Hi; do you have any specific recommendations? We are reevaluating our Docker strategy, perhaps to include things like Swarm or Kubernetes as distribution options.

Here's what we do now...

https://github.com/inedo/proget-docker

Hi @mikael ,

We have no plans for this and honestly, I don't think I'd recommend setting up a tool like ProGet in this manner.

Outside of some very specialized use cases (like setting up labs for testing, or nodes in a ProGet Enterprise Edge Computing Edition) I don't really see the benefits. Only headaches.

It might sound fine on paper, but every company that has set it up this way has regretted it. And you will to. The reasons they want "fully reproducible configuration" is usually:

• so we can store configuration in versioned code
• so we can easily replicate it in a testing environment
• so we can easily migrate/move to a new server

Those seem nice, but it totally fails in practice in my experience.

First, you can't "rollback" most configuration. Say you fat-finger a configuration file and delete half your feeds. There go all your packages. And when you realize you've got gigabytes/terabytes of content to deal with, plus all the metadata in storage, this is a huge headache.

The configuration you can make idempotent (say, permissions/users) is so much more a pain to work with than a UI. Again, more error prone you lose all the benefits of visual cues, input verification, etc. You fat-finger the wrong setting, and you get some obscure error instead of a helpful red box next to the text box.

The regret comes in realizing they've created a buggier environment that isn't properly tested, and is somehow less "portable" than an ordinary installation. A year later, when the new team comes in, they usually have to figure out how to "undo" it -- and you can probably guess why we need to get involved to untangle the mess.

Cheers,
Alex