Thanks to all the comments and feedback from the last post I think I have a better understanding now of the benefits of CQRS (separate from the benefits of Event Sourcing). I’m going to try and sum it up here, and point out some areas where I could still use some advice:
CQRS Benefits
Sounds like the primary benefit of CQRS as an architecture is it allows you to create a simpler domain model by sucking out everything related to queries. I can definitely see the benefit to this, in general the domain logic related to commands is the high-value behavior in the software, but the logic required to service the queries would add a lot of low-value “noise” to the domain model that would dilute the high-value (command) behavior – sorting, paging, filtering, pre-fetch paths, etc. Also the most appropriate domain structure for implementing commands might not be the most optimal for implementing queries. To paraphrase Greg, this usually results in a domain model that is mediocre at both, piss-poor at one, or more likely piss-poor at both commands and queries.
Not only will you be able to simplify your domain model by pulling out all the query logic, but at least a handful of commands in most systems will probably be “pass-though” type commands with little to no logic that just generate events. If these can be implemented directly in the command-handler and never touch the domain model, this allows you to slim down the domain model even more.
Also, if you were to do event sourcing without CQRS, you no longer have a database containing the current state (only the domain model would) which makes it difficult (or impossible) to support ad-hoc querying and/or reporting that is common in most business software.
Of course CQRS provides some great scalability benefits, not only scalability but I have to assume that it provides extremely low latency for most operations, especially if you have an asynchronous event bus.
I know Greg says that you get a 3x scaling (Commands, Queries, Client) of your ability to perform parallel development, but IMHO, it seems like it only provides 1.5x scaling since even without CQRS you’re going to have your client loosely coupled to your domain - which is still a great benefit to be able to realize.
Questions / Concerns
If all the queries against an aggregate get pulled out to the Query layer, what if the only commands for that aggregate can be handled in a “pass-through” manner with the command handler directly generating events. Is it possible to have an aggregate that isn’t modeled in the domain model? Are there any issues or downsides to this?
I know in the feedback from my previous posts it was suggested that having one domain model handling both commands and queries requires implementing a lot of traversals between objects that wouldn’t be necessary if it was only servicing commands. My question is, do you include traversals in your domain model based on the needs of the code, or based on the conceptual domain model? If none of my Commands require a Customer.Orders traversal, but the conceptual domain includes the concept of a set of orders belonging to a customer – should I model that in my domain model or not?
I like the idea of using the Query side of the architecture as a place to put junior devs where the risk of them screwing something up has minimal impact. But I’m not sold on the idea that you can actually outsource it. Like I said in one of my comments on my previous post, the code to handle a query and generate DTO’s is going to be dead simple, but the code to process events and apply them to the tables on the query side is going to require a significant amount of domain knowledge to know which events to listen for to update each of the de-normalized tables (and what changes need to be made when each event is processed). I don’t know about everybody else, but having Indian/Russian/whatever outsourced developers have to do anything that requires significant domain knowledge has never been successful in my experience. And if you need to spec out for each new query which events to listen to and what to do with each one, well that’s probably going to be just as much work to document as it would be to just implement it.
Greg made the point in a comment that doing an aggregate query like “Total Sales By Customer” is going to be inefficient if you use event sourcing but not CQRS. I don’t understand why that would be the case. I imagine in that case you’d simply have a method/property on the Customer object that calculated total sales for that customer by enumerating over the Orders collection. Then the application services layer would generate DTO’s off of the Customers collection that included say the CustomerID, CustomerName, TotalSales, or whatever the case may be. As long as you use a snapshotting implementation, I don’t see why that would be anymore inefficient in a DDD+Event Sourcing implementation than in a typical DDD implementation.
Like I mentioned in my last post I still have some questions about query logic that haven’t been answered yet, but before I start asking those I want to make sure I have a strong grasp on what benefits CQRS provides. My main concern with the query logic was that I know I could just toss it all into the query side, but I was concerned that I would be losing the benefits of using CQRS in the first place if I did that. I want to elaborate more on this though with some example situations in an upcoming post.
