We have some communication patterns where it would be desirable not to wait for a send timeout if there is no connected peer -- instead we would prefer to fast fail so that the application can do something else (like print an error message and exit, or return an HTTP error over another socket, or something).

These situations are a little more brittle, since a temporary disconnect can cause a hard failure, but for our use cases at RackTop, it's better to just fail right away.