amazon.aws aws_ec2_inventory: An option to sort by hostname

Summary

The current implementation of ec2_inventory plugin is sorting the hosts by InstanceId. This is good enough to ensure stable ordering of the hosts within a hostgroup as long as no more nodes are added or deleted. However this implementation becomes an issue when the nodes in the group are not equal, for example, one node is supposed to be a master or initial node and treated differently than others.

With static inventories the problem has a simple solution: taking the first node of the group (e.g. hostvars[groups['db_cluster'][0]] where db_cluster includes db1, db2 and db3). This doesn't work well with aws_ec2_inventory as the first node in the group could be any of the three.

One solution could be sorting the hostnames in-place, e.g. db_primary_node: "{{ groups['db_cluster'] | sort | first }}" however we would have to do this everywhere, in all playbooks, roles and inventories where we rely on the host order. Another slightly relevant problem is run_once implementation (and everything else that is using the same logic) which makes the tasks run only on the first node in the group but skipped if the first node is skipped (see https://github.com/ansible/ansible/issues/19966).

An easy fix for these would be an option to sort hosts by their hostnames within a group.

Issue Type

Feature Idea

Component Name

aws_ec2_inventory, ec2_inventory

Additional Information

A naive hard-coded implementation:

diff --git a/collections/ansible_collections/amazon/aws/plugins/inventory/aws_ec2.py b/collections/ansible_collections/amazon/aws/plugins/inventory/aws_ec2.py
index 3806a0a2..113aba23 100644
--- a/collections/ansible_collections/amazon/aws/plugins/inventory/aws_ec2.py
+++ b/collections/ansible_collections/amazon/aws/plugins/inventory/aws_ec2.py
@@ -615,7 +615,7 @@ class InventoryModule(BaseInventoryPlugin, Constructable, Cacheable):
                     instances.append(i)
                     ids_to_ignore.append(i['InstanceId'])
 
-        instances = sorted(instances, key=lambda x: x['InstanceId'])
+        instances = sorted(instances, key=lambda x: self._get_hostname(x, self.get_option('hostnames')))
 
         return {'aws_ec2': instances}