Install
Choose how to install the operators; the platform resource is the same in every case.
Method |
For |
Page |
|---|---|---|
OLM |
OpenShift (OperatorHub and the console form) |
|
OLM |
Kubernetes with OLM installed |
|
Kustomize |
Kubernetes without OLM |
|
Ansible |
Any of the above, from a playbook |
This documentation is public. During this phase, the images in ghcr.io/koiosplatform and the
koios-installer repository are private and require access granted by the Koios project. Registry
read access and repository read access are separate permissions: a registry token alone does not
let you clone the installer repository.
To request access, write to the Koios project at asield03@gmail.com.
All methods use the installer repository (koios-installer): one tag per release with the
OLM catalog, the operators rendered for Kustomize, platform examples and the Ansible collection, all
pinned to the images of that release.
After repository access has been granted, clone the installer for your release:
git clone --branch <version> https://github.com/KoiosPlatform/koios-installer.git
cd koios-installer
Requirements
OpenShift 4 (tested on 4.21) or Kubernetes (tested on k3s 1.37 with kustomize, and with OLM v0.46.0).
Access to the private registry
ghcr.io/koiosplatform(Registry access).A default StorageClass. Automation hub stores content on a ReadWriteMany volume by default; without one, use a single-node volume (
hub.file_storage_access_mode: ReadWriteOnce).
Lab observation
In the test clusters, a complete platform (gateway, controller, EDA and hub with managed PostgreSQL and Redis) ran on one worker with 16 GiB of memory. This is a lab observation, not a production minimum or sizing recommendation. Size the platform for its workload, storage, availability needs, and the resources consumed by job and activation pods.
Install modes with OLM
Scope |
Operators run in |
They manage |
|---|---|---|
Cluster-scoped (recommended) |
|
Every namespace |
Namespaced |
The platform namespace ( |
That namespace only |
Only one of the two can exist in a cluster: both own the same APIs.