in this case, you can refer to the ideas of TCC (Try, Confirm, Cancel).

it is impossible for multiple data stores to control through a single transaction. To control it manually

you define what consistency is first. If that API has huge side effects, such as launching a nuclear bomb, success or failure is unlikely to roll back

implementing transaction solutions across mysql and redis must be very complex. final consistency scheme is recommended.

final consistency can be achieved by message queue decoupling. Basic idea:

@Transactional
doSomething{
    
      
}

then start another service to receive messages

{
    redis    
}

points to pay attention to:

1. Database write operations and send messages are transactional, and kafka has the impression that there are transactional messages.
2. sequence of messages: make database modifications to the same data to ensure that the order of the redis write operation service is also the original modification order.