Juju 1.25.10 is running hooks prior to additional network interfaces being up

[James Hebden]

Present in an OpenStack cloud provisioned with Juju 1.25 and the 17.02 series OpenStack charms, deploying services to LXC containers.

Charms deployed to Xenial LXCs, which are then related to hacluster services, also running under Xenial LXCs, fail to detect network interfaces in ha-relation-joined, causing all VIPs to attempt to land on the interface configured in `vip-interface` on the related service. This causes a problem when multiple VIPs are the be configured on different network interfaces, as secondary interfaces configured in the /etc/network/interfaces.d/post-cloudinit.cfg file, which is generated by cloud-init during final stages of the boot, are not up and running before juju hooks start firing.

Running the same hook after system boot shows the correct number of VIPs being configured on the correct interfaces.

An example of this is when relating nova-cloud-controller to hacluster on a xenial LXC, with one VIP on eth0, and another on eth1.
The VIP to CIDR to interface mapping is occuring correctly, and has been tested manually by using the netaddr and netifaces python modules, as is present in the nova_cc_hooks.py file in the nova-cloud-controller charm. Plotting out systemd dependency graphs, it seems that the jujud for the machine and service do not wait for the network to be fully initialised, or for cloud-init to be complete.

[John A Meinel (jameinel)]

We're unlikely to do updates to the 1.25 series at this time, but if we do, we can keep this in mind.

[Anastasia (anastasia-macmood)]

Juju team can only address bugs in Juju 1.25 that are Critical and do not have work-around.

As (a) this bug is not Critical; (b) there is a workaround - system reboot, see description; (c) this behavior is not present in Juju 2.x; I am marking it as "won't fix".

Thank you for reporting and for finding the workaround \o/