cache session information somehow

[ktdreyer]

Currently every Ansible module authenticates its own unique Koji client session. This means we always send the login RPC to Koij Hub every time we want to make a change.

This is not slower than invoking the koji CLI many times, but we may be able to do better.

When we do a client login to the Koji Hub at the /ssllogin endpoint, we get back a XML-RPC response with a session-id (int) and a session-key (string). Example:

cbs -d --debug-xmlrpc hello

The Koji client then uses these two values to authenticate all the requests going forward. Subsequent HTTP requests for XML-RPC look like this:

/kojihub?session-id=123456&session-key=1234-asdf&callnum=0

where callnum is an integer that increases for every authenticated call.

In koji-ansible, we could return session-id, session-key, from a koji_login module, and have all the other modules use these variables if they are defined for the profile.

Note, we would also need to cache callnum in a way that every other koji-ansible module can increment it.

[ktdreyer]

Ansible includes "connection" plugins that might help here. I have not looked into this in depth, but one that looks interesting is "httpapi". Maybe we need a "koji" connection plugin.

https://github.com/network-automation/httpapi/blob/master/playbook.yml is one example of using httpapi.

[ktdreyer]

"httpapi" looks a bit too specific for Cisco/network devices.

However, this post describes how to create a custom connection plugin that caches a login credential: https://steampunk.si/blog/let-us-give-ansible-a-rest/ .

When we use a connection plugin, all the Koji client traffic originates from the host that runs Ansible. Right now we have some users that run the koji-ansible modules on remote nodes (via Ansible's usual SSH mechanism), and some other users running koji-ansible in containers, so we need to evaluate how to make this work in all cases.

[ktdreyer]

One project to watch here is the new Ansible "turbo" module. There was a presentation at AnsibleFest 2020. A bit unclear how widespread and battle-tested that is at this point, but an interesting approach to caching.