Forgive me if this is answered elsewhere. I promise I've performed the requisite searching, lurking, and cursing.
Why does it seem like there's no way to properly automate the management of this software? Sure we've spent man-*years* in effect getting Ansible to deploy a cluster with adequate functionality, but even that can't touch the immediate config drift required just to add/remove resources, roles, and manage service run states. (Who the hell thinks poking around in a UI is operationally sound?)
We've leveraged some of the functionality exposed by the API into some python based modules, but they're not idempotent and add terrible complexity to the stack. Why build config management if the effort to replicate it is as costly and error-prone as going without automation completely?
Yes, we took a look a Director. Is that a joke? Just spinning up the tiniest of POCs resulted in useless silent failures with no useful output or results AND several expensive AWS instances left spinning idly. And nothing about Director fits into any kind of formal CM workflow I can see. Looks more like another UI contrivance around some API calls and a set of 'bootstrap' scripts that are just a notch up from raw shell scripts.
Sorry if this sounds like just a bunch of bitching and moaning. We've spent WAY too much time trying to automate our hadoop implementations and it's never not a freaking nightmare.
Thank you for the feedback. I have forwarded the information to the appropriate teams to make them aware.