Feature Request: Automatic Failover for Overlapping CIDRs (High Availability)

[3azmeo]

Summary

I am requesting the implementation of automatic routing failover when multiple sites (Newt agents) advertise the same Private Resource (CIDR or Host). Currently, if the active site goes down, traffic does not switch to the available backup sites.

Motivation

I am migrating from Twingate to Pangolin for a self-hosted Zero Trust setup and need redundancy (High Availability).

Currently, the documentation states that Pangolin "arbitrarily chooses a single site" when resolving overlapping networks. In my testing, if I have two gateways (Site A and Site B) advertising the same subnet (e.g., 10.0.0.0/24), the client binds to one. If that specific gateway goes offline (container stop or network failure), the connection times out and does not switch to the healthy gateway.

This creates a single point of failure, making it impossible to run a resilient network compared to other solutions like Twingate Connectors or Tailscale Subnet Routers.

Proposed Solution

The client (or controller) should be "health-aware" regarding overlapping routes:

1. Route Redundancy: When the client receives the map, it should be aware that multiple sites offer a route to the same destination.
2. Active Failover: If the connection to the currently selected site fails (handshake timeout or agent disconnect), the client should automatically update its routing table to use the next available site advertising that resource.
3. Priority (Optional): Ideally, allow setting a "Priority" field in the dashboard to prefer one site over another (Primary vs. Backup).

Alternatives Considered

1. Manual Intervention: Currently, I have to manually log in to the dashboard and delete the resource associated with the dead site to force traffic to the healthy one.
2. Hypervisor HA: Using Proxmox HA to restart the Newt LXC/VM on another node. This is slow (reboot time) and does not solve network-level disconnects where the VM stays running but loses internet.

Additional Context

I verified this behavior by deploying Newt on two separate physical hosts advertising the same CIDR.

• Step 1: Client connected successfully via Host A.
• Step 2: I stopped the Newt container on Host A.
• Result: Connectivity was lost completely. Traffic did not reroute to Host B despite Host B being online and configured correctly.