Support High Availability of Private MetalnetLBs across Multiple AZs

Summary

This is an issue concerning the High Availability (HA) of Private MetalnetLBs in the net-dp-service. The current behavior of the service does not support ECMP (Equal-Cost Multi-Path), which limits the ability to use multiple private load balancers in a HA mode.

In the scenario of Private MetalnetLBs, traffic is not routed through the router. As a result, only a single copy of a Private MetalnetLB can exist within a given VNI, in a given VM-cluster. This restriction impedes the ability to spread PrivateLB across multiple Availability Zones (AZs).

Basic example

In cases where there are multiple AZ/VM-clusters, each having its own copy of a given API-PrivateLB (in turn a copy of PrivateMetalnetLB), where the same Network is spread across multiple clusters/AZs, we encounter conflicts. This setup leads to the creation of multiple PrivateMetalnetLBs within a single VNI. Consequently, the load balancing functionality might not perform as expected.

Motivation

To ensure effective load balancing and the high availability of Private MetalnetLBs across multiple clusters and AZs.

Begin by tracing net-dp-service's Private MetalnetLB handling and its ECMP and routing behavior; the issue names no files or tests. Understand how VNI, VM-cluster, and Availability Zone placement currently interact, then define the multi-cluster scenario and validation needed for Private MetalnetLB high availability without conflicts.