I am wondering, what is a good approach to handle time-series data in an Akka Actor application?
My current approach does not involve actors. Time-series entries/values are represented as simple numeric or case-class instances. Those instances are both written, or read from the database (Cassandra) and then processed further using Akka Streams.
I am wondering if there is a more Akka dogmatic way to handle time-series data (that makes sense). For example, each time-series belongs, in DDD terms, to an aggregate. This aggregate entity could be represented as an Actor. For example, it could have the current time-series data as state, then to be queried by consumers instead of the database, and passivated while not in use.
I don’t think the latter approach is sensible, because you’d have to replay all time-series data to recreate the actor (after passivation) even if you only need a subset and, as far as I know, you’d have to give up on low-level primitives in the database, and use the more space- and CPU intensive serialized versions of events when reading from the database. As far as I understand is, this is one of the reasons why CQRS is used in the the first place.
Maybe we don’t have to use Actors for everything, and I don’t see much wrong with the database approach. Still, in hope that I might learn something new, I’m curious if I could improve my approach above reading and writing everything directly to the database. Maybe there is a nice pattern that could improve this kind of thing.