ERROR: gpfdist server closed connection (seg30 slice1 x.x.x.x:xxxxx pid=5896)(Hint The root cause is likely to be an overload of the ETL host or a temporary network glitch between the database and the ETL host causing the connection between the gpfdist and database to disconnect.;
2021-04-29 13:31:00 20132 INFO [0:821823:54:97] session already ended. return empty response (OK) 2021-04-29 13:31:00 20132 INFO [0:821823:54:97] HTTP EMPTY: ::ffff:x.x.x.x GET 2021-04-29 13:31:00 20132 INFO [0:821823:54:97] sent EOF: succeed 2021-04-29 13:31:00 20132 INFO [0:821823:54:97] request end 2021-04-29 13:31:00 20132 INFO [0:821823:54:97] detach segment request from session 2021-04-29 13:31:00 20132 INFO [0:821823:54:97] SSL cleanup and free 2021-04-29 13:31:00 20132 INFO [0:821823:54:97] successfully shutdown socket 2021-04-29 13:31:00 20132 WARN [0:821823:54:97] gpfdist read unexpected data after shutdown 2021-04-29 13:31:00 20132 INFO [0:821823:54:97] unsent bytes: 0 (-1 means not supported) 2021-04-29 13:31:00 20132 INFO [0:821823:54:97] successfully closed socket
This is due to network slowness between whichever server is running gpfdist and the Greenplum cluster causing timeouts that lead to premature EOFs eventually killing the gpfdist command.
Adding the '-w' switch into the gpfdist command with an acceptable value, such as 'gpfdist -w 60' will add an extra 60 second wait time to the gpfdist command before it closes each session.
This value is discussed more in the Utility Guide linked below: