Hi Chris, On Sat, Feb 17, 2018 at 11:37 PM, Chris Marusich <cmmarus...@gmail.com> wrote: > Hi, > > At FOSDEM, some of us discussed "orchestration", which means something > like "how to deploy services to more than 1 machine in a coordinated > fashion". Many people contributed to the discussion. I took notes. > I've thought about this more, reviewed the "wip-deploy" branch, and > written up my thoughts in the attached file. > > It's a rough sketch of ideas, biased with my own opinions and > experience, but I think it's good enough to share. I invite you to > improve upon it: share your own thoughts, hack some code together, and > just iterate on this a bit, so we can make some progress. > > Hopefully, we can agree on a basic design and get a working proof of > concept. Then we can make a blog post about it!
These are good notes, thanks for sharing them! One additional use-case I would consider for an orchestration tool would be so-called "immutable deployment", where virtual machines are replaced entirely rather than updated in-place. This is commonly used for deploying web applications into auto scaling groups (where the actual number of hosts at any given time is dynamic) using a "blue-green" deployment technique (in a nutshell it's a double buffer that allows updating the application without downtime and allows easy rollback in the event the deploy breaks critical functionality). This is the kind of thing that I do at my day job, and we are but one of many companies that do things this way. - Dave
