Skip to main content

Finalization, Know What it Costs

This is a post about object finalization in .NET.
Finalization is not as inexpensive as we think, it increases the pressure put on GC.All objects that need finalization are moved into a finalizable queue and the actual finalization happens in a separate thread. Because the objects full state may be needed, the object itself and all the object it points to are promoted to the next generation (this is needed so that GC does not clean these objects off in the current round), and these objects are cleaned up only after the following GC.
Due to this reason, resources that need to be released should be wrapped in as small a finalizable object as possible, for instance if your class needs a reference to an unmanaged resource, then you should wrap the unmanaged resource in a separate finalizable class and make that a member of your class and furthermore the parent should be a non-finalizable
class. This approach will assure that only the wrapped class (class that contains the unmanaged resource) will be promoted.And as mentioned earlier the finalization occurs in a separate thread and furthermore there is only one finalization thread hence if a finalizer causes this thread to block, all the other objects needing finalization in the queue will not be called and your application will
leak !!!.

Therefore, when implementing finalizers, keep it as simple as possible.A good practice to follow when the lifetime of an object is explicitly know, is to implement the dispose pattern, by implementing the IDisposable interface and calling it explicitly; or use the using statment to wrap you created instance, so that once the scope of the object is no longer valid the dispose method of the object is called.
A time may arise when you may at some places know the lifetime of the object but at some point you may not; in this case put your clean up code in a separate function, call that function in the finalizer and also in your dispose method, additioanlly call the GC.SupressFinalization() in your dispose method after the clean up method; this will inform GC not to run the finalizer.

Comments

  1. A good one related to your post:

    http://www.cs.inf.ethz.ch/ssw/files/GC_in_NET.pdf

    Thanks.

    - Prabath

    ReplyDelete

Post a Comment

Popular posts from this blog

Hosting WCF services on IIS or Windows Services?

There came one of those questions from the client whether to use II7 hosting or windows service hosting for WCF services. I tried recollecting a few points and thought of writing it down.
WCF applications can be hosted in 2 main ways- In a Windows service- On IIS 7 and aboveWhen WCF was first released, IIS 6 did not support hosting WCF applications that support Non-HTTP communication like Net.TCP or Net.MSMQ and developers had to rely on hosting these services on Windows Services.With the release of IIS 7, it was possible to deploy these Non-Http based applications also on IIS 7. Following are the benefits of using IIS 7 to host WCF applications
Less development effort
Hosting on Windows service, mandates the creating of a Windows service installer project on windows service and writing code to instantiate the service, whereas the service could just be hosted on IIS by creating an application on IIS, no further development is needed, just the service implementation is n…

Task based Asynchronous pattern, Async & Await and .NET 4.5

One of the key features in .Net 4.5 is to write asynchronous programs much easier. So if I was to write asynchronous programs in .Net 2.0/3.5, I would either follow the event based model or the callback based model. For an example, a synchronous method that does intensive work (say the DoWork()) can be made asynchronous by using the following patterns
1) Implementing the IAsyncResult pattern. in this implementation, 2 methods are exposed for the DoWork() synchronous method, the BeginDoWork() and the EndDoWork() method. The user will call the BeingDoWork() passing in the required parameters and a callback of the delegate type AsyncCallback(IAsyncResult). The BeginDoWork() will spawn a new thread a return control back to the user. Once work is completed in the spawned method, as a last step, it will call the inform the AsyncResult implementation, which in turns will call the EndDoWork() (which is the callback that was passed in to the BeginDoWork()).
2) Implementing the event pattern. Her…

MEF (Managed Extensibility Framework), .NET 4, Dependency Injection and Plug-in Development

Almost after .Net 4 was released I remember reading about MEF (Managed Extensibility Framework), this was a framework for developers to compose their application with required dependencies. At first this looks like the Unity Container used for dependency injection, but MEF is much more than a dependency container but there is nothing stopping you from using MEF as a dependency injector.I remember around 5 years back when I was in a project that created a framework that allows developers to plug-in there modules as WinForm screens. The developer would create a set of screens with the intended functionalities and the drop this component in the bin folder of the framework and then will move on to do some painful configurations for the framework to pick up the module. Dropping the component into the bin folder and doing a bit of configuration is all that s needed for the framework to pick up display the screens. Typically, the configurations would also contain metadata about the screen.Al…