Whenever a urlRetrieve() fails due to an unexpected error (ENOTFOUND, ETIMEDOUT, Socket hang up, etc.), the domain handler and the global exception handler would detect this and not crash the server, however the dirty internal state would somehow prevent future HTTP requests from completing successfully.
Removed domain usage since that feature is marked "unstable" and is rumored to be marked for deprecation in future versions of node. Using the "error" event of the request object itself, which means errors are local in scope and won't pollute global state. This should have been the solution originally, but when urlRetrieve() was written, I was not as familiar with node.
3689aafe3b