Polling Duplex Service Failure in Out of Browser Silverlight Client on Mac

Question

We have a Silverlight client which runs "out-of-browser" on a Mac. This client consumes a WCF service through a polling duplex binding.

In the client I am listening to the "Faulted" event exposed by the "InnerChannel" property of the System.ServiceModel.DuplexClientBase derivative which represents the service in the client side.

After exactly one minute this "Faulted" event is triggered and after that the channel is not working anymore, i.e. when the server tries to send a message through the callback channel it gets a timeout exception.

Here is a theory I have: I suspect that the underlying polling operation in the client has a timeout of 1 minute. In the server side the serverPollTimeout property of the pollingDuplexHttpBinding section is set to more than one minute. This means that the server holds a poll request for more than one minute if it has nothing to tell to the client during that time. I suspected that this revealed to the timeout in the client poll message. To test my theory, I've reduced the serverPollTimeout setting to less than a minute and indeed – the problem is not shown.

In the client-side, there is the PollingDuplexBindingElement.ClientPollTimeout property which is according to this blog exactly the setting which should tell the client to wait for more than a minute. The default for this setting is 5 minutes and I've even set it explicitly – but the problem still remains (without the workaround as described above).

Please note that this problem happens only on Mac out of browser client.

To sum-up, here are my questions:

• How/where can I see a descriptive error message which tells exactly what is the problem here?
• Why does it happen only in Mac out of browser client?
• Can someone confirm my theory?
• If my theory is true – how can I really set the timeout for the polling request in the client?

Answer 1

After a long thread regarding this with Microsoft support here are the conclusions regarding this issue:

• The problem is relevant also for "regular" WCF calls. If we call a regular WCF operation from SL Mac OOB, it will timeout after exactly one minute, even-though the timeout is set to be higher
• Microsoft confirmed that 60 seconds is the default timeout on Mac OS and that the SL runtime doesn't call the required Mac OS API to set it to higher value even-though the SL client code uses the documented SL API for changing the timeout. They say it’s "by design" since they are not aware of any scenarios where the server might take that long even in a long poll