understand the argument for atomized storage

[vlcn-io]

understand the argument for atomized storage

[tantaman]

Given we can do automated migrations in most cases (https://aphrodite.sh/docs/migrations) atomized storage doesn't seem that useful.

Even if you go a datomic route, your application still has to have a schema (the shapes of records that functions expect). You're just hiding the problem of migrations and making it less explicit, giving an illusion of safety.

I.e., rather than catching inconsistencies near the data itself (table level) you're letting them trickle through your application and scattering != null, x?.foo everywhere with a few x.foo! bombs.

Other problems is that the datomic model is unfamiliar to developers and requires a new mental model to craft performant queries. Understanding when to join vs sub-select vs where exists vs left join, etc. is hard enough in a traditional sql model without then splitting it apart. Some background: datomic style approaches save their data in sql under the hood.