> > I am conflicted on this - I know the limitations with adding additional > interfaces. > I'd almost argue for requiring number of interfaces/IPs to be set at machine > instantiation and still handing out addresses to all interfaces via DHCP, but > not > all OSes will handle that cleanly.
As you probably know CS hands out IPs via DHCP to multiple NICs today, and it is challenging to work reliably since different guest OS's have different behavior with respect to setting/re-setting the default route in DHCP replies. Today CloudStack has some knowledge of guest OS to make this work but we've seen bugs with that where the list of guest OS behavior is not right. Those are fixable. But there are other cases (e.g. boot from ISO, a user changes the DHCP client post-create, or the user upgrades the OS which changes the DHCP client) where I don't think there is a reasonable fix. So I think we need at a minimum an option to disable DHCP reply on NICs that are not the default route. We should even consider a behavior change to make that the only option. -kevin
