The UCP endpoint.
The address, prefixed with `ucx://` to use.
Whether to deserialize data in :meth:`distributed.protocol.loads`
The read-write cycle uses the following pattern:
Each msg is serialized into a number of "data" frames. We prepend these
real frames with two additional frames
1. is_gpus: Boolean indicator for whether the frame should be
received into GPU memory. Packed in '?' format. Unpack with
``<n_frames>?`` format.
2. frame_size : Unsigned int describing the size of frame (in bytes)
to receive. Packed in 'Q' format, so a length-0 frame is equivalent
to an unsized frame. Unpacked with ``<n_frames>Q``.
The expected read cycle is
1. Read the frame describing if connection is closing and number of frames
2. Read the frame describing whether each data frame is gpu-bound
3. Read the frame describing whether each data frame is sized
4. Read all the data frames.This method overrides distributed.comm.core.Comm.local_address.
This method overrides distributed.comm.core.Comm.peer_address.