The order team renames a field in an event. The change looks harmless: the producer compiles and the business concept seems the same. But analytics and notifications receive a shape they no longer understand. The sending team may not notice the break until someone asks why the reports stopped moving.
Events remove direct calls, not contracts. Schema Registry helps manage message structures and versions. It is useful infrastructure, but a message that passes schema validation can still carry the wrong business meaning.
What a registry actually records
A registry stores schemas, versions and compatibility rules. Producers and consumers can check which fields exist, which are optional, and whether an application can read a message written with another version.
Compatibility has direction. Sometimes a new reader must handle old data; sometimes an old reader must survive new data. If events are retained for replay, both questions may matter. Teams need to choose the policy that matches the lifecycle of their messages.
The type is unchanged; the meaning is not
Imagine that status remains a string. Previously, “confirmed” meant an order was accepted. Now the producer uses it only after payment clears. A schema check may pass unchanged while a warehouse acts at the wrong moment.
Important events therefore need stable semantics as well as compatible fields: which business fact occurred, what the publisher guarantees and what consumers are allowed to infer from it.
Changing the contract without a big-bang migration
A safer change often adds a new optional field first, lets consumers adopt it, and removes the older form only after migration. A genuinely breaking semantic change may deserve a new event version or a separate event type.
Someone must own that transition and know the affected consumers. Automated compatibility checks can reject invalid formats in CI, but they cannot negotiate deadlines with an external partner or determine whether a new business meaning is safe.
When the registry is worth operating
If a small number of services know all their consumers, contract tests and direct coordination may be enough. Another infrastructure dependency can add more cost than protection.
As independent teams and consumers multiply, central schema validation can catch many accidental breaking changes before deployment. It still does not prove that the right business process occurred after the message was delivered.
Schema Registry protects the structure of the agreement. People remain accountable for what the agreement means.