I can replicate the problem on KIND using the configuration shown in K8S-2322 to add the admission controller that OCP is using with the existing chart version.

I can confirm that the change resolves the issue to allow a cluster to deploy. This includes the node selector option in the values file provided: the only tweak I had to make to the values file was the server image name (not using RHCC so should be the default of couchbase/server:6.6.2) and the storage class was commented out so it used the default.

I will ask QE (Roo Thorp) to verify the fix on OCP as well: https://github.com/couchbase-partners/helm-charts/pull/58 

I've tested Patrick's change on OCP - labelling one of the nodes as the customer does (tier=app), and then helm installing using their values.yaml from above (I changed the storage class so that the cluster could install properly, but that's the only change I made). I can see that the operator & DAC install on the labelled node - the couchbase cluster pods are spinning up on other nodes, which I think is expected as the nodeSelector was just for the DAC and operator.

The updated chart version should now be available as version 2.2.006, make sure to do a `helm repo update` to pick it up as the latest version.

https://github.com/couchbase-partners/helm-charts/releases/tag/couchbase-operator-2.2.006 

Confirmed with Tin Tran that this is all ok now.

Done in 2.2, removing the label.