Some of you sometimes think why your application eat up until 1 GB-2GB memory on production server. After looking through the code, doing some profiling, maybe shaking your head a bit, you've figured out what the issue is and now you need to give some feedback.
In this tutorial, I will show some tips that you can follow to reduce your memory usage on production server and keep your ASP.NET MVC codebase working as you’d expect.
1. Understand the queries in your problem domain
The root cause of the support ticket I received was a simple case of fetching too much data from the database, causing obscene amounts of memory usage.
It's a common enough issue. You're building a simple blog, it has posts and it has media (images, videos, attachments). You put a Media array onto your Post domain object. Your Media domain object has all the image data stored in a byte array. Since you're using an ORM, there's a certain way you need to design your domain model to play nice; we've all experienced this.
public class BlogPost {
public ICollection<BlogMedia> Media { get; set; }
}
public class BlogMedia {
public byte[] Data { get; set; }
public string Name { get; set; }
}
There's nothing absolutely wrong with this design. You've modeled your domain accurately. The problem is, when you issue a query through your favorite ORM, it eagerly loads all the data associated with your blog post:
public IList<BlogPost> GetNewestPosts(int take) {
return _db.BlogPosts.OrderByDescending(p => p.PostDate).Take(take).ToList();
}
A seemingly innocuous line (unless you've been bitten), a sneaky monster is lying in wait with big consequences if you haven't disabled lazy loading or didn't tell your ORM to ignore that big Data property on blog media.
It's important to understand how your ORM queries and maps objects and make sure you only query what you need (for example using projection).
public IList<PostSummary> GetNewestPosts(int take) {
return _db.BlogPosts.OrderByDescending(p => p.PostDate).Take(take).Select(p => new PostSummary() {
Title = p.Title,
Id = p.Id
}).ToList();
}
This ensures we only grab the amount of data we really need for the task.
It's OK to have more than 5 methods on a repository; be as granular as you need to be for your UI.
2. Don't call your repositories from your views
Consider this line in an MVC view:
@foreach(var post in Model.RelatedPosts) {
...
}
It seems innocent enough. But if we take a look at what exactly that model property is hiding:
public class MyViewModel {
public IList<BlogPost> RelatedPosts {
get { return new BlogRepository().GetRelatedPosts(this.Tags); }
}
}
Your "view model" has business logic in it on top of calling a data access method directly. Now you've introduced data access code somewhere it doesn't belong and hidden it inside a property. Move that into the controller so you can wrangle it in and populate the view model conciously.
This is a good opportunity to point out that implementing proper unit tests would uncover issues like this; because you definitely can't intercept calls to something like that and then you'd realize injecting a repository into a view model is probably not something you want to be doing.
3. Use partials and child actions to your advantage
If you need to perform business logic in a view, that should be a sign you need to revisit your view model and logic. I don't think it's advisable to do this in your MVC Razor view:
@{
var blogController = new BlogController();
}
<ul>
@foreach(var tag in blogController.GetTagsForPost(p.Id)) {
<li>@tag.Name</li>
}
</ul>
Putting business logic in the view is a no-no, but on top of that you're creating acontroller! Move that into your action method and use that view model you made for what it's intended for. You can also move that logic into a separate action method that only gets called inside views so you can cache it separately if needed.
//In the controller:
[ChildActionOnly]
[OutputCache(Duration=2000)]
public ActionResult TagsForPost(int postId) {
return View();
}
//In the view:
@{Html.RenderAction("TagsForPost", new { postId = p.Id });}
Notice the ChildActionOnly attribute. From MSDN:
Any method that is marked with ChildActionOnlyAttribute can be called only with the Action or RenderAction HTML extension methods.
This means people can't see your child action by manipulating the URL (if you're using the default route).
Partial views and child actions are useful tools in the MVC arsenal; use them to your advantage!
4. Cache what matters
Given the code smells above, what do you think will happen if you only cached your view model?
public ActionResult Index() {
var homepageViewModel = HttpContext.Current.Cache["homepageModel"] as HomepageViewModel;
if (homepageViewModel == null) {
homepageViewModel = new HomepageViewModel();
homepageViewModel.RecentPosts = _blogRepository.GetNewestPosts(5);
HttpContext.Current.Cache.Add("homepageModel", homepageViewModel, ...);
}
return View(homepageViewModel);
}
Nothing! There will not be any performance gain because you're accessing the data layer through a controller variable in the view and through a property in the view model... caching the view model won't help anything.
Instead, consider caching the output of the MVC action instead:
[OutputCache(Duration=2000)]
public ActionResult Index() {
var homepageViewModel = new HomepageViewModel();
homepageViewModel.RecentPosts = _blogRepository.GetNewestPosts(5);
return View(homepageViewModel);
}
Notice the handy OutputCache attribute. MVC supports ASP.NET Output Caching; use it to your advantage when it applies. If you are going to cache the model, your model needs to essentially be a POCO with automatic (and read-only) properties... not something that calls other repository methods.
Conclusion
I hope with tutorial above, it will help you to minimize your memory usage on the server.