Fix nova scheduler down after first docker restart

Due to a Docker bug [1] we cannot use Docker to send
SIGHUP to the container because it will mark it as
stopped.
This patch sends the signal directly to the process,
bypassing Docker.

'changed_when: false' is also removed from the
relevant task as it definitely changes the state.
In the future we could do the refresh only if
there really is a need for another one.

[1] https://github.com/moby/moby/issues/11065



Change-Id: Ief73bbd24568d6941384ea3330ab45f11aa42d37
Co-authored-by: Radosław Piliszek <radoslaw.piliszek@gmail.com>
Closes-Bug: #1845244

ansible/roles/nova/tasks/refresh_scheduler_cell_cache.yml

@@ -3,7 +3,10 @@
 # in nova scheduler.
 - name: Refresh cell cache in nova scheduler
   become: true
-  command: docker kill --signal HUP nova_scheduler
-  changed_when: False
+  # NOTE(yoctozepto): Normally we would send the signal via Docker but, due to a
+  # Docker bug (https://github.com/moby/moby/issues/11065), this might cause the
+  # container to be stopped if we restart Docker or reboot the server as we
+  # use the 'unless-stopped' restart policy by default.
+  shell: "kill -HUP `docker inspect -f '{% raw %}{{.State.Pid}}{% endraw %}' nova_scheduler`"
   when:
     - inventory_hostname in groups['nova-scheduler']