DISCLAIMER: THE ALL-IN-ONE (AIO) OCP DEPLOYMENT IS AN UNSUPPORTED OCP 3.11.x CONFIGURATION INTENDED FOR TESTING OR DEMO PURPOSES.

A common request from customers is how to run the actual Red Hat OpenShift  Container Platform (OCP) bits in a single node. This request often comes from customers that need to support training environments, dedicated single-user development environments or from technical architects wanting to validate concepts without deploying a full multi-node cluster. There are many options available for developers, from Minishift to CodeReady Workspaces. These are supported options which are a great solution for application developers that want to deploy to the platform. The use cases not addressed with these solutions are more platform and infrastructure related.

For example, use cases well suited for this OCP AIO configuration arelike:

  • Training OCP operations teams.
  • Enabling and managing multiple identity providers.
  • Configuring and managing NetworkPolicy objects.
  • Configuring advanced inbound traffic options, such as the usage of ExternalIPs, etc.
  • Testing certain OpenShift-SDN features or other CNI plugins.

In short, use cases around platform operations and administrative tasks.

There are more business related use cases where an All-in-One installation of OpenShift can be useful, such as:

  • Using OCP as universal CPE (uCPE) platform.
  • Using OCP for Cloud-Native far Edge setups.
  • Using OCP for Cloud-Native network appliances.

NOTE: At the time of this writing, OCP AIO it is not a commercially supported deployment. This blog explores the intended use case for this project so architects and others can begin experimenting with the possibilities of a single node installation of OpenShift.

The Design

The following conceptual diagram illustrates the design constructs. This is a single machine running OCP and hosting an NFS server to use as the persistent storage provider for the platform and applications.

The host characteristics that I tested are:

  • VM or bare-metal
  • 6 vCPU (should work with fewer, but services like Prometheus and ElasticSearch should be configured with a lower limit of milicores)
  • RHEL 7.6 minimal install
  • 8GB RAM
  • 40GB Drive (the OS with the full OCP installation consume ~16GB but you may want to have enough for your apps and containers)

The OCP setup for the All-in-One (AIO) used the following parameters:

  • AIO node name: ocp.example.com
  • AIO node IP: 192.168.1.30
  • App wildcard subdomain: *.apps.ocp.example.com
  • NFS Server configured to export: /srv/nfs
  • Deployment can use docker or CRI-O as container runtime
  • Configure AIO node with DNS resolving the Node FQDN and wildcard subdomain.

DNSMASQ as Lab DNS

Since OCP requires FQDN with proper DNS resolution, I configured an external dnsmasq instance with the following entry:

$ cat /etc/dnsmasq.d/ocp.example.com.conf

address=/ocp.example.com/192.168.1.30

Due to the way dnsmasq works this will resolve any subdomain under the *.ocp.example.com to the host IP enabling the *.apps.ocp.example.com to also work.

The AIO Inventory file

Note: Since this configuration is intended for lab use, some of the services (i.e. Prometheus, ElasticSearch, Metrics, etc) have been configured to use ephemeral storage.

The core section of the inventory file is as follows:

 

###########################################################################

### OpenShift Hosts

###########################################################################

[OSEv3:children]

nfs

masters

etcd

nodes

[nfs]

ocp.example.com

[masters]

ocp.example.com

[etcd]

ocp.example.com

[nodes]

## All-In-One with CRI-O

#ocp.example.com openshift_node_group_name='node-config-all-in-one-crio' openshift_node_problem_detector_install=true

## All-In-One with Docker

ocp.example.com openshift_node_group_name='node-config-all-in-one' openshift_node_problem_detector_install=true

 

Note: For a complete reference inventory file, refer to this GitHub repo.

Preparing the AIO environment

  1. Ensure the AIO node can ssh passwordless to itself.
$ ssh-keygen

$ ssh-copy-id ocp.example.com

  1. Register the host
# register each host with RHSM

subscription-manager register --username=<user_name> --password=<password>

# pull subscriptions

subscription-manager refresh

# identify the available OpenShift subscriptions

subscription-manager list --available --matches '*OpenShift*'

# assign a subscription to the node

subscription-manager attach --pool=<pool_id>

# Disable all RHSM repositories

subscription-manager repos --disable="*"

# Enable only repositories required by OpenShift

subscription-manager repos \

   --enable="rhel-7-server-rpms" \

   --enable="rhel-7-server-extras-rpms" \

   --enable="rhel-7-server-ose-3.11-rpms" \

   --enable="rhel-7-server-ansible-2.6-rpms"

  1. Update system to latest patches
# Update RHEL

yum -y update

# Reboot with updated Kernel

reboot

  1. Install required installation tools
<span>$ yum -y install atomic-openshift-clients openshift-ansible</span>

Installing the AIO environment

The inventory file can either go into /etc/ansible/hosts or into any folder you have access to. For the purpose of this blog, I assume the inventory file is in the local directory ./inventory_aio.

Note: For the setup documented in this blog I’m using NFS as the OCP AIO persistent storage and the ansible playbook will assume /srv/nfs is the NFS server path.

  1. Install prerequisites
<span>ansible-playbook -i inventory_aio /usr/share/ansible/openshift-ansible/playbooks/prerequisites.yml</span>
  1. Deploy OCP AIO
<span>ansible-playbook -i inventory_aio /usr/share/ansible/openshift-ansible/playbooks/deploy_cluster.yml</span>

Summary

OpenShift (OCP) as a superset of Kubernetes brings many additional integrations and capabilities to this container native infrastructure. When considering non-traditional use cases of Kubernetes,  integrations with additional tools and functionalities not offered by the core Kubernetes will be required. Fortunately, OpenShift is extremely modular, so you can add or remove functionality as you go.

In this blog, we have a documented process to install the full OCP featuresets and capabilities in an AIO configuration. You may want to use this simply as a  practice platform, where configurations can be tested, or for training classes. You could even explore ways to use OCP on universal CPE environments. The only limits are your imagination.