Swallowing the CommClosedError when the stream handler closes may remove relevant exception information

[hendrikmakait]

In

distributed/distributed/core.py

Lines 1027 to 1033 in 5708bdf

	except CommClosedError:
	closed = True
	logger.info(
	"Connection to %s has been closed.",
	comm.peer_address,
	)
	break

we always swallow the CommClosedError without even logging it. This can be problematic because we use this error both for the happy and the sad path of stream handler closing. For example, in convert_stream_closed_error (

distributed/distributed/comm/tcp.py

Lines 130 to 143 in 5708bdf

	def convert_stream_closed_error(obj, exc):
	"""
	Re-raise StreamClosedError as CommClosedError.
	"""
	if exc.real_error is not None:
	# The stream was closed because of an underlying OS error
	exc = exc.real_error
	if isinstance(exc, ssl.SSLError):
	if exc.reason and "UNKNOWN_CA" in exc.reason:
	raise FatalCommClosedError(f"in {obj}: {exc.__class__.__name__}: {exc}")
	raise CommClosedError(f"in {obj}: {exc.__class__.__name__}: {exc}") from exc
	else:
	raise CommClosedError(f"in {obj}: {exc}") from exc

), we may attach relevant exception information to the CommClosedError but will subsequently ignore it in handle_stream.

Ideally, we could differentiate between when stream handler closing is expected and when it's unexpected and log more information in the latter case.