Challenges of Multi-Tenant, Enterprise ASP.NET Applications
I’ve been doing ASP.NET for a while, mostly at my current employer where we make large-scale online banking web sites. During that, what I’ve noticed more and more as new features come out for ASP.NET is that there’s a heavy focus on rapid application development - drag, drop, and ship - and less around the idea of creating a commercial application using ASP.NET. There are a few products like Community Server and DotNetNuke out there, but not many (or not as many as there could be) and I’d wager a lot of it has to do with lack of framework support for that sort of app.
To put what I’m talking about in context, let me first describe at a high level the kind of application I’m working on. Customers might want to have us host the application for them or they might want to host it themselves, so it needs to be something fairly easy to deploy. In a hosted environment, the customers want to be able to change their settings easily, so there’s a sort of “configuration user interface” that has to be put in place. Changes might include not only application settings, but text that appears on the various pages, so localization comes into play. It needs to be easily upgraded, deployed, and managed, so you don’t want a full copy of the application out there for every customer; you want a single copy with different IIS apps pointing at it… but that means the application has to support multi-tenancy (you can’t just stick all your config in web.config because there’s only one, right?). In a custom deployment, the application will be taken by a team, put into the customer’s environment, and programmatically customized, which means it needs to have a lot of extensibility points.
So, with that context, here are the big challenges.
Everything in the .NET framework assumes there’s only one tenant running on the application. When you ask for a configuration value, the value comes from The One Configuration Source and that’s that. There’s no qualifier in there anywhere to say “I want this configuration value for this specific tenant.” You have to write that. If you want a default value for all tenants on the app and the ability for individual tenants to override the value, you have to write that. Same thing with localization - you can’t say “I want this string for this tenant.” There’s one big bucket of resources and that’s it.
The lack of multi-tenancy support is pervasive and means a lot of work for the development group that wants to have a multi-tenant app. That’s unfortunate, particularly in light of the “Software as a Service” push that was going on just a couple of years back. What ever came of that?
Where do resources get stored? In compiled assemblies, right? What if I need some text on a page changed at runtime but for security reasons I don’t want to be recompiling assemblies and deploying them on the fly? (Multi-tenancy really hurts here since you can’t have a different set of resource assemblies per tenant.) There’s no out-of-the-box alternative to storing localized resources. You want to store things in SQL Server? You get to write that. Want the out-of-the-box stuff (like the ASP.NET localization expressions) to work with it? You get to write the factories and providers for that, too.
The whole ASP.NET theming thing is broken. Not “broken” in that it doesn’t work, but “broken” in that there are actually two different ways to theme things - skins and master pages. And they sort of work together, but when you define a single “style” for your pages, you have to manually track that “Style X means Skin Y and Master Page Z.” That’s crap.
Don’t forget each tenant wants their own theme, too.
There are a lot of things that you might want to configure in an app. Unfortunately, the place that stuff gets stored by default is in an application configuration file. In the filesystem. You want to give someone an interface to configure things, you either have to create a configuration service that stores things in a database and make your interface (and your app) talk to that proprietary service OR you have to allow your interface to somehow update web.config on the fly. In some cases, you can’t escape web.config - for example, if someone enables/disables a feature that means you need to register/unregister an HttpModule, you can’t do that because you can only register modules through web.config.
Oh, and throw in that multi-tenancy thing, too.
ASP.NET apps basically aren’t extendable at the page level. You can’t “derive and override” markup. If you want to interject your own logic, you have roughly three choices:
- Put code blocks inside the markup.
- Override the page class and change the markup to inherit from your custom page.
- Try to anticipate what people might want to extend and allow plugins through inversion of control, Microsoft Extensibility Framework, or some similar approach.
None of those are terribly great. Options one and two have you changing the ASPX markup, which makes it impossible to track what has been customized on that application instance (and is difficult to manage on a per-tenant basis) and option three quickly leads to YAGNI as you try to make everything infinitely extensible.
This actually has a direct impact on…
Deployment and Upgradeability
So, you put together your web app installer, run the MSI, and it puts a bunch of markup and config in the filesystem and some assemblies in the “bin” folder. Six months later, an implementation team has customized this thing using the “extensibility points” you’ve provided above, and they need to upgrade the base application.
Which markup files did they change? What config settings did they change in web.config? It becomes a tedious task of manually merging markup and config. (This is something that users of Subtext and other blogging engines are familiar with, too.)
Could you track checksums on the markup files and compare whether they’ve changed or not? You could… but you’d have to track every checksum for every file for every version ever released because someone might skip upgrading from 1.1 to 1.2 and go directly from 1.1 to 1.3.
Could you compile the pages? Sure, but that not only affects your extensibility (see above) but still requires those markup placeholder pages.
Pages aren’t the only things out there in the filesystem, though. Don’t forget your skins, master pages, and other markup files. In some cases, you can’t even move the locations.
Things in the filesystem that aren’t binary end up being problematic from a deployment and upgrade standpoint. You can address some of this with a custom VirtualPathProvider that serves things from embedded resources, but there are still some things you can’t hide behind a VPP - web.config, for example, and skins.
Why Isn’t This Stuff Addressed?
There are a lot of challenges with making large scale, multi-tenant applications. The above items aren’t an exhaustive list, but they’re some of the more obvious issues. Why hasn’t this been addressed in the framework? Is the majority case really the IT guy dragging a couple of grids and a DataSource onto a page and publishing the app right from Visual Studio? Or is it a case of self-fulfilling prophecy, where the features aren’t there so people don’t make these apps… and because people aren’t making these apps, the features aren’t considered important so they aren’t there?