Episode 46 - Crud Forms, Hidden Coupling, and the Schema Migration Myth

Episode Summary

After a running joke about Germany’s 4-day internet outage, Martin and Adam dig into one of the most recurring questions they get: should a big, nasty crud form become one giant event or many? That leads straight into event versioning — why centralized upcaster functions are hidden coupling in disguise, and how Axon Framework v5’s dynamic converters validate the per-projection approach they’ve already been using. They close by tearing into a LinkedIn take that “event sourcing schema migration is an unsolved problem,” calling traditional migration tooling a stack of band-aids over a dysfunction nobody actually cured.

Main Discussion Points

• The Evergreen Crud Form Question — One of the most recurring questions they get: should submitting a large, multi-section form produce one giant event or several domain events?
• Design for UX, Not Domain Purity — Expanding a section, filling it out, and collapsing it is itself an event; you don’t need to be a domain expert, just think about the user’s cognitive load section by section.
• Wizards Work the Same Way — Multi-step wizards can be treated as automation performed by a human, tracked by the same kind of to-do-list read model a saga would use.
• Avoid the Other Extreme Too — A single command exploding into fifteen events is just as much an anti-pattern as one 300-field mega-event.
• Events Capture a Decision, Not All the Data — Large blobs, documents, or images should be stored elsewhere with the event holding a reference back to them, git-style.
• Giant Events Are Coupling Through the Back Door — When dozens of slices depend on the same sprawling event, adding one field turns into a breaking change for everyone.
• Upcaster Functions Are Hidden Coupling — A single global function deciding how every old event version gets transformed is a centralized decision that violates their own “don’t repeat yourself” rhetoric.
• Per-Projection Converters, Credit to Yordis — Each read model should decide for itself how to interpret an old event version instead of relying on one shared upcaster.
• Axon v5 Validates the Pattern — Axon Framework’s new dynamic converters attach per-event, letting each projection request whichever version it needs.
• The “Solved Problem” Pushback — A LinkedIn argument claimed event sourcing schema migration kills adoption because migrations are unsolved; Martin and Adam call traditional migration tooling a stack of band-aids, not a cure.

Design for the User’s Cognitive Load, Not Domain Purity

“The true spirit of this complex form is going through it a section at a time… don’t try to be a domain expert, be a UX expert as to what the human needs to attain when they’re looking at a wall of blank fields.” - Adam

Neither One Mega-Event Nor Fifteen Tiny Ones

“What I also don’t like is you have this one command and then you have 15 events coming out of that. That’s also not a good pattern. So it fully depends on what you’re doing.” - Martin

Upcasters Are a Global Decision Disguised as Infrastructure

“Upcaster functions… that is a form of coupling because the way I want to translate the absence of a date of birth actually is on the responsibility of the individual read models or projections that I have. Huge shout out to Yordis for pushing us on our own rhetoric about don’t repeat yourself.” - Martin / Adam

Schema Migration Is Not a Solved Problem — It’s a Thousand Band-Aids

“It’s a solved problem with a trap. It’s a solved problem with like a thousand band-aids. If you ever saw a soccer ball that’s been kicked around for 10 years in a family and it has duct tape basically replaced the leather — band-aid over band-aid is what they’re talking about.” - Adam

Key Takeaways

1. Big form submissions should mirror UX sections, not become one giant event — treat each expand/fill/collapse step, or wizard step, as its own event.
2. Don’t over-correct into fifteen events per command either — one command exploding into a pile of granular events is its own anti-pattern.
3. Events capture a decision, not necessarily all the data behind it — large blobs and documents belong in dedicated storage, referenced from the event.
4. Sprawling multi-field events create coupling through the back door — when many slices depend on one giant event, adding a field becomes a breaking change for everyone.
5. Upcaster functions centralize a decision that isn’t yours to make globally — each projection should decide for itself how to interpret an old event version.
6. Axon Framework v5’s dynamic converters validate the per-projection approach — converters attach per event so each projection pulls whichever version it needs.
7. Traditional schema migration tooling is a stack of band-aids, not a cure — Flyway- and Liquibase-style tools manage the dysfunction; they never remove it.
8. Event-sourced systems make rollbacks and shifting requirements dramatically cheaper — you’re not stuck reconciling new-schema data against an old schema after a bad deploy.