Okd: Documentation missing for All-in-one OKD 4

[mangelajo]

Describe the bug
The documentation for OKD 4 doesn't explain how to deploy an all in one deployment. We were planning to use something like that in the CI pipeline of the submariner project http://github.com/submariner-io to reduce the cost of deploying a fully fledged OKD 4

Version
OKD 4 documentation

How reproducible
N/A

Log bundle
N/A

[vrutkovs]

It should be as simple as setting

compute:
- name: worker
  replicas: 0
controlPlane:
  name: master
  replicas: 1

in install-config.yaml, but the docs should also mention that the host cannot be modified via MCO - and the cluster cannot be updated

[kai-uwe-rommel]

Again, documentation about this is important. Currently, the docs state that you need at least three master notes ...
I am also waiting for non-HA setups for some cases, so I'm keen on that.
I am also wondering if setting the # of masters to 1 and no workers if then the deployment of monitoring will work?
Because that currently installs with multiple instances.
I have already tried installing cluster-logging with only 1 instance (only one instance of Elasticsearch) and at least that failed.

[kai-uwe-rommel]

and the cluster cannot be updated

This will probably remain a fundamental limit? As there is no redundancy to perform a rolling updae?

[vrutkovs]

I am also wondering if setting the # of masters to 1 and no workers if then the deployment of monitoring will work?

Yes. Master would become schedulable and run all pods

This will probably remain a fundamental limit? As there is no redundancy to perform a rolling updae?

Correct. MCO requires maxUnavalable to be set for a pool (minimum 1) during upgrade, which is impossible to satisfy in single master deployment

[kai-uwe-rommel]

I'll give it a try one of these days.

[vrutkovs]

Fixed by https://github.com/openshift/okd/pull/253