I've had a couple eMMC go bad on single board computers. I initially only noticed the off-by-1 read errors when the radio data that was being recorded by the computers (8kB/s for 2+ years) as comma separated values started sometimes getting commas turned into the next character up in ascii which broke my visualization python script.

It's definitely something you learn the first time you use eMMC (or any low end storage tech).

what volume of writes per day, in MB, should you be expecting on a top of rack switch, though?

does it have some logging features you can't turn off? using as an example a cisco nexus 3064 I have sitting around here, all of the logging is done to an off-system destination over the network.

I can think of very few legitimate or sound network engineering practices that would require 1U, mostly layer-2 feature set switches to be writing anything more than a few dozen KB per day to their internal storage.

I can't say that I'm surprised to hear of something failing in a new and unique way in an edgecore switch, since the company (Accton) has also released a real dumpster fire of a series of outdoor radios in their Ignitenet brand, which have a spectacular failure rate.