I'm just finishing up building my first deployed WCF service, and over the past couple of months I've learned a lot about WCF - what works and what doesn't from a practical standpoint.  Unfortunately there are still a few areas in which I'm fuzzy.

I was originally treating my WCF proxies like ASP.NET 2.0 web service proxies.  Create a proxy and then let the GC handle it.  This was really useful because I was just injecting the service instance into each of my MonoRail controllers using Windsor.  Testing was nice and easy, and things looked good.

Until last week... I discovered a nasty bug in my web application.  The bug was, that under load, the WCF service would stop responding and timeout.  Come to find out that it's not only recommended to close your client side proxies, but required (at least when using Net TCP binding).

This turned my beautiful code with its inverted dependencies into a mess.  I had to start asking the container on every call for a new proxy since it was getting closed after every remote call.

IMyService svc = IoC.Resolve<IMyService>();
svc.SomeRemoteCall();
svc.close();

I had to do it this way because the remote calls were wrapped in a local service helper object that didn't have any idea about the unit of work, i.e. where the best place to put the Close() call.

When I have some time this week I plan on building a WCF proxy helper that will create a proxy unit of work on a per HTTP request basis.  When the HTTP request comes in, the proxy is created, and when the HTTP request finishes the helper will call Close() on the proxy.  This should allow me to remove the ugly static IoC dependency and use Windsor to inject the proxy instance.