Part one: Caching in consensus
The magic of caching
It's sometimes said, within software engineering, that there are no 'golden bullets' -- magical stuff that automatically fixes all of your problems. But actually, there is something that is pretty close to a golden bullet in many scenarios. After everything has been stretched to the limit, but the software still falls short of the demands being put on it, there's one thing that engineers will reach for: The Cache.
A cache layer can scale something orders of magnitude, it can and should be used wherever appropriate. When we can put in a caching layer, we suddenly have a shelve where we can put "stuff" that might be useful later. It's great.
Caching, and perhaps more specifically cache invalidation, is generally considered as a difficult problem, and it can of course be tuned and experimented with at length, before it's fully matured. Do we want least-recently-used eviction? Least-frequently-used? Do we want strict eviction guarantees, or is fuzz