TL;DR there is a new testing tool for Juju and you should check it out at https://github.com/juju-solutions/matrix
Paraphrasing Netflix, "If you can't run chaos monkey in production you're not really prepared to deal with failure". And it's true. It quickly becomes a question of constraints if doing this sort of thing is actually possible. Maybe it costs too much to scale out at this level, to provide full HA with no SPoF in your deployment. And if that isn't something you alone are willing to pay for that is fair, but these are Juju Charms, shared DevOps. Because of that we can share the costs and know that if we do want to scale up, if we do demand HA services ensure that they survive failures we can work with others to ensure the Charms are up to snuff. If we're lucky the Charm already are. But luck and production software are fickle friends, so how can we know that our deployment will survive a real world runtime, or better yet, the real world runtime that is production? Traditionally that has fallen on individual Charm authors to test their software. Did they test the upgrade from one version to another? Did they ensure that all the units could survive reboot? And many other things. If you begin to feel skeptical that each Charm you depend on has an author that is as paranoid as you are about their production deployment as your are about yours then you want a way to verify, before deploying, upgrading or otherwise operating a new Charm. We've felt skeptical too, we've wanted proof. Most people can never ask, "Did this deployment work in Production elsewhere?". That is a function of Juju and Charms, the elevation of deployment from custom wizardry to the wisdom of the crowds. A Juju deployment can and should work for many people, its test cases can and should be as rigorous as the most demanding. We are not there today, but we're working on it and we'd love your input. This project was introduced at the last Charmers summit and has come some distance since then but is still pretty alpha. That is good and bad for you. Today this isn't automatically gating access to the Charm store's stable channel. Today this doesn't have a full suite testing all the operations that Juju supports. Tomorrow it can and will. With your help we're developing a testing framework that automatically tells you how Charms and Bundles perform under all sorts of adverse conditions. Do we expect everything to work when machines and networks are failing underneath them? No, but, and this is the most important part, we expect that for any of those failures the people maintaining the Charms and upstream software should be able to derive useful insights from this tool. We call it Matrix (as in "welcome to the real world") and it can do things like run automatic tests on all the basic operation Juju supports, like deploy, upgrade, and scale. In addition it can introduce random or fixed failures on those deployments to both better understand how the software works (and to improve the Charms) and to act as a gate to the Charm store. We are interested in your feedback, it is still early days so you can have plenty of input on how it works. https://github.com/juju-solutions/matrix Thanks, The Matrix Team.
-- Juju mailing list Juju@lists.ubuntu.com Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/juju
