Increase compute capacity of CI shared runner managers
During rolling deployments, one runner manager at a time is unavailable. As the other runner managers take on more jobs from the queue, they tip over into CPU saturation:
Each runner manager appears to be unavailable for several hours, which might be causing prolonged periods of CPU saturation.
If this is correct, then the likely mitigation is to scale the fleet vertically or horizontally.
@steveazz @tmaczukin @dawsmith