Zero Downtime Updatability for dp-service

[guvenc]

There is the need for a concept of "Zero downtime updatability" for dp-service. At the moment update of dp-service would cause all the running connections break and re-establish. This may not be desirable in many customer use-cases, although the down-time would be less than a minute with the current dp-service implementation.

Other than developing a concept in dp-service and implementing it, the following ideas could be a workaround till a final solution is available:

• A concept of VM immigration where all VMs can be immigrated to another node without traffic disruption and dp-service can be updated on a VM-free node.
• If the workloads in the VM are running using Kubernetes then the workload pods can be "drained" to other nodes and dp-service can be updated on the node where the pods are completely drained.

[guvenc]

a workaround and a single digit downtime for dp-service upgrade provided with #340
CC: @dnlvanco