Add weight decrease to keepalived checks

Currently keepalived ping checks don't have any weight configuration,
which means that if a ping check fails, the VIP will transition to FAULT
state. This also means, that if ping check fails for all keepalived
instances, the VIP will be unreachable. In the general use case this is
undesirable as we don't want for example hitting a icmp rate limit or a
routing issue causing problems to just reach the ping endpoint to cause
a full outage.

This commit instead makes keepalived use weight decrease to still keep
the behavior of failing VIP over if a node loses connectivity, but if
all nodes are having failed connectivity VIP will still be kept up. The
value of 101 is selected to make sure a failed node will always have a
lower effective priority than any node that has not failed.

Change-Id: I618329785e86c8a6b1f4a41b90601c2a908f3fea

inventory/group_vars/haproxy/keepalived.yml

@@ -37,6 +37,7 @@ keepalived_scripts:
     src_check_script: "{{ playbook_dir }}/../scripts/keepalived_pingable_check.sh"
     interval: "{{ keepalived_ping_interval }}"
     instance: external
+    weight: -101
     fall: 2
     rise: 4
   pingable_check_script_internal:
@@ -45,6 +46,7 @@ keepalived_scripts:
     src_check_script: "{{ playbook_dir }}/../scripts/keepalived_pingable_check.sh"
     interval: "{{ keepalived_ping_interval }}"
     instance: internal
+    weight: -101
     fall: 2
     rise: 4