A while ago I observed an interesting problem due to a routing misconfiguration. An asymmetric route was introduced across two discrete switches/routing instances. The route was then imported in a separate location.
When I began troubleshooting I was under the impression that the importer preferred one route over another because of admin distance or some other metric, or perhaps it was some sort of BGP esoterica you’d only encounter if you were using VRFs and MP-BGP to learn routes. However, the truth was probably something more prosaic.
Both routes were static routes. This meant their learning mechanism across importers was the same. The reason the importer preferred one route over another was probably due to one entering the routing table before the other.