From downstream bug report:
While deploying an Overcloud that is utilizing IPv6 addressing, we sometimes see where the pinging to the controllers fails as part of the deployment validation which then results in a failed deploy. Through troubleshooting, we've determined that pings to the gateways do not fail and when pinging the gateway, it appears to "fix" the ability to ping to the Controller Nodes.
Within our team we've tried to figure out why this would be the case and one of our team members pointed out that IPV6 does not have ARP, it has neighbor solicitation and router solicitation, and sometimes neighbor solicitation isn't on or doesn't work, in which case you -have- to establish connectivity to the router before any other host communication works. He said this only applies to non eui64 (mac address embedded in ipv6 address) setups.
We're not sure if that's what is happening here, however, we do know if we continually run an Ansible Playbook that connects tot he overcloud nodes and has them ping their default gateways during deployment, the deployment succeeds 100% of the time.
In looking at the all_nodes.sh validation script:
/usr/share/openstack-tripleo-heat-templates/validation-scripts/all-nodes.sh
We noticed that it is performing 2 tests, ping_controller_ips and ping_default_gateways, in that order. We believe flipping these around will correct this issue.
This BZ is to have those 2 tests swapped so the ping of the gateways happens before pinging the controllers.
Fix proposed to branch: master
Review: https:/