Previously I bloggged about the cool things I did with NixOS. After publishing the post, my friend Anton promptly asked:

Wondering what’s your professional take on NixOS. Would you give it a short for a small-to-medium size server fleet provisioning? It felt rather involved and not very mature when I looked at it. Kind of a commitment, too.

Here is my response:

My journey to NixOS has been bumpy ride: it’s been over a year since I looked at first, and I still sometimes feel I did not escape the beginner level. The learning curve is steep, and it is best to take it on gently or have a good mentor nearby. I started by installing NixOS on my primary laptop, which was a mistake. The annoyance of “I can do this in Debian in 5 seconds, and I am an hour in without an end of sight in this thing” was very discouraging at times.

I reinstalled my laptop back to Debian and took a few slow months to provision 2 personal servers (the thing that’s detailed in the blog). Taking it slow has been fantastic experience. The folks in Matrix are very helpful where documentation, especially high-level, is patchy. Now I feel comfortable enough to retry NixOS on my laptop again.

Recently I realized that what I originally perceived as immaturity later turned out lack of knowledge and/or lack of high-level documentation. Technicals are good. Granted, I have found some bugs (though trivially fixed), but they mostly come from the power to configure it and thus the huge surface area. Also, variety does not help: for example, there are 10 deployment tools in the wiki (“nixops related” counts too). It is hard to choose when I don’t know what to expect, much less know what’s possible. It is also nontrivial to ask for a “high-level” advice: a beginner will just tell their favorite system, not knowing the trade-offs or alternatives. An expert will tell “depends on what you want to do”. Moving beyond such answer requires time and a beverage, which brings it’s own constraints. In this concrete case, I spent quite some time learning krops, which later turned out to be a dead-end. Later moved to deploy-rs, which turned out to be a good decision so far.

As far as recommendations go. For smaller companies, especially where developers are also taking care of operations/deployments/infrastructure, I can’t recommend NixOS enough. For medium-large size companies it would certainly bring a lot of value (I can already see how many things mine or my sister-team at Uber had to re-implement which come out of the box in NixOS), but, like with anything that has a different paradigm, requires a mentality shift, which may be very hard organizationally.

There is at least one large-ish company I know that uses NixOS (proof). I did not look, I found it by accident. I also know a few folks in Tweag; their primary consulting stream is helping companies onboard to Bazel and/or Nix. They won’t tell who they are, but there are “quite a few, of different sizes, flying under the radar”.