Feature Overview

Operators (both CVO managed and OLM managed) are used to control the deployment, configuration and life cycle management of OpenShift components, layered products and products from our ecosystem. 

Red Hat Operators are very opinionated (which is good) but typically expose a limited number of configuration knobs and other logic. And while this works in many cases, we have customers that require more control. User changes that may be required to work around lack of a config option or to recover from an outage often requires the operator to be 'unmanaged' so the underlying Kube resources can be modified. This can lead to blocking of upgrades.

The Operator Framework defines a maturity model that really starts to get interesting in Level 4 and 5. Most CVO Operators can be graded at Level 3 and while they are excellent at running/upgrading/recovering their software stack, they could be enhanced to do things like:

• auto-tuning themselves based the size of the cluster (eg. Single Node OpenShift)
• reconfigure based on load they are experiencing (eg. Monitoring getting tons of user workload metrics)
• reconfigure based on cluster configuration (eg. being managed by ACM)
• auto-prune based on overall etcd object count or user defined settings for keeping history

Goals

• Explore a mechanism to provide a middle ground between managed and unmanaged - an ability to define what is supported and what is not as a result of changes done by users above and beyond documented and exposes operator knobs. A sort of partially supported unmanaged middle ground. 
• Mature operators by adding auto-tuning and auto-pruning
• Deliver a design and a reference implementation for a TBD operator.