In newton we are setting both nova::host and neutron::host values
explicitly to the fqdn.
This can cause problem during upgrade from Mitaka. The previous host
value (defaulting to python socket.gethostname) could return only the
hostname[0]. It means that during upgrade we are changing this
identifier. At restart nova/neutron creates *new* agents. Those
agents are then unaware of existing workload.
For neutron, the problem is that due to [1] and the fact the L3 agents
are in HA mode, the previous defined workloads on those agent get lost
and FIPs become unreachable.
For nova it's no longer possible to send commands to (before upgrade)
existing vm anymore.
This patch checks the current live value of the host parameter through
a fact and set the nova::host and neutron::host value to it if we are
not in a deployment (upgrade/update)
For nova, we directly use nova-manage to get the current live value.
Using the mysql parameter directly has the advantage that it's defined
on all types of node (controller *and* compute). As a matter of fact
the required auth parameters are usually not defined on compute node.
For neutron, when auth is available in the configuration (on
Controller) we use that. There is no neutron-manage equivalent here
so we use the nova value when auth is unavailable. When host is unset
they both use python.gethostname, so it should be the same value.
Using auth on controller add another level of confidence though. And
the controller are where the l3 agents are, so better be safe than
sorry.
This patch is newton only as it's where we are setting for the first
time this parameter. After that (ocata on) we use[2] to make sure
that those parameters are never rewritten.
[0] https://bugzilla.redhat.com/show_bug.cgi?id=1499201
[1] https://review.openstack.org/#/c/560855/
[2] need to be backported to ocata https://review.openstack.org/#/q/I8f075a5ad869ef0dc72a700dcb7be0b6efca787a
Partial-Bug: #1763322
Change-Id: Ieb92ff161d1684c214382c5eb6b5949efc3fe75c
6938b5d1f2