.NET on Pivotal Cloud Foundry

In my latest post, I tested Lattice.cf, the single VM smaller brother of Pivotal Cloud Foundry (PCF). Considering a full installation of PCF has a footprint of about 25 Virtual Machines (VM) requiring a total of over 33Gb RAM, 500+ Gb storage, and some serious CPU power, it’s not hard to see why Lattice is more developer friendly. However, that wasn’t my real motivation to try it out: more important was Lattice’s incorporation of the new elastic runtime architecture codename ‘Diego‘ which will replace the current Droplet Execution Agent (DEA) based runtime in due time.

For me, the main reasons to get excited about Diego are two-fold:

  • Diego can run Docker containers: I demoed this in my latest post
  • Diego can run Windows (.NET) based workloads

In this post, I’ll demo the Windows based workloads by deploying an ASP.NET MVC application, which uses the full-fledged production ready .NET stack and requires a Windows kernel – as opposed to .NET Core which is cross-platform, runs on a plentitude of OSes, and is very exciting, but not production ready yet.

Diego on PCF

At this point we have to resort to PCF as Lattice can’t run Windows workloads. This is because Lattice’s strong point (all services in 1 VM) is also its weakness: since Lattice incorporates all required services in a single Linux VM it quite obviously loses the ability to schedule Windows workloads.

Let’s take a quick look at the Diego architecture overview:

Diego architecture overview

Diego architecture overview

Diego consists of a ‘brain’ and a number of ‘cells’. These cells run a container backend implementing Garden – a platform-neutral API for containerization (and the Go version of Warden). The default backend – garden-linux – is Linux based and compatible with various container technologies found on Linux (including Docker).

As soon as we run the various services in the overview on seperate VMs (as we do on PCF), it becomes possible to provision a Windows cell. ‘All’ we need is a Windows backend for garden and the various supporting services, and we should be good to go. Preferably in the form of a convenient installer.

One problem remains: we still need the Diego runtime in Pivotal Cloud Foundry. Kaushik Bhattacharya at Pivotal supplied me with an internal beta of ‘Diego for PCF’ (version 0.2) which I could install in my Pivotal on VMware vSphere lab environment. Installed, the Ops Manager tile looks like this:

Diego for PCF tile

Diego for PCF tile

And the default VMs and resources that come with Diego for PCF (notice the single default cell which is a garden-linux cell):

Diego for PCF default resources

Diego for PCF default resources

garden-windows cell

To provision a Windows cell, we have to create a new VM manually and install a Windows server OS, and run a setup powershell script as well as the installer with the correct configuration. I’ll skip the details here, but when all is done, you can hit the receptor service url (in my case https://receptor.system.cf.lab.local/v1/cells), and it should show 2 cells: 1 installed as part of Diego for PCF, as well as the one we just created:

        "cell_id": "DiegoWinCell",
        "zone": "d48224618511a8ac44df",
        "capacity": {
            "memory_mb": 16383,
            "disk_mb": 40607,
            "containers": 256
        "cell_id": "cell-partition-d48224618511a8ac44df-0",
        "zone": "d48224618511a8ac44df",
        "capacity": {
            "memory_mb": 16049,
            "disk_mb": 16325,
            "containers": 256

Intermezzo: Containerization on Windows?

There must be some readers who are intrigued by the garden-windows implementation by now. After all, since when does Windows have support for containers? In fact, Microsoft has announced container support in the next Server OS, and the Windows Server 2016 Technical Preview 3 with container support was just released. However, this is not the ‘containerization’ used in the current version of garden-windows.

So how does it work?

Some of you may know Hostable Web Core: an API by which you can load a copy of IIS inside your process. So what happens when you push an app to Cloud Foundry and select the windows stack, is that app is hosted inside a new process on the cell, in which it gets its own copy of IIS after which it’s started.

I know what you’re thinking by now: “but that’s not containerization at all”. Indeed, strictly speaking it isn’t: it doesn’t use things like cgroups and namespaces used by Linux (and future Windows) container technologies in order to guarantee container isolation. However, from the perspective of containers as ‘shipping vehicles for code’ it’s very much containerization, as long as you understand the security implications.

Deploying to the Windows cell

Deployment to the Windows cell isn’t harder than to default cells, however, there are a couple of things to keep in mind:

  • as the windows stack isn’t the default, you have to specify it explicitly
  • as for now the DEA mode of running tasks is still default, you have to enable Diego support explicitly

Using the Diego Beta CLI the commands to push a full .NET stack demo MVC app are as follows (assuming you cloned it from github):

cf push diegoMVC -m 2g -s windows2012R2 -b https://github.com/ryandotsmith/null-buildpack.git --no-start -p ./
cf enable-diego diegoMVC
cf start diegoMVC

After pushing and scaling the Pivotal Apps Manager:

Pivotal CF Apps Manager with DiegoMVC .NET  app

Pivotal CF Apps Manager with DiegoMVC .NET app

And the app itself:

DiegoMVC .NET application on Windows cell on Pivotal CF

DiegoMVC .NET application on Windows cell on Pivotal CF


Diego for PCF is still in internal beta, but soon Pivotal Cloud Foundry will have support for running applications using the full .NET stack.


AppPool crashes and RapidFail protection

Yesterday, one of our production sites began to crash at random intervals. We managed to narrow the issue down to one specific user logging in at the time, and clicking on a number of (again random) pages.

Post-mortem debugging using crashdumps and WinDbg showed the last exceptions on the stack to be (again random) and pretty minor.

The only thing they had in common was that they were unhandled, and so ended up in the Application_Error method of the Web project’s HttpApplication derived class.

So what happened ?

In the end it boils down to a feature in Internet Information Services called “Rapid Fail Protection”. If enabled (default), the application pool will stop and serve 503 Service unavailable responses when it sees X unhandled exceptions in Y minutes (both configurable).

rapid fail protection

Of course the best fix is to properly catch exceptions, however, if you ever have a case of Application Pools stopping under mysterious circumstances, check if you have Rapid Fail protection turned on.

Self-hosting ASP.NET MVC

Recently I was considering the technology to use for the GUI of a windows desktop client application I’m working on in my spare time. The standard picks are obviously WinForms or the more modern WPF. However, I have some problems with them in this case:

  • the technologies above are (windows) platform dependent, while for instance HTML5 isn’t.
  • the application I’m working on is a debugger. Perhaps details about that will follow in a future post, but I would like to be able to run it remotely from an arbitrary machine (and OS).
  • I really appreciate the power of modern javascript libraries such as knockout.js, and I can’t use those with WPF or WinForms.

Now, for Windows 8, the store applications can be developed in HTML5 already, while desktop apps can’t. Of course I could resort to creating a real web application (hosted in IIS), but that would require the debugger host machine to have IIS installed. The same is true for the rehostable web core.

There is an open source library – NancyFX – which can be self-hosted, and I was tempted to use that but it had some unknowns coming from the MVC experience: controllers are modules in Nancy, and I couldn’t quite find what existing functionality in MVC was or wasn’t available in the Nancy library.

So with all other options out of the window, I set out to self-host ASP.NET MVC4.

Surprisingly, much of the stuff that’s needed to do this is undocumented: while the MVC pipeline is well known, the internals of what happens between the TCP port and the first entry point in ASP.NET aren’t. However, logic dictates:

  1. there should be an open TCP/IP port listening somewhere
  2. there should be a dedicated app domain for the web application
  3. somehow the request received on the port should be transferred to the web application http runtime

1. Receiving requests

The traditional way of listening for web requests was to simply open a socket on port 80 using the Windows Socket API (WinSocks), and run the HTTP protocol stack on top of this in your app in user mode. However, this had severe drawbacks, which led to the creation of a dedicated kernel HTTP protocol stack – http.sys – which does all the low level HTTP work, and lies directly on top of the TCP/IP protocol stack.

In the user mode space, applications communicate with http.sys through the Windows HTTP Server API. IIS uses this API, and if we want to self-host ASP.NET, we will have to as well. Fortunately, the .NET framework includes a wrapper class around this API: System.Net.HttpListener

Using HttpListener to process requests should then be easy. I started by implementing a listener on top of HttpListener, and using Reactive Extensions to push the incoming request context to clients:

internal class Listener : IDisposable
    private HttpListener _listener = null;
    public IObservable IncomingRequests { get; private set; }

    internal Listener(int port)
        _listener = new HttpListener();
        _listener.Prefixes.Add(string.Format("http://localhost:{0}/", port));
        _listener.AuthenticationSchemes = AuthenticationSchemes.Anonymous;

        IncomingRequests = _listener.GetContextsAsObservable().ObserveOn(NewThreadScheduler.Default);

    public void Dispose()
            if (_listener == null) return;
            _listener = null;
        catch (ObjectDisposedException)

internal static class ListenerExtensions
    private static IEnumerable Listen(this HttpListener listener)
        while (true)
            yield return listener.GetContextAsync().ToObservable();
    internal static IObservable GetContextsAsObservable(this HttpListener listener)
        return listener.Listen().Concat();

Now all client code has to do is create a new listener, and subscribe to the requestcontext stream:

private Listener _listener;

private void SetupListener(int port)
    _log.InfoFormat("Setting up new httplistener on port {0}", port);
    _listener = new Listener(port);

    _log.InfoFormat("Start forwarding incoming requests to ASP.NET pipeline");
    (c) =>
        catch (Exception ex)
            _log.Error("Exception processing request", ex);
    (ex) => _log.Error("Exception in request sequence", ex),
    () => _log.Info("HttpListener completed"));
    _log.Info("Completed httplistener setup");

The real magic happens in the yet to be implemented ProcessRequest method. However, before feeding requests to the ASP.NET pipeline, we first have to set that up in its own AppDomain. When hosting your project in IIS, your application is typically bound to a dedicated application pool (AppDomain). In theory the web application could be run in the already running default app domain, however, in this post I’ll try to stay as close as possible to the IIS approach.

System.Web contains the class ApplicationHost which allows you to do exactly what we want: create a new AppDomain, and specify a user supplied type (class) which should be instanced there to bootstrap the domain.

Again, in theory, you could use any class, but there are two things to keep in mind:

  • you want to communicate with the new AppDomain from your default AppDomain
  • you don’t want this bootstrapping instance to be garbage collected as it’s the root object in the new domain

The classic solution for the first issue is to use a class derived from MarshalByRefObject, as this will automagically enable remoting by RPC between your AppDomains. Another – more modern – option would be to use WCF, but I didn’t check if that works in practice yet.

The second issue is fixed by explicitly telling the remoting infrastructure the object’s lease is never expiring (which ties its actual lifetime to the lifetime of the new AppDomain).

The code below demonstrates this:

public class AppHost : MarshalByRefObject
    //factory method
    private static AppHost GetHost(string virtualDir, string physicalPath)
        // Fix for strange CreateApplicationHost behavior (can only resolve assembly when in GAC or bin folder)
        if (!(physicalPath.EndsWith("\"))) physicalPath += "\";

        // Copy this hosting DLL into the /bin directory of the application
        var fileName = Assembly.GetExecutingAssembly().Location;
            var folderName = string.Format("{0}bin\", physicalPath);

            //create folder if it doesn't exist
            if (!Directory.Exists(folderName)) Directory.CreateDirectory(folderName);

            //copy file
            File.Copy(fileName, Path.Combine(folderName, Path.GetFileName(fileName)), true);

            //and all assemblies
            var pathToAppHost = Path.GetDirectoryName(fileName);
            foreach (var fn in Directory.EnumerateFiles(pathToAppHost, "*.dll", SearchOption.TopDirectoryOnly))
                File.Copy(fn, Path.Combine(folderName, Path.GetFileName(fn)), true);
        catch { }

        return (AppHost)ApplicationHost.CreateApplicationHost(typeof(AppHost), virtualDir, physicalPath);

    //set an infinite lease
    public override object InitializeLifetimeService()
        return null;

The code before the actual CreateApplicationHost call in the factory method requires an explanation: while the CreateApplicationHost is a convenient method, it’s hardwired to only search for assemblies in the bin folder relative to the physical path of the web project (or the GAC). Rick Strahl mentions this on his blog, and in fact if you check the framework reference sources, or inspect the assemblies with ILSPY you’ll discover a hardwired bin reference.

So for a quick fix, in the code above I just copy the relevant assemblies to that folder. A more elegant solution would be to do a bit more work and create a new AppDomain using AppDomain.CreateDomain(…) and tell it where to find assemblies yourself, or even better, override the AssemblyResolve eventhandler.

Next we want to create an instance of your web/MVC project’s main class, which is some class – in global.asax.cs – derived from HttpApplication, so we also need a method on AppHost which does this:

private HttpApplication _mvcApp;

private void HostMvcApp() where T: HttpApplication
    //usually IIS does this, but we need to serve static files ourselves

    _mvcApp = Activator.CreateInstance();

This will bootstrap your MVC application and call into it’s Application_Start method where you can register routes, areas, bundles, etc as usual.

Since the ASP.NET pipeline does not serve static files, we need to do this ourselves. I do this here by registering a StaticFileHandlerModule, which we’ll implement in a future post.

So now we have hosted the MVC application and set up a HTTP listener. The only thing that’s left is to connect these two together so the ASP.NET pipeline will handle these requests, an undertaking which turns out to be more complex than it sounds.

The way in which requests should be handed to the HttpRuntime is through a call to the static ProcessRequest method:

public static void ProcessRequest(HttpWorkerRequest wr)

The problem is: HttpWorkerRequest is abstract and has over 70 methods that should be implemented to be able to pass every possible request to the ASP.NET pipeline (the HttpRuntime will call into those methods to find details about the request).

So whatever host calls into the HttpRuntime will have to provide its own implementation of HttpWorkerRequest for wrapping its http requests.

We can check those who did by searching derived types:


All except the SimpleWorkerRequest are IIS specific implementations, which tightly integrate with the proprietary and native code IIS engine. The SimpleWorkerRequest itself is an implementation which when used to wrap HTTP requests allows you to send simple requests into the ASP.NET pipeline. However, that’s about it: the concept of streams, certificates (security), completely misses, so it’s not much more than a proof of concept: it won’t enable you to unleash the full power of the ASP.NET engine.

So we’re out of luck, we have to make our own implementation. An (incomplete) example can be found in an old msdn magazine.

Wrapping the request and sending it into the pipeline then looks like this:

private void ProcessRequest(HttpListenerContext context)
   _log.DebugFormat("Processing request");
   var wr = new HttpListenerWorkerRequest(context, VPath, PPath);
   HttpContext.Current = new HttpContext(wr);

   _log.DebugFormat("Finished processing request");

But it fails at runtime…

It turns out we run into another problem: since the request is received in the default AppDomain, and processed in the ASP.NET AppDomain, we have to pass either the HttpListenerContext – like we did above – or our implementation of the HttpWorkerRequest between AppDomains. No matter how you do this (MarshalByRefObject, WCF), it requires serialization. And guess what ? None of these is easily serializable, for one because they contain the request and response streams…

At this point I decided it would save me a lot of work if I moved the Listener we created into the ASP.NET AppDomain as well, and do away with the serialization mess.

So ultimately, I ended up with a factory method which:

  1. creates a new appdomain and hosts my implementation of HttpApplication
  2. sets up a HttpListener in this new AppDomain
  3. wraps the HttpListenerContext in a custom implementation of HttpWorkerRequest
  4. sends it into the HttpRuntime

And finally…it works…I managed to have a fully functioning MVC application hosted in-process with my main application.

There are some loose ends to tie up though, like serving static files. I may touch on that in a follow up, but if you want to try it yourself, try the hostasp NuGet package incorporating these concepts right now.

UPDATE: or try the source code


One (AppFabric) workflow with multiple persistence stores

For a new project, we decided to model and implement the business processes of our customers using Windows Workflow Foundation 4.

To do this we use a single workflow definition in the form of a workflow service host (.xamlx) in IIS, which depending on the type of process invokes different final actions. To persist our workflow instances and manage their lifecycle we use Windows Server AppFabric.

Initially, we started with the default configuration: a single persistence database separate from our customer databases, but this presented a problem: in rollback scenarios, only rolling back the database of one customer would cause the workflows and the customer’s other business data to get out of sync. One solution would be to also rollback the workflow database, however, in a scenario with multiple customers who all use the same persistence database this is no option either.

In the end, the best approach would be to have a customer’s business data and workflow persistence data in the same location, so both can be managed at the same time. However, AppFabric doesn’t have built-in support for the scenario in which we have a single WF service definition, which is exposed to all customers, which then use their own instance stores for persistence.

So we had to create support for this scenario ourselves and the schematic solution consisted of the following steps:

  • Create a factory for the xamlx service, which enables us to set the persistence store for a new workflow service host in code based on the customer.
  • Tell the factory which customer should be serviced for a request. The only way in which we can do this is by exposing the factory on a different url for every customer.
  • To expose the factory dynamically based on the customers in the database, we used a virtual path provider.

Technically what we did was:

1 – In Global.asax of the web project register add the virtual path provider:

protected override void OnApplicationStarted()
    HostingEnvironment.RegisterVirtualPathProvider(new MyPathProvider());

2 – The virtual path provider uses a simple parser – MyServiceUrlParser (not shown) – to find out which customer is connecting (unique id in the url), and if it ends on “svc” it returns a virtual file (MyServiceFile) with a service host definition. The first two overrides are necessary for serving the file, the last two for preventing IIS from throwing exceptions for not finding the folder or file in cache:

public class MyPathProvider : VirtualPathProvider
    public override bool FileExists(string virtualPath)
        return !string.IsNullOrWhiteSpace(MyServiceUrlParser.Parse(virtualPath).Service) || base.FileExists(virtualPath);

    public override VirtualFile GetFile(string virtualPath)
        var match = MyServiceUrlParser.Parse(virtualPath);

        if (match.ServiceType.EndsWith("svc")) return new MyServiceFile(virtualPath);
        if (match.ServiceType.EndsWith("xamlx")) return base.GetFile("~/Workflows/MyFlow.xamlx");

        return base.GetFile(virtualPath);

    public override System.Web.Caching.CacheDependency GetCacheDependency(string virtualPath, System.Collections.IEnumerable virtualPathDependencies, DateTime utcStart)
        return MyServiceUrlParser.IsValidServiceUrl(virtualPath) ? null : base.GetCacheDependency(virtualPath, virtualPathDependencies, utcStart);            

    public override bool DirectoryExists(string virtualDir)
        return MyServiceUrlParser.IsValidServiceUrl(virtualDir) || base.DirectoryExists(virtualDir);

3 – The virtual file (svc) provided at this customer dependent path, this will trigger the IIS to load the MyFlow.xamlx from the customer specific url. Since our virtualpath provider is setup to return MyFlow.xamlx fom any xamlx request, it will service the one workflow for all customers:

public class MyServiceFile : VirtualFile
    public MyServiceFile(string virtualPath) : base(virtualPath) { }

    //return the same factory and service for the same url...only the factory will specify another persistance db based on the url
    public override Stream Open()
        var serviceDef = new MemoryStream();
        var defWriter = new StreamWriter(serviceDef);

        // Write host definition

        serviceDef.Position = 0;
        return serviceDef;

4 – finally the factory for setting the persistence store:

public class DynamicHostFactory : WorkflowServiceHostFactory
    protected override WorkflowServiceHost CreateWorkflowServiceHost(WorkflowService service, Uri[] baseAddresses)
        var host = new WorkflowServiceHost(service, baseAddresses);

        host.DurableInstancingOptions.InstanceStore = SetupInstanceStore(baseAddresses);

        return host;

    private static SqlWorkflowInstanceStore SetupInstanceStore(Uri[] baseaddresses)
        //do stuff based on the url

And there we have it, a single xamlx workflow definition with a dedicated persistence database per customer.