I think the additional features of devise that are missing from guardian are worth thinking through in terms of schedule and implementation time, but rollout and unread? These are both so trivial to write that I’m half-surprised that there’s even a gem for it in Ruby. I’d hardly put this down as a black mark against the elixir system—indeed, I think its symptomatic of what’s wrong with the Ruby ecosystem that the cost of an additional dependency would make these gems worthwhile.

For me the biggest pain point in moving my app over is the absence of a clean newrelic integration. There do exist a couple of hex packages that attempt to use newrelic’s apm library, but that’s a NIF strategy I’m not really comfortable with in production yet. I’m using Elixometer instead of exometer now; its decidedly more streamlined to use, but the jury’s still out on an overall monitoring and instrumentation strategy for me.

FWIW I think the dependency evalution process you went through here is a sound strategy for deciding whether to do a green-field rewrite of an app in Elixir, but that is not usually the way I’d encourage a team to go. feature-by-feature migration I’ve found is a more realistic path, or at least has been for me.