CDH 5.11 and Impala 2.8 is pretty old. You should try the latest versions.   In the logs, there is one failed fragment instance and one timeout fragment instance. The failed one is on host hadoop-slave22.use1.data.ripple.com. You should check impalad logs on it for more details. The timeout one has instance_id=d6447c0a5ed591c4:47ac77680000001b. The first part (d6447c0a5ed591c4) is the same for this query. The last part (47ac77680000001b) is the id of the fragment instance. You can check previous logs to see where this instance is scheduled. Then check impalad logs of the scheduled host. Usually this might due to network saturation. Later Impala versions have more RPC improvements, e.g. KRPC. ... View more