ec2 host variables lost with refresh_inventory

[Jeff Hardy]

Hello all,

I have been working with a few different patterns for launching and configuring ec2 hosts, first with add_host, and more recently working with refresh_inventory.  The problem I am having is that all ec2_ host variables are lost with the introduction of the refresh_inventory task.  Sample playbook:

---
- hosts: localhost
  connection: local
  gather_facts: false
  roles:
    - aws_ec2    # standard ec2 deployment based on include vars
  tasks:
    - meta: refresh_inventory

- hosts: tag_Name_somehosts
  tasks:
    - debug: var=hostvars[inventory_hostname]
    - debug: var=ec2_tag_Name

Running the playbook specifies an inventory directory, containing ec2.py, a static hosts laying out some groups, group_vars, etc:

ansible-playbook -i inventory ec2.yml

I can see in the debug output that the ec2 host variables are lacking, yet all the tag, instance, and sg-based groups exist.  This fact is also proven because the host declaration itself for the second play relies on one of those groups.  But the second task echoing out the ec2_tag_Name fails:

ok: [10.1.1.1] => {
    "ec2_tag_Name": "VARIABLE IS NOT DEFINED!"
}

Of course if I leave off the refresh, the playbook will succeed and those variables exist, except for that very first run where the instances were not in inventory and the group does not exist:

PLAY [tag_Name_somehosts] **************************************************
skipping: no hosts matched

As an aside, when I introduced refresh_inventory to the typical add_host playbook pattern, that re-init of inventory seemed to eliminate the ad-hoc ec2launch group I had created (probably expected).

Ultimately, if any tasks or roles in the second post-ec2-provisioning play rely on ec2_* variables, I am afraid refresh_inventory is deleting them.  I flipped on and off gather_facts for the second play and it made no difference.

Has anyone else run into this or have an idea whether I should be doing something differently? 

-Jeff

[Barry Kaplan]

I am seeing similar behavior with 2.1. If I use refresh_inventory some of the hosts are removed and plays subsequently fail.

[Jeff Hardy]

I should have indicated, I see the same with both 2.0.2 and 2.1.0.  When I first ran into the issue, the only thing I found related was https://github.com/ansible/ansible/issues/15115, which does not seem to be the same problem.