SPIKE: Central API ML clustering

[pinpan]

The Central API ML should be able to run off-platform with HA support.

Load balancing and fail-over capabilities are an integral part of the API ML GW when routing to onboarded services. When run off-platform however, the API ML GW component doesn't provide load balancing between its own multiple instances. The HA needs to be provided by an external proxying component. Possible options are:

• Run API ML in orchestrated containers - Kubernetes.
• Deploy a general proxy in front of API ML GW to distribute the load and bridge instances failures.
• Delegate the load balancing to the client application. If the client provides fail-over it should be able to catch failures in API ML GW instance and repeat the request through a different instance of the client, which can be configured to communicate with a different dedicated instance of the API ML GW.