Reputation: 773
I have a WCF duplex service with a 2 way method called GetList, plus a one way callback method ItemUpdated.
I've detected a deadlock in the following situation:
I could set the ConcurrencyMode of my service to Reentrant, but that's not what I need. What I need is to make sure that my client is not processing any callback from the service before calling GetList.
So I can synclock an object to make sure that this doesn't happen. Is that my better alternative?
Just to clarify what my current design is, my service is actually a windows service that does some processing on a list of objects on a schedule and when each item is updated it raises the event ItemUpdated. This service has a WCF service facade that allows one or many consoles (clients) to subscribe to its events and see what is going on in the service.
Upvotes: 0
Views: 693
Reputation: 773
The problem was not really on the service but on the callback handler in my client. This is basically what was happening:
So I ended up changing the ConcurrencyMode in the callback using the CallbackBehavior attribute. The service remains with ConcurrencyMode = Single.
Upvotes: 0
Reputation: 5801
InstanceContextMode = Single together with ConcurrencyMode = Single causes WCF to serialize all the calls to your single instance. In other words WCF has created a lock to enforce serialization. This is the cause of the deadlock. Adding yet another lock might work. But you are proposing a client lock... the client shouldn't have to change its threading behavior to keep the service from deadlocking. The service needs to be more robust than that.
This is a service issue. You should create a more robust service.
I would suggest changing the service to remove the serialization and prevent the possibility of deadlock entirely. My suggestion (in order that I would try):
Change the service's InstanceContextMode to PerCall. The "preferred" design is to have the service be stateless. Create a different service instance for each call removes the forced serialization. Store all your state information in a common store (for example a database or cache).
Have the service make the ItemUpdated callback on a different thread. This allows the service thread to complete before the callback is done. The problem with this solution is that if the ItemUpdated callback throws an exception the service is no longer waiting to catch and handle it. This means good error handling is harder.
See Discover Mighty Instance Management Techniques For Developing WCF Apps for more information.
Upvotes: 2