The newest version of cmapi change an exception type to ConnectionRefusedError from socket.timeout. Essentially triggering workernode/ loadbrm to reload every 10-20 seconds.
During this time, on large S3 installations with slow networking, DBRMs, specially the extent map might not be downloaded/processed in time so the map can be truncated at run time share memory.

The solution is to avoid ConnectionRefusedError error during startup polling to cause this restart/reloading.

/usr/share/columnstore/cmapi/cmapi_server/managers/process.py : 173

173                 except (socket.timeout,ConnectionRefusedError) :

    174                     logging.debug(

    175                         f'"{name}" {workernodes[name]["IPAddr"]}:'

    176                         f'{workernodes[name]["Port"]} not started yet.'

    177                     )

    178                 else:

    179                     # delete started workernode from workernodes dict