Let me start with an explanation of our bespoke setup (we use spring boot 3.x):
We have an aggregate in a backend service. The aggregate receives commands via a websocket connected to a web app frontend. We have mechanisms in place to ensure that an aggregate only receives commands from a single websocket at a time.
The advantages of this setup: 1) We get low latency by using the websocket
2) If the websocket is disconnected, e.g., due to a container crash, the
frontend can connect to another backend instance where the aggregate is
When the frontend reconnects, the aggregate is reconstituted from a list of
events we’ve stored in a standard postgres database. Further commands are
directed to this reconstituted instance. In this manner, reconstitution is done
once per websocket and not for each command.
Currently, we use an in-house implementation. I’m wondering if we
could use the Axon framework instead. My idea is that we could do some hybrid between the Event sourcing repository and an ‘in-memory’ repository.
- All domain events on the aggregate will be stored in the event store.
- On the creation of the aggregate, we store it in the ‘in-memory’ repository.
- On reconstitution, we let the event sourcing repository reconstitute the aggregate and store it in the ‘in-memory’ repository.
- When we receive commands, we fetch the aggregate from the ‘in-memory’ repository.
- We will not need the Axon server component.
Of course, there’s also a need to introduce command handlers and so on, but my main headache is how to do the reconstitutions, when I don’t want to do them for each and every command.
I’ll be grateful for advice on this approach. I’m in doubt whether I need special aggregate factories, so please include pointers for this if applicable.