift/okd
disconnected_registries.mirrors	Array of strings. URL to mirror registry. Example: registry.okd4.example.com:5011/openshift/okd

Pull-through-cache

In disconnected environments you normally have a docker registry which will supply all the OKD images. The pull-through-cache will simulate that and should also, if you have a faster cache than your Internet connection, improve your installation time.

There is a Makefile under hack/docker-proxy that can create the certificates and docker containers needed for that. The certificates will be placed under openshift-ca and will need to be copied to the docker machine. Note that the hostname registry.okd4.example.com needs to be setup in DNS.

Usage	description
make ca	creates both CA and registry certificates
make quay.io	creates the docker proxy registry for quay.io
make docker.io	creates the docker proxy registry for docker.io

Preparation

1. Pull your RedHat pull secret and place in file pull-secret.

2. If you want github integration, create a file called "github-config.json" with similar content as this:

   {
     "clientID": "< github client ID >",
     "clientSecret": "< github client secret >",
     "organizations": [
       "< your github organization >"
     ]
   }

3. Set up DNS with the following entries (of course, adjust addresses to your infrastructure):

   hostname	Address
   infra1.okd4.example.com	192.168.60.180
   api-int.okd4.example.com	CNAME infra1.okd4.example.com
   api.okd4.example.com	CNAME infra1.okd4.example.com
   apps.okd4.example.com	CNAME infra1.okd4.example.com
   *.apps.okd4.example.com	CNAME infra1.okd4.example.com
   master1.okd4.example.com	192.168.60.181
   master2.okd4.example.com	192.168.60.182
   master3.okd4.example.com	192.168.60.183
   worker1.okd4.example.com	192.168.60.184

   If you're using pihole (as I do), increase rate limiting, create /etc/dnsmasq.d/99-openshift.conf with the following content and restart dns (pihole restartdns)

   address=/.apps.okd4.example.com/192.168.60.180
   

   If your DNS cannot handle wildcards, add these entries as CNAME, pointing to app.okd4.example.com:

   • alertmanager-main-openshift-monitoring.apps.okd4.example.com
   • canary-openshift-ingress-canary.apps.okd4.example.com
   • console-openshift-console.apps.okd4.example.com
   • downloads-openshift-console.apps.okd4.example.com
   • grafana-openshift-monitoring.apps.okd4.example.com
   • oauth-openshift.apps.okd4.example.com
   • prometheus-k8s-openshift-monitoring.apps.okd4.example.com
   • thanos-querier-openshift-monitoring.apps.okd4.example.com

4. Create a new image for the rasperry pi with enabled ssh and boot it up.

5. Create a bootable USB from the correct version of Fedora CoreOS.
   (At the time of writing, the current working release is 36.20220716.3.1)

   ./openshift-install/openshift-install coreos print-stream-json | jq -r '.architectures.x86_64.artifacts.metal.formats.iso.disk.location'

6. Set up static DHCP entries for the machines, matching IP addresses above.

Python PIP proxy

If you are using a proxy for PIP, you can configure this by defining the PIP_INDEXURL and PIP_CERT environment variables in the .env or ${CLUSTER_NAME}/.env file.

Ansible CA

If you are downloading tools from a self-hosted proxy that is using a custom CA, you need to configure Ansible to use this CA.

You can do this by defining the SSL_CERT_FILE environment variable in the .env or ${CLUSTER_NAME}/.env file.

Run playbook

It's time to run the playbook. There are a number of steps that will be completed:

1. Cluster configuration will be created and added into ignition files.
2. The infrastructure node (RPI) will be setup.

This is how to run the playbook:

make dependencies
CLUSTER_NAME=example make cluster

Or if you wish to run the playbook directly:

ansible-playbook -i inventories/example -v deploy-okd.yml --extra-vars "use_control_plane_nodes_for_compute=true argocd=true"

Once the playbook tell you to, boot the masters on Fedora CoreOS USB.

To switch keyboard mapping on CoreOS, do the following:

$ sudo localectl set-keymap se

Then start the installation process.

# Use hostname from DHCP or DNS
$ bash -c "$(curl -fsSL http://infra1.okd4.example.com:8080/install.sh)"

# Manually set hostname
$ bash -c "$(curl -fsSL http://infra1.okd4.example.com:8080/install.sh)" -s master[1-3]

Once the installation has finished, remove the bootable media and reboot.

Verify that the master is trying to pull the secondary ignition from https://api-int.okd4.example.com:22623.

Once all masters are waiting for the secondary ignition, continue the playbook which tell you to boot the first worker machine on Fedora CoreOS USB and start the installation for the bootstrap process:

# Use hostname from DHCP or DNS
$ bash -c "$(curl -fsSL http://infra1.okd4.example.com:8080/install.sh)"

# Manually set hostname
$ bash -c "$(curl -fsSL http://infra1.okd4.example.com:8080/install.sh)" -s bootstrap

Once the installation has finished, remove the bootable media and reboot.

After some time, you will be able to login via ssh to the bootstrap machine and follow the installation:

$ ssh [email protected]
The authenticity of host 'worker1.okd4.example.com (192.168.60.184)' can't be established.
ECDSA key fingerprint is SHA256:Z3edOf5ImnxO/x9tchkto5LoEQIaFm8DT/7zyGj5r6g.
Are you sure you want to continue connecting (yes/no/[fingerprint])? yes
Warning: Permanently added 'worker1.okd4.example.com,192.168.60.183' (ECDSA) to the list of known hosts.
This is the bootstrap node; it will be destroyed when the master is fully up.

The primary services are release-image.service followed by bootkube.service. To watch their status, run e.g.

  journalctl -b -f -u release-image.service -u bootkube.service
Fedora CoreOS 34
Tracker: https://github.com/coreos/fedora-coreos-tracker
Discuss: https://discussion.fedoraproject.org/c/server/coreos/

Last login: Sat Sep  4 05:55:40 2021 from 192.168.40.50
[core@nucbootstrap ~]$ journalctl -b -f -u release-image.service -u bootkube.service

When the installation has started, continue the playbook. When the playbook detects that the installation is finished, the playbook will continue with post-installation configuration.

You can now open the cluster console by opening https://console-openshift-console.apps.okd4.example.com.

Have fun with your cluster!

Adding the bootstrap or any other node to the cluster

Once the cluster has been correctly installed, shutdown the bootstrap node, remove the partitions and reinstall using this command:

# Use the generic ignition file for the role, "master" or "worker"
$ bash -c "$(curl -fsSL http://infra1.okd4.example.com:8080/install.sh)" -s worker

Since this has not been prepared in the cluster earlier, you need to approve the certificate requests.

This is easiest done with the issueCertificates.sh script. Note that there is normally three certificate per node.

$ ./issueCertificates.sh
./issueCertificates.sh                                                                                           
No pending certificate requests
...
No pending certificate requests
No pending certificate requests
Approving pending certificate requests...
certificatesigningrequest.certificates.k8s.io/csr-qmkv6 approved
No pending certificate requests
Approving pending certificate requests...
certificatesigningrequest.certificates.k8s.io/csr-wfnjq approved
certificatesigningrequest.certificates.k8s.io/csr-qftjq approved
No pending certificate requests

Network booting nodes

The OKD nodes can be booted over the network with iPXE.

Variable	Description	Default
network_boot_enabled	Enable configuration for network booting.	false
network_boot_coreos_proxy_scheme	Override coreos proxy scheme (http/https).
network_boot_coreos_proxy_hostname	Override coreos proxy hostname.
network_boot_coreos_proxy_port	Override coreos proxy port.
network_boot_coreos_proxy_path	Override coreos proxy path.

Boot order

Configure the following boot order for the machines:

1. Hard Disk (Boot from disk after install)
2. Optical (e.g. boot from iPXE iso)
3. USB (e.g. boot from iPXE USB stick)

Required DHCP configuration

Required DHCP information:

DHCP field	Description	Example
Hostname	Provide hostname using DHCP.	bootstrap
Domain	Provide domain using DHCP.	okd4.example.com
Filename	Default BIOS file name.	lbs.okd4.example.com:8081/ipxe/autoexec.ipxe

How it works

PXE

PXE network booting is not supported at the moment.

iPXE

iPXE can be booted from an iso, USB drive, etc.

More information: https://ipxe.org/

1. iPXE will start and get a DHCP assignment. If your DHCP provider does not set the hostname and domain, you can manually set them with

   $ set hostname master1
   $ set domain okd4.example.com
   

2. iPXE will first load http://infra1.okd4.example.com:8081/ipxe/autoexec.ipxe This can be done manually with

   $ chain http://infra1.okd4.example.com:8081/ipxe/autoexec.ipxe
   

3. Which will load http://infra1.okd4.example.com:8081/ipxe/$hostname.$domain.ipxe
   • Hostname and domain values are taken from DHCP.
4. Which will load CoreOS with the nodes ignition file. e.g. http://lbs.okd4.example.com:8080/ignition/bootstrap.okd4.example.com.ign

Local testing using molecule

Test the ansible scripts locally using molecule.
Molecule uses vagrant and virtualbox to create virtual machines to perform the installation on.

Dependencies

• Virtualbox
• Vagrant
• Molecule

VirtualBox

https://www.virtualbox.org/wiki/Downloads

macOS

brew install --cask virtualbox

Vagrant

https://developer.hashicorp.com/vagrant/downloads

macOS

brew install hashicorp/tap/hashicorp-vagrant

Molecule (and python dependencies)

https://molecule.readthedocs.io/en/latest/

make local-dependencies

Usage

Test

To perform a full test, run:

make molecule-test

Manual testing

These are the commands used for testing manually:

molecule create
molecule converge
molecule verify
molecule destroy

To run a specific scenario:

molecule create -s <scenario name>
molecule converge -s <scenario name>
molecule verify -s <scenario name>
molecule destroy -s <scenario name>