Enable dynamic changes in the router for non-kube environments

[nluaces]

Implements #2337
Branch based on #2334, the proper commits that implement this issue are bb8ab35 and the following.

flowchart TD
   id1([new listener resource is created])-->|triggers|InputResourceHandler-->Q{site exists?}
   Q -->|Yes| A[Refresh Router Config]
   Q -->|No| B[Bootstrap]

Loading

flowchart LR
  SystemAdaptorHandler-->|has a callback|RouterStatusHandler
  SystemAdaptorHandler-->|spawns|processRouterConfig-->id2([reconciliates the router config file with the router])

Loading

The InputResourceHandler was modified to not bootstrap in case a site is already created in the namespace.

EDIT: A new flag called --reload-type is available in the skupper system install command. It has priority over the env. variable.

[fgiorgetti]

My impression so far is that it is on the right track.
Really nice work @nluaces!

A few other comments I have so far:

• The site name in the router config is changing (we need to check if this will cause issues to the console or vanflow metrics)
• As I already added, the system controller update and the update of the router container for the controlled namespaces is something we need to figure out still

[nluaces]

• The site name in the router config is changing (we need to check if this will cause issues to the console or vanflow metrics)

I checked that the name of the site was not changing when the controller just refresh (for example, when a listener has been created); do you have any examples for me to understand this better?

[fgiorgetti]

The system stop command should use a similar validation done to the system reload command, which performs the following check:

        if nsPlatform != currentPlatform {
            return nil, fmt.Errorf("existing namespace uses %q platform and it cannot change to %q", nsPlatform, currentPlatform)
        }

At present if someone runs system stop --platform podman but the actual platform is docker, the container is being left behind.

[fgiorgetti]

Just before doing this, I'd recommend checking if the CONTAINER_ENDPOINT environment variable is set, if so, just return its value.

[fgiorgetti]

I believe you can keep this one. If the CONTAINER_ENDPOINT is set, use it, otherwise what you have looks correct.

[fgiorgetti]

I believe that this is misplaced. When the system-controller is running with reload-type auto, the validation above will always fail, if the system-controller is running with docker.

If this gets moved to internal/cmd/skupper/system/nonkube/system_stop.go (ValidateInput), I believe it should produce the desired effect.