Now that the userbase is so large, we need some persistence for hydrated user data. Badger is a possible candidate, or just moving to Redis. Both support TTLs. We'd probably still need some caching for hydration errors etc.
https://github.com/dgraph-io/badger