Steps to reproduce:

• build etcd and etcdctl from master or take latest release
• run etcd
• run ETCDCTL_API=3 etcdctl watch --prefix ""
• SIGSTOP etcd, e.g. using ctrl+z in terminal

Expected outcome:
After some timeout, the watch will fail. This would allow the caller to propagate an error further up the stack, telling clients to retry elsewhere where there might be a healthy connection.

Observed outcome:
The watch blocks forever.