debian_hypervisor: synchronize team0_x port state with OVS

On hypervisor shutdown, OVS is stopped a long time before the team0_x
ports are shutdown as well.

In the meantime:
* The switching done by OVS is stopped
* The corosync totem stops circulating and nodes are disconnected from
  the pacemake cluster which leads to resources being stopped.
* After a while, RTSP detects this and change ports configuration to
  avoid the shutdown node and corosync/pacemaker start back with the
  left nodes.

To avoid this, use systemd services to shutdown team0_x ports as soon as
OVS is stopped to signal RTSP to change the configuration "instantly"
and thus, avoid the corosync totem timeout.

Signed-off-by: Yoann Congal <[email protected]>

roles/debian_hypervisor/files/team0_x@.service

@@ -0,0 +1,20 @@
+# Systemd unit wrapping around team0_* ports.
+# This can be used to synchronize ports link state with the service that use
+# them.
+
+[Unit]
+After=network.target
+DefaultDependencies=false
+Conflicts=shutdown.target
+Before=shutdown.target
+StopWhenUnneeded=true
+
+[Service]
+Type=oneshot
+RemainAfterExit=yes
+ExecStart=networkctl up %i
+ExecStop=networkctl down %i
+
+[Install]
+# empty: This unit must be activated via Wants/Requires of other units that need
+# it

roles/debian_hypervisor/tasks/main.yml

@@ -210,9 +210,17 @@
     group: root
     mode: 0644
   register: ovsvswitchd
+- name: Copy [email protected] file
+  ansible.builtin.copy:
+    src: [email protected]
+    dest: /etc/systemd/system/[email protected]
+    owner: root
+    group: root
+    mode: 0644
+  register: team0_x_service
 - name: Restart ovs-vswitchd
   ansible.builtin.systemd:
     state: restarted
     daemon_reload: yes
     name: ovs-vswitchd
-  when: ovsvswitchd.changed
+  when: ovsvswitchd.changed or team0_x_service.changed

roles/debian_hypervisor/templates/ovs-vswitchd_override.conf.j2

@@ -1,2 +1,6 @@
 [Service]
 Slice=ovs.slice
+
+[Unit]
+Requires=team0_x@{{team0_0}}.service
+Requires=team0_x@{{team0_1}}.service