In brief, wifi routers need to generate unique crypto keys. The programmers started off using /dev/random, observed that blocking, and switched to /dev/urandom. The problem was that they were generating the keys at first boot, before any entropy had accumulated, with the result that many, many devices wound up with similar or identical keys.
The right answer is Linux's new getrandom syscall, whose default mode "blocks if the entropy pool has not yet been initialized", but thereafter acts like /dev/urandom. But on systems where that syscall is not available, I'm skeptical that using /dev/urandom is sane design.
To put it another way: /dev/urandom is effectively a service that can be in a "starting" or "started" state. When you, as a service, depend on another service, the idiomatic thing to do is to teach your init/supervisor daemon about that dependency. Then, instead of your service coming up and sitting around doing voodoo to your pipes/sockets to try to figure out whether your dependencies are up, your supervisor can just delay starting your service until it knows the dependent service is up.
(Or, to get really clever about it, you could just make /dev/urandom itself a socket attached to a "service" whose job is to, on startup, do a blocking-seed of the "real" /dev/urandom device, and thereafter become a dumb pipe over to the "real" /dev/urandom device. Then all your services just block on reading /dev/urandom [the socket] the first time, with the first one of them to try reading from it in fact causing the reseeding to happen. Socket activation!)