Windows: management dial stuck on metric-preferred NIC with no upstream (no interface fallback)

[siraykut]

Describe the problem

On Windows with advanced routing (IP_UNICAST_IF), the client binds management/signal gRPC dials to the metric-preferred interface from the routing table. If that NIC is up and has a default route but no working upstream connectivity, dials time out and NetBird never falls back to another interface that can reach the management server.

User-context HTTPS to the same host can succeed (Windows chooses a working path), while the NetBird service (LocalSystem) fails with:

failed creating connection to Management Service: create connection: dial context: context deadline exceeded
nbnet.NewDialer().DialContext: dial tcp <mgmt-ip>:443: i/o timeout

To Reproduce

1. Windows host with two connected IPv4 interfaces, both with a default route:
   • Ethernet: lower interface metric, no working internet to the management host
   • Wi-Fi (e.g. phone hotspot 172.20.10.0/28): higher metric, can reach management over HTTPS
2. Install NetBird client (reproduced on 0.76.1) pointed at a self-hosted management URL on :443
3. Run netbird up
4. Observe management dial timeouts; netbird status stays Management: Disconnected

Expected Behavior

Management dial should succeed via a working interface (Wi-Fi in this case), even if a lower-metric NIC has a non-working default route.

Are you using NetBird Cloud?

No — self-hosted (netbird.planatechnologies.io).

NetBird version

0.76.1 (windows/amd64)

Evidence / findings

Probe	Result
Test-NetConnection mgmt:443 (user)	success
Invoke-WebRequest https://mgmt (user)	HTTP 200
curl --interface <wifi-ip> https://mgmt	HTTP 200 (~0.37s)
curl --interface <ethernet-ip> https://mgmt	connection timed out
NetBird service dial via nbnet.NewDialer()	i/o timeout ~20s, then canceled

Workaround that immediately fixed it:

Set-NetIPInterface -InterfaceAlias Ethernet -AddressFamily IPv4 -InterfaceMetric 80
Set-NetIPInterface -InterfaceAlias Wi-Fi -AddressFamily IPv4 -InterfaceMetric 10
Restart-Service Netbird
netbird up

After that: Management/Signal Connected, peers 4/4.

Relevant code path

• client/grpc/dialer_generic.go → nbnet.NewDialer().DialContext
• client/net/dialer_init_windows.go → Control = applyUnicastIFToSocket
• client/net/net_windows.go → selectInterface / GetBestInterfaceFunc
• client/internal/routemanager/systemops/systemops_windows.go → GetBestInterface returns only the first sorted candidate

Proposed fix

1. Expose all usable candidate interfaces (best-first), not only the top one
2. On Windows advanced-routing dials, try candidates with a split timeout budget so the first dead NIC cannot consume the entire context deadline
3. Log when a fallback interface succeeds

I will open a PR with this approach.