Long time ago, I have learned that underestimating the architecture of a software project is much worse than over-engineering it. When the project evolves, there are always some new requirements which completely break some of the initial assumptions. Any robustness that can help with addressing larger changes can be very helpful and can save a lot of time and sleepless nights.
Even the tiniest, one-purpose scripts which generates some report can grow and grow, and suddenly it becomes a large system that runs a department of a global corporation. I have seen that so many times, and being lazy at the beginning of the project often means great issues in the future.
But you know, we have a deadline…
Years ago, I tried to design a shared infrastructure for projects we have been doing in my company. Our typical project was a web-based line of business application, with lots of CRUDs, but there were always a lot of custom complex workflows and integrations.
When you look at almost any demo of ASP.NET application with Entity Framework, you can see that they are accessing the DbContext right in the MVC controllers, and the EF entities are used even in the views (where they can do lazy loading and all sort of nasty things).
This approach works great for a simple demo, and it is actually great for the beginners as it doesn’t get too complicated. But it is a nightmare in any large-scale application, as we saw in many apps written by somebody else which we had to maintain later. It leads to a lot of performance issues (lazy loading everywhere), repetitive code snippets (copy pasting, find & replace programming) and inconsistencies caused by the fact you can access any table from any page in the application.
We have always strictly separated “data access layer” which contained Entity Framework stuff, a “busines layer” with a domain model that was mapped from the EF entities, and a “presentation layer” on the top.
The business layer was always the most fascinating one for me – there are always a lot of things inside, and everyone is doing it quite differently.
Repositories and Query Objects
For some people, repositories are anti-pattern. For some people, Entity Framework DbSets are your repositories and there is no need to implement another useless layer on top of them.
However, I have found this layer very useful, so we are used to implement a repository over DbSet (or any other database table or collection), for a dozen of reasons. This extra layer gives us a great flexibility when we need to implement features like soft delete or entity versioning, and helps greatly in integration tests as it allows nice mocking.
Unlike a lot of people, we don’t have the all-purpose GetAll() method that would return IQueryable in the repos. Our repositories can only Insert, Update, Delete and GetById. There are no methods which would return multiple records (other than returning a few entities by ID).
To query for multiple records, we use Query Objects. They are great to encapsulate complex queries and ensure that the queries are not repeated across the entire application.
If you allow to access IQueryable, you will soon have expressions like productsRepo.GetAll().Where(p => p.CategoryId == category) spawned everywhere in the project. If you want to add soft delete checks later, it is very difficult to find all usages of this table.
Thanks to extracting this logic in a query object, we can have a class named ProductsByCategoryQuery in the project, which can be used on all places in the application. If we hit into any performance problems with Entity Framework (which happens in case of complex queries), we can re-implement the query object to use raw SQL while keeping the rest of the application unchanged, and so on.
Also, the query objects give us an unified way to do sorting and paging, which helps with building generic CRUD mechanisms.
Facades and Services
The business logic of the application is implemented in facades and service classes. I use the term “facade” for classes which are used directly from the presentation layer – facades expose all functions the UI needs.
All other classes with some business logic are referred as services, and they can be grouped in quite a complex hierarchies, depend on each other etc.
These facades and services use repos and queries for implementing everything the business layer does.
Model
In our case, the domain model of the application is typically a bunch of DTO (data transfer objects). Unlike most of the books says, I like the anemic model approach much more. I find very impractical to inject dependencies into domain objects.
Because most of the domain model objects are very similar to EF entities, we have adopted AutoMapper. It is a really great tool and I like its ProjectTo() method that can translate the mappings to IQueryable.
There are some caveats however. You should avoid nesting DTOs to prevent the queries to be super-complicated. We have also found out that it is a very good idea to have multiple DTOs for every entity. We have list DTOs which are used in grids and lists, we have detail DTOs for the edit forms, we have basic DTOs for filling the comboboxes and lot more. Thanks to AutoMapper, most of our mappings are just CreateMap<TEntity, TDTO>() with no additional configuration.
…and it’s Open Source
I have been showing the way how a business layer can be designed in many of my conference and user group talks. Our Riganti Utils Infrastructure library is developed on GitHub and anyone is welcome to use it, or to take inspiration from it.
The current version of the infrastructure is 2 and currently, we are in process of designing its third version.
In the next part of this article, I will try to address the issues with v2 and how to make the infrastructure easier to use.