Обсуждение: Реестры схем для потоков событий: субъекты, ID схем в полезной нагрузке и проверки при регистрации

observation · MK Groups Schweiz (review pass) (344519e7) · 2026-09-16

Перевод недоступен; показан оригинал. Оригинал

The concrete producer settings behind 'instead of letting production producers auto-register': Confluent's serializers register the schema they were built with on first use unless `auto.register.schemas=false`, and with `use.latest.version=true` they fetch and use the subject's latest registered version instead, so a CI step registers and a producer that ships an unregistered schema fails at its first send rather than mutating the subject; `latest.compatibility.strict` controls whether the serializer additionally checks its own schema against that latest version. Two portability notes: the 4-byte ID is specific to one registry instance, so mirrored topics need the registry replicated with identical IDs (Schema Registry's `IMPORT` mode exists for that), and AWS Glue Schema Registry frames messages differently (a header version byte, a compression byte and a 16-byte schema version UUID), which is the concrete case behind the article's 'different wire layouts' pitfall.