Fix #_write_to_socket in connection

By placing the NBIO_READ_EXCEPTIONS above the NBIO_WRITE_EXCEPTIONS, an
Errno::EWOULDBLOCK or Errno::EAGAIN would trigger a wait for the socket
to be readable. This is causing timeout errors in our testing.

When calling #write_nonblock, the appropriate action is to wait for the
socket to be writable, not readable. Ensuring that we handle the write
exceptions before the read exceptions accomplishes this.

Author: quixoten

lib/redis/connection/ruby.rb

@@ -104,14 +104,14 @@ def _write_to_socket(data)
         begin
           write_nonblock(data)
 
-        rescue *NBIO_READ_EXCEPTIONS
-          if IO.select([self], nil, nil, @write_timeout)
+        rescue *NBIO_WRITE_EXCEPTIONS
+          if IO.select(nil, [self], nil, @write_timeout)
             retry
           else
             raise Redis::TimeoutError
           end
-        rescue *NBIO_WRITE_EXCEPTIONS
-          if IO.select(nil, [self], nil, @write_timeout)
+        rescue *NBIO_READ_EXCEPTIONS
+          if IO.select([self], nil, nil, @write_timeout)
             retry
           else
             raise Redis::TimeoutError