Description of problem:

In an OCP 4.16.x (reproduced with 4.14.37, 4.16.24, and 4.16.45) when setting the routingViaHost parameter to true the cluster breaks. We tried this on day2 (installing with routingViaHost: false and then moving it to true) and on day0 (adding the setting via install manifests). In both cases, the cluster breaks.

We were able to reproduce this 100% of the time in a single stack IPV6 cluster connected through proxy.

When installing with false the cluster deploys and works fine, after changing to true the cluster operators will degraded.

When installing with true, the cluster won't finish the deployment.

This same configuration works fine in 4.18.x in day2 and day0.

Version-Release number of selected component (if applicable): 4.14.x / 4.16.x

How reproducible: Always

Steps to Reproduce:

Day 2:

1. Install single-stack IPv6 cluster connected through proxy

2. Modify network.operator config to .spec.defaultNetwork.ovnKubernetesConfig.gatewayConfig.routingViaHost: true

3. Wait for the change to rollout and the cluster operators will start degrading

Day 0:

1. Add the following as install extra manifest:

apiVersion: operator.openshift.io/v1
kind: Network
metadata:
  name: cluster
spec:
  defaultNetwork:
    ovnKubernetesConfig:
      gatewayConfig:
        routingViaHost: true 

2. The install won't finish.

Actual results: 

routingViaHost cannot be set to true

Expected results:

routingViaHost can be set to true

Additional info:

The must gather cannot be collected since pods cannot start once the cluster degrades. I can provide access to an environment within RH's VPN. The MG i was able to run seems incomplete, but this is the link just in case.

https://drive.google.com/file/d/1v3O8IKi4Wy2X_RUSyPgaP_HHnlaqGsDE/view?usp=drive_link

Affected Platforms:

OCP 4.16.x
OCP 4.14.X

Is it an

1. customer issue / SD

We identified this while trying to fix an RDS deviation for a telco customer https://issues.redhat.com/browse/ECOPS-509

This is in a customer's lab, no support case opened. Let us know if you need one.