I am running Spark 1.5.2 on Yarn. My job consists of a number of SparkSQL transforms on a JSON data set that I load into a data frame. The data set is not large (~100GB) and most stages execute without any issues. However, some more complex stages tend to lose executors/nodes regularly. What would cause this to happen? The logs don’t give too much information -
15/11/19 15:53:43 ERROR YarnScheduler: Lost executor 2 on ip-10-0-0-136.ec2.internal: Yarn deallocated the executor 2 (container container_1445957755572_0176_01_000003) 15/11/19 15:53:43 WARN TaskSetManager: Lost task 142.0 in stage 33.0 (TID 8331, ip-10-0-0-136.ec2.internal): ExecutorLostFailure (executor 2 lost) 15/11/19 15:53:43 WARN TaskSetManager: Lost task 133.0 in stage 33.0 (TID 8322, ip-10-0-0-136.ec2.internal): ExecutorLostFailure (executor 2 lost) 15/11/19 15:53:43 WARN TaskSetManager: Lost task 79.0 in stage 33.0 (TID 8268, ip-10-0-0-136.ec2.internal): ExecutorLostFailure (executor 2 lost) 15/11/19 15:53:43 WARN TaskSetManager: Lost task 141.0 in stage 33.0 (TID 8330, ip-10-0-0-136.ec2.internal): ExecutorLostFailure (executor 2 lost) 15/11/19 15:53:43 WARN TaskSetManager: Lost task 123.0 in stage 33.0 (TID 8312, ip-10-0-0-136.ec2.internal): ExecutorLostFailure (executor 2 lost) 15/11/19 15:53:43 WARN TaskSetManager: Lost task 162.0 in stage 33.0 (TID 8351, ip-10-0-0-136.ec2.internal): ExecutorLostFailure (executor 2 lost) 15/11/19 15:53:43 WARN TaskSetManager: Lost task 153.0 in stage 33.0 (TID 8342, ip-10-0-0-136.ec2.internal): ExecutorLostFailure (executor 2 lost) 15/11/19 15:53:43 WARN TaskSetManager: Lost task 120.0 in stage 33.0 (TID 8309, ip-10-0-0-136.ec2.internal): ExecutorLostFailure (executor 2 lost) 15/11/19 15:53:43 WARN TaskSetManager: Lost task 149.0 in stage 33.0 (TID 8338, ip-10-0-0-136.ec2.internal): ExecutorLostFailure (executor 2 lost) 15/11/19 15:53:43 WARN TaskSetManager: Lost task 134.0 in stage 33.0 (TID 8323, ip-10-0-0-136.ec2.internal): ExecutorLostFailure (executor 2 lost) [Stage 33:===============================> (117 + 50) / 200]15/11/19 15:53:46 WARN ReliableDeliverySupervisor: Association with remote system [akka.tcp://sparkExecutor@ip-10-0-0-136.ec2.internal:60275] has failed, address is now gated for [5000] ms. Reason: [Disassociated] - Followed by a list of lost tasks on each executor.