I attach also my issue, which is similar, I think a solution is to override the _receive method of the gremlin-python connection.py module, simply trying a reconnection before the 'finally' statement which puts it back into the pool.

similar issue - connection pool not automatically repaired