Open Stack

Hi,

Starting sometime earlier in the week, gate jobs started failing that were 
running in the OSIC cloud, due to a loss of connectivity to VMs running there 
when Neutron L3 was configured.  I wanted to send some information out on what 
we found in case other deployment tools trip over the same issue.

The problem was the VMs in OSIC are IPv6-only by default, so must be reachable 
using a global IPv6 address.  At boot, everything was fine - IPv6 address and 
default route were configured, but when IPv6 forwarding was enabled, poof!  The 
problem is that the default behavior in Linux is to remove any IPv6 default 
routes when forwarding is enabled, and that action caused connectivity loss to 
systems outside the OSIC datacenter using IPv6.  Luckily there's a sysctl to 
control this, and there is a patch out [1] that is working it's way through the 
check queue now.

So if you're creating puppet, chef, etc, scripts and deploying in an IPv6-only 
environment, you might need a few tweaks to not hit the same issue.

-Brian

[1] https://review.openstack.org/#/c/359490/