I think it depends a bit on which in part of the ecosystem you work. I do not do web applications at all, but use Rust for machine learning. Things like ndarray, petgraph, and the Tensorflow bindings have been very stable for me.

The only large change that I had to make over the last year due to ecosystem changes was going from error-chain to failure. Of course, this was not strictly necessary, but failure seems to be more popular now and I generally prefer it over error-chain.