Django. I'm not doing software these days but still following the news etc., and it seems to me that the general knowledge of the tool is still operable since very early releases. I'm not very familiar with Rails itself but when an environment changes very quickly it's both a big distraction and a great inconvenience. I've last worked with django around January 2014, today when I look at a Django app it's the same beast but a bit better. I was also interested in node in its first days, then about 0.12 I lost my interest. Each time I look at it it's a new thing. If Rails is more like node than Django, that's a problem for a beginner, as they'd rather spend time learning fundamentals of web dev rather than learning a breaking change every month.
The upgrades have progressively become much less painful. Rails 1 to 2 was a big deal, as the change introduced the RESTful API. Rails 3 was slightly less violent, but the asset pipeline was still a pretty big deal. Rails 4 had a lot of API cleanup/simplification. A lot of meta programming magic was removed in favor of simpler code, which resulted in some rewriting of existing code.
Rails 5 was the easiest upgrade yet. ActionCable was the big change, but it’s an addition to Rails, not a replacement for some existing tech. As a result, it doesn’t disturb much existing code.
This has been my experience. I've been working on a large Rails project that started on Rails 4 beta in ~2014. Since then it's been incredibly stable and the upgrade paths have been quite simple from 4.0 to 4.1, then 4.2 and now we're on 5.0 with a few minor updates left before moving to 5.1.
Everything would have worked fine if we didn't keep things up to date, but just about everything in the software world needs to be continuously updated due to security updates/patches, etc.
In comparison to Python, where things just work for years without being touched. Upgrading can get you a better api or performance for a specific workload but rarely will it be necessary.