Require that routes have matching clusters

[Tratcher]

Today routes and clusters are validated individually before officially loading the model. However, this step does not validate if every route references a real cluster. When the model is rendered later it's possible that a route will be created that has no matching cluster.

Today this is allowed and then checks are done in middleware to short circuit routes without a cluster. https://github.com/microsoft/reverse-proxy/blob/d9eeb548817364c7199acadd177f699781091a48/src/ReverseProxy/Middleware/DestinationInitializerMiddleware.cs#L32-L37

I reviewed this with @davidni and he said the expected scenarios for this (a more dynamic pipeline) never materialized.

Proposal:

• Check that each route references a valid cluster before accepting the config as valid.
• Remove the associated checks from middleware.

[Kahbazi]

While we are enforcing the validation, how about to also throw an error if a Cluster has no Destination?

[Tratcher]

While we are enforcing the validation, how about to also throw an error if a Cluster has no Destination?

I'm not sure, destinations can be much more ephemeral. E.g. You may create the route and cluster first, and then incrementally add destinations as they come online. The system already has to handle the case where there are no healthy destinations, so having no destinations isn't much worse.

[samsp-msft]

From an extensibility perspective, I would prefer the validation to be done as late as possible, preferably at pipeline stages that need data. That way people can add their own stages to the pipeline that may add/remove or otherwise change the data attached to an individual request.

[karelz]

Triage: We should validate earlier -- we can loosen it later, when extensibility features require that.

[samsp-msft]

We should discuss this in the context of extensibility around cluster selection.