Implement basic k8s operator

[dbrewster]

Will assume custom code is built in the eidolon image. Machine name === namespace One machine per namespace

The operator will:

• Have CRD's for every resource type in eidolon and needs to be built dynamically. Question: Can we do this without typing and can this be totally dynamically based on the version / container of Eidolon. What do we do if the user defines a new spec on an agent? etc...
• The only resource that triggers a new "pod" will be associated with the machine. Every other resource is assigned to the machine. Question: How do we validate the resources separately so that we can report errors for each resource? I.e. when an agent fails to start or when a component crashes...
• Updating a resource ALWAYS upgrades / restarts the deployment.
• The machine name will === the namespace at this time.
• Every resource will be applied with kubectl

[dbrewster]

Maybe we can cycle the server in stages. I.e. when something new is applied:

• If server running, verify with existing server
• If not, create a bare server with container and verify with that.
• We need to verify with the same container that is in the machine so if the machine changes we need to shutdown the old machine first.

[dbrewster]

Another idea. Just add a /health endpoint to the server. It will report the health of every component.

So the process is:

1. Given the current set of config, create a ConfigMap resource for those. Create pod for the machine. Set the state of pod to initializing.
2. If pod in initializing, start the code, set state to starting for every resource type.
3. Call /health until the server responds or times out.
4. Set the state of every component, including the machine, based on this output.
5. Return done if every state is == running or error.

I think this is the right approach.