On Tue, Feb 2, 2016 at 3:09 AM, Antti Kantee <[email protected]> wrote: > Humh, same thing. Apparently a /32 subnet mask is a thing: > https://code.google.com/p/google-compute-engine/issues/detail?id=77 > > Never encountered that one before. > > So, I guess someone(tm) has to hack the dhcp client to add an explicit route > to the gateway in the /32 case.
Yeah the networking sounds like it uses a not super popular DHCP RFC. I'm not sure I understand how adding the explicit route would fix the issue. I'm looking at rumprun/buildrump.sh/brlib/libnetconfig which I suspect is the correct place to adjust this, but the output from my instance was: dhcp: vioif0: adding IP address 10.240.0.4/32 dhcp: vioif0: adding route to 10.240.0.1/32 dhcp: vioif0: adding default route via 10.240.0.1 Which seems to me that it already added an explicit route to the gateway. So it's not clear to me what the fix is. Any suggestions? > Guess I didn't test networking when I was testing that virtio-scsi works on > GCE ... > > EC2 definitely works. The server I started in August or September seems to > still be up: http://52.2.126.166/ > (well, ok, you need to use "ec2" instead of "iso" for ec2, and a xen build > instead of hw, but almost the same thing) That's good to know, thanks! I'll keep trying GCE for now because it's what I use normally, but I guess I can fall back to ec2 if it doesn't work.
