Reacting on data changes and publishing those changes as GraphQL events with subscriptions can be hard, especially in a multi-service environment with multiple databases or when scaling your GraphQL server with multiple instances. GraphQL clients shouldn't miss events or receive them twice, no matter how your backend architecture looks like or what trouble (service goes down, database connection lost, ...) they might have when serving a subscription request.In this talk, I will show you, how Debezium and Apache Kafka can help you building reliable subscriptions from changes in your database. Debezium is a change data capture (CDC) tool that can forward changes from a database' transaction log in to the Kafka message broker.In my talk I will use a GraphQL backend implement in Java with "Spring for GraphQL", but as Debezium and Kafka are not tied to java the idea is usable also with other GraphQL frameworks and programming languages. You do not need to have knowledge of Java or Spring for GraphQL" to understand the talk.