labs: sleep in shutdown_controller.sh

After building the cluster with the two additional snapshots on the
controller VM, we end up with a dying nova-compute service.

Sleeping 2+ seconds after booting into compute and before rebooting
controller fixes the problem: nova-compute keeps running and its status
in nova-manage service list will go to XXX and back to :-) without user
intervention.

Change-Id: I8ba20c77392b0966d6febfc300790cf1943f2ad5

labs/scripts/shutdown_controller.sh

@@ -8,8 +8,22 @@ indicate_current_auto
 
 exec_logfile
 
-echo "Shutting down the controller node."
+# At this point in the cluster build, we just rebooted the compute VM to take
+# a snapshot, and we are about to reboot the controller node for the same
+# purpose.
+#
+# About a minute after we reboot the controller, the status of nova-compute
+# (according to nova-manage service list) becomes "XXX".
+#
+# If we sleep for 2 seconds now, before rebooting the controller, the
+# nova-compute service on the compute node will keep running and the status
+# will automatically return to ":-)" after some time (may take several
+# minutes). If we don't sleep here, the nova-compute service on compute will
+# die within a few minutes (needs manual service restart or a compute node
+# reboot).
+sleep 2
 
+echo "Shutting down the controller node."
 ssh \
     -o "UserKnownHostsFile /dev/null" \
     -o "StrictHostKeyChecking no" \