工作单元位于领域层。
 
ABP的数据库连接和事务处理:
1,仓储类
ASP.NET Boilerplate opens a database connection and begins a transaction when entering a repository method.
当一个仓储方法开始执行时,ABP打开一个数据库连接并启用一个事务。
当仓储方法结束时,事务会被自动提交,数据库连接也会自动释放。
If the repository method throws any exception, transaction is rolled back and the connection is disposed. 
一旦仓储方法抛出任何异常,事务就会回滚,连接也会被释放。
In this way, a repository method is atomic (a unit of work). 

一个仓储方法就是一个工作单元。

If a repository method calls another repository method (in general, if a unit of work method calls another unit of work method), both uses same connection & transaction. The first entering method manages connection & transaction, others use it.
假如仓储方法调用另一个仓储方法,它们使用的是同一个连接,同一个事务。第一个被调用到的仓储方法负责管理连接和事务,另一个仓储方法则单使用不管理。
 
2,应用服务类
publicclass PersonAppService : IPersonAppService
{
privatereadonly IPersonRepository _personRepository;
privatereadonly IStatisticsRepository _statisticsRepository; publicPersonAppService(IPersonRepository personRepository, IStatisticsRepository statisticsRepository)
{
_personRepository = personRepository;
_statisticsRepository = statisticsRepository;
} publicvoidCreatePerson(CreatePersonInput input)
{
var person = new Person { Name = input.Name, EmailAddress = input.EmailAddress };
_personRepository.Insert(person);
_statisticsRepository.IncrementPeopleCount();
}
}

在应用层里的service类中,一个服务方法就是一个工作单元。

原理跟仓储方法是一模一样的。
 
3,工作单元(直接操作Unit of work)
一个做法是使用特性:
[UnitOfWork]
publicvoidCreatePerson(CreatePersonInput input)
{
var person = new Person { Name = input.Name, EmailAddress = input.EmailAddress };
_personRepository.Insert(person);
_statisticsRepository.IncrementPeopleCount();
}

上面的CreatePerson方法使用了[UnitOfWork]特性标签。

Thus, CreatePerson methods becomes unit of work and manages database connection and transaction, both repositories use same unit of work. Note that no need to UnitOfWork attribute if this is an application service method. See 'unit of work method restrictions' section.
CreatePerson方法称为工作单元,它负责管理数据库连接和事务,_personRepository和_statisticsRepository共用同一个工作单元。
要注意的是,如果这是一个应用服务方法,则不需要加UnitOfWork特性标签了。后文将会提及UnitOfWork特性应该用在什么地方。
 
另一个做法是使用 IUnitOfWorkManager.Begin(...)方法:
publicclass MyService
{
privatereadonly IUnitOfWorkManager _unitOfWorkManager;
privatereadonly IPersonRepository _personRepository;
privatereadonly IStatisticsRepository _statisticsRepository; publicMyService(IUnitOfWorkManager unitOfWorkManager, IPersonRepository personRepository, IStatisticsRepository statisticsRepository)
{
_unitOfWorkManager = unitOfWorkManager;
_personRepository = personRepository;
_statisticsRepository = statisticsRepository;
} publicvoidCreatePerson(CreatePersonInput input)
{
var person = new Person { Name = input.Name, EmailAddress = input.EmailAddress }; using (var unitOfWork = _unitOfWorkManager.Begin())
{
_personRepository.Insert(person);
_statisticsRepository.IncrementPeopleCount(); unitOfWork.Complete();
}
}
}

You can inject and use IUnitOfWorkManager as shown here. Thus, you can create more limited scope unit of works. In this approach, you should call Complete method manually. If you don't call, transaction is rolled back and changes are not saved. Begin method has overloads to set unit of work options.

It's better and shorter to use UnitOfWork attribute if you don't have a good reason.

在上面代码的Myservice构造函数中IUnitOfWorkManager会被注入。使用using关键字,可以在一个限定的代码域内使用工作单元。在代码域的最后必须以Complete方法结束。如果不调用这个方法,事务就会回滚。

如无很好的理由,最好还是使用UnitOfWork特性标签,而不是使用_unitOfWorkManager.Begin()
 
更多工作单元细节
 
禁用工作单元
[UnitOfWork(IsDisabled = true)]
publicvirtualvoidRemoveFriendship(RemoveFriendshipInput input)
{
_friendshipRepository.Delete(input.Id);
}

无事务的工作单元

[UnitOfWork(false)]
public GetTasksOutput GetTasks(GetTasksInput input)
{
var tasks = _taskRepository.GetAllWithPeople(input.AssignedPersonId, input.State);
returnnew GetTasksOutput
{
Tasks = Mapper.Map<List<TaskDto>>(tasks)
};
}

这是特殊情况下才使用的东西,可能是因为高并发,希望不要因为使用事务锁住了某些数据表和数据列。

 
一个工作单元调用另一个工作单元
If a unit of work method (a method declared with UnitOfWork attribute) calls another unit of work method, they share same connection and transaction. First method manages connection, others use it. This true for methods run in same Thread (or in same request for web applications). Actually, when a unit of work scope begins, all codes executing in same thread shares same connection and transaction until the unit of work scope ends. This is true both for UnitOfWork attribute and UnitOfWorkScope class. If you create a seperated Thread/Task, it uses own unit of work.
同上面仓储方法里的描述,会共用同一个连接和事务。但是如果两个工作单元方法使用不同的线程,则不会共用。
 
自动保存
[UnitOfWork]
publicvoidUpdateName(UpdateNameInput input)
{
var person = _personRepository.Get(input.PersonId);
person.Name = input.NewName;
}
O/RM framework keep track of all changes of entities in a unit of work and reflects changes to the database.

如上面代码,我们不需要在代码里写_personRepository.Update方法,因为ORM框架会自动追踪变化并反映到数据库中。

 
IRepository.GetAll() 方法
[UnitOfWork]
public SearchPeopleOutput SearchPeople(SearchPeopleInput input)
{
//Get IQueryable<Person>var query = _personRepository.GetAll(); //Add some filters if selectedif (!string.IsNullOrEmpty(input.SearchedName))
{
query = query.Where(person => person.Name.StartsWith(input.SearchedName));
} if (input.IsActive.HasValue)
{
query = query.Where(person => person.IsActive == input.IsActive.Value);
} //Get paged result listvar people = query.Skip(input.SkipCount).Take(input.MaxResultCount).ToList(); returnnew SearchPeopleOutput { People = Mapper.Map<List<PersonDto>>(people) };
}
When you call GetAll() out of a repository method, there must be an open database connection since it returns IQueryable. This is needed because of deferred execution of IQueryable. It does not perform database query unless you call ToList() method or use the IQueryable in a foreach loop (or somehow access to queried items). So, when you call ToList() method, database connection must be alive.

在仓储方法、应用服务方法外面,如果使用到IRepository.GetAll()之类涉及到数据库读写的方法,必须加上UnitOfWork特性标签。

 
工作单元特性标签的限制
  • All public or public virtual methods for classes those are used over interface (Like an application service used over service interface).
  • All public virtual methods for self injected classes (Like MVC Controllers and Web API Controllers).
  • All protected virtual methods.
UnitOfWork特性标签可以用在上面列举的方法上。
建议把方法都写成virtual的。
UnitOfWork特性不能用于private的方法,因为
ASP.NET Boilerplate uses dynamic proxying for that and private methods can not be seen by derived classes. UnitOfWork attribute (and any proxying) does not work if you don't use dependency injection and instantiate the class yourself.
 
选项
publicclass SimpleTaskSystemCoreModule : AbpModule
{
publicoverridevoidPreInitialize()
{
Configuration.UnitOfWork.IsolationLevel = IsolationLevel.ReadCommitted;
Configuration.UnitOfWork.Timeout = TimeSpan.FromMinutes(30);
} //...other module methods
}

SaveChanges

sometimes, you may want to save changes to database in middle of a unit of work operation. In this case, you can inject IUnitOfWorkManager and call IUnitOfWorkManager.Current.SaveChanges() method. An example usage may be saving changes to get Id of a new inserted Entity in EntityFramework. Note that: if current unit of work is transactional, all changes in the transaction are rolled back if an exception occurs, even saved changes.
 
有时候我们想在工作单元还没结束时就保存变化,那么这时就要使用IUnitOfWorkManager.Current.SaveChanges() 方法。
但是要注意的是,即便执行了SaveChanges,最后若有异常抛出,事务发生回滚,SaveChanges也会回滚。
 
事件
三种事件:Completed,Failed ,Disposed
publicvoidCreateTask(CreateTaskInput input)
{
var task = new Task { Description = input.Description }; if (input.AssignedPersonId.HasValue)
{
task.AssignedPersonId = input.AssignedPersonId.Value; _unitOfWorkManager.Current.Completed += (sender, args) => { /* TODO: Send email to assigned person */ };
} _taskRepository.Insert(task);
}
 
 
05-04 06:31