All other evidence being equal this scenario suggests very possible disk level problem on the slow host. You could try brute force and restart the Impala Daemon on that host or do some light diagnostics, like hdparm or fio, depending on how much time you want to invest. host one - RowsReturned: 52,115 (52115) - RowsReturnedRate: 4868064 per second (4868064) host 2 - RowsReturned: 52,319 (52319) - RowsReturnedRate: 955 per second (955) If these obvious disk checks don't show up anything useful , send full query profile of a fast query and the slow.
... View more
