Hi there... The old post closed, but I found an answer, so I will repost. The question was:
----------- I have recently been noticing some WCF services (hosted in IIS 6) crashing out on Windows Server 2003 R2 SP2 (with KB 913384 applied). I have run IISState, WinDBG, etc on IIS, but nothing so far has helped come up with the answer. The event log gives the following three errors: 1. (System Log) .NET Runtime version 2.0.50727.63 - Fatal Execution Engine Error (7A05F093) (80131506) 2. (Application Log) A process serving application pool X suffered a fatal communication error with the World Wide Web Publishing Service. The process id was N. The data field contains the error number. (0000: 8007006d) 3. (Application Log) A process serving application pool X terminated unexpectedly. The process id was N. The process exit code was '0x80131506'. The http error logs contain several entries similar to: "Connection_Abandoned_By_AppPool X" There is nothing in the http logs for each service that looks out of the ordinary (each wcf service is in its own app pool). Basically, it appears as if .Net is crashing out in system libraries (going by the stack traces that I produced), and at the moment, no one has a fix...! The stack traces don't show any thrown exceptions, stack overflows, access violations, etc, so I am not too sure how to go about diagnosing this error now. Any help appreciated! ----------- The error turned out to be within the DataContractSerializer and the NetDataContractSerializer. The .Net runtime (mscorwks.dll) was crashing out when invoking a dynamically generated serializer to implement the WriteObject() method. This is causing the runtime to exit, rather than throwing a catchable exception (System.ExecutionEngineException). Installing .Net 3.5 has fixed this problem (tested on several servers). The installer for .Net 3.5 pushes out service pack 1 for 2.0 and service pack 1 for 3.0. The service pack for .Net 2 replaces the mscorwks.dll (from 2.50727.62 to 2.50727.1442), which appears to fix the bug. It also appears that responses are coming back slightly faster now too.
