On 26/08/16 17:18, Merlijn Sebrechts wrote: > > The talk Mark gave at cfgmgmtcamp in Ghent clearly left a strong > impression. It still comes up in conversations with devops folks. > However, such a conversation quickly goes into the direction of "It > sounds cool but people will never trust other people's ops code. Look > at the poor quality of containers in docker hub. I don't think Juju > can change that". > > Juju has great answers to this problem. I think the biggest one is > that Juju makes it easy for charms to live in the upstream's repo. > "You can trust the quality of the Openstack charms because they are > created by the Openstack team." Juju's interfaces make this really > easy. Upstreams don't have to think about the millions of possible > combinations of services. They just focus on their own service and > their job ends where the interface begins. The poor dependency > management of other devops tools make this a lot harder. >
That's exactly right, you have the whole story in a nutshell :) Juju encapsulates ONLY the application. You don't get all the ancillary decisions that the ops needed to take, like which monitoring system they like, built into the charm. You JUST get the operations of that application. That means the operations are immediately more likely to be useful to someone else, and hence reusable. It also means that the upstream, who really knows the application well, can steer all the operators of the application onto a common set of patterns and behaviours. Juju also encourages the encapsulation to be blind as to the model in which it will be deployed. The charm doesn't know in advance if its going to be in production or in devel, if it will be at large scale or small scale. That again makes it much easier to build a community around the charms that reuses one another's work for a wide range of operational scenarios. I think the most important step now is the upstreaming of important charms. But we can't do that quietly from a central office - we need to figure out how to inspire upstreams to see these benefits for themselves. I see very encouraging stuff in OpenStack (the charms are now officially upstream, maintained as part of the OpenStack project, albeit alongside a million other projects in OpenStack) and in Kubernetes (the charms are in the primary upstream repos). Apache Bigtop is an obvious place to make good progress but I get the impression that's stalled on a bunch of people asking "why?". Seems like Merlijn you could provide a strong answer to that :) > Another thing that comes up a lot is that companies already use a > combination of different devops tools and it's getting harder and > harder to get them to work together. Juju is a great solution to this > problem: Want to get a docker container to talk to a service that's > managed by your own snowflake in-house devops tool? No problem, wrap > them both in a Charm and they can work together! Yes exactly! > Not much of a conclusion here, I just wanted to let you know what I > hear a lot in conversations about Juju. It might be good to highlight > these qualities when you do another one of those mind-blowing talks. Will do. I believe there is another ConfigMgmtCamp coming up! Mark -- Juju mailing list [email protected] Modify settings or unsubscribe at: https://lists.ubuntu.com/mailman/listinfo/juju
