You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Support multiple network interfaces per workload instance #442
Compute instances are limited to one network interface. This prevents network appliances and other multi-homed workloads from connecting to separate tenant networks—for example, a VyOS instance with management, WAN, and LAN interfaces.
Desired outcome
Customers can declare multiple named network interfaces on a workload. Each interface is attached to its selected network, receives independent addressing and status, and keeps a stable guest-visible identity across instance replacement.
One interface must be designated as primary for default routing and primary-IP presentation. Additional interfaces should be governed by project quota, while every instance continues to receive its first interface by default.
Acceptance criteria
A workload can request more than one named network interface, subject to the selected runtime class's supported limit.
Every declared interface is provisioned, attached, and reported independently.
The primary interface and default-route behavior are deterministic.
Adding, removing, or changing interfaces results in a safe instance rollout with no orphaned interface allocations.
Quota can limit additional interfaces per project without hiding the API behind a feature flag.
CLI output and documentation make all interfaces and their readiness visible.
End-to-end coverage demonstrates a multi-homed VM or sandbox attached to at least three networks.
Dependencies
Requires corresponding multi-interface support in the cell networking path, including Multus injection, Galactic attachment lifecycle, and the active runtime provider.
Activity