See https://github.com/operator-framework/operator-lifecycle-manager/issues/2778 for upstream issue.

Essentially, due to the size of the OLM CSV CRD growing over time, it has now crossed over a certain threshold where it cannot be applied successfully to a cluster. This limits the ability of users to upgrade OLM in-place (which is not currently supported, but desirable) and upstream users of OLM to install the software more generally.

This may be accomplished by limiting the depth of the codegen for the OpenAPI spec that gets generated – in particular removing descriptions of embedded fields in the CSV spec like the deployment. This can be accomplished via toggling the controller-tools settings in the script OLM uses to generate its CRDs or by editing the generated schema after the fact using a text processing tool like jq/yq/etc.

A/C:

• Make CSV schema smaller so kubectl apply succeeds
• Document the process as it may be helpful for other teams with similar issues