[BenoitP]

I did something akin to that with the following monstrosity of DevOps:

- set up Kafka

- set up Flink

- CDC on your DB, into Kafka

- have your server maintain a websocket to clients, read Kafka from it, and dispatch messages according to a field from Kafka

- put your reaction logic into Flink, insert into the Kafka topic read by the server

I even went to opening JIRA tickets in Apache Flink, so that you may choose the Kafa topic to insert to dynamically. In case you have multiple servers to dispatch messages to.

It was very clunky, and a pain to operate. However it was short-lived and is no longer a concern of mine. What I got away from that: adding a time to data serving is just multiplying the complexity of data bugs you can have.