We've recently observed a race condition between the assemblers coming up and connected to the geometry device, and the geometry device sending its geometry. This MR is an attempt at avoiding this race condition by waiting (up to 3s) for the assemblers to be active.
