Here is the trace I get from the command line: [Stage 4:================> (60 + 60) / 200]15/12/07 18:59:40 WARN YarnSchedulerBackend$YarnSchedulerEndpoint: ApplicationMaster has disassociated: 10.0.0.138:33822 15/12/07 18:59:40 WARN YarnSchedulerBackend$YarnSchedulerEndpoint: ApplicationMaster has disassociated: 10.0.0.138:33822 15/12/07 18:59:40 WARN ReliableDeliverySupervisor: Association with remote system [akka.tcp://sparkYarnAM@10.0.0.138:33822] has failed, address is now gated for [5000] ms. Reason: [Disassociated] 15/12/07 18:59:41 WARN ReliableDeliverySupervisor: Association with remote system [akka.tcp://sparkExecutor@ip-10-0-0-138.ec2.internal:54951] has failed, address is now gated for [5000] ms. Reason: [Disassociated] 15/12/07 18:59:41 ERROR YarnScheduler: Lost executor 3 on ip-10-0-0-138.ec2.internal: remote Rpc client disassociated 15/12/07 18:59:41 WARN TaskSetManager: Lost task 62.0 in stage 4.0 (TID 2003, ip-10-0-0-138.ec2.internal): ExecutorLostFailure (executor 3 lost) 15/12/07 18:59:41 WARN TaskSetManager: Lost task 65.0 in stage 4.0 (TID 2006, ip-10-0-0-138.ec2.internal): ExecutorLostFailure (executor 3 lost) … …
On Dec 7, 2015, at 1:33 PM, Cramblit, Ross (Reuters News) <ross.cramb...@thomsonreuters.com<mailto:ross.cramb...@thomsonreuters.com>> wrote: I have looked through the logs and do not see any WARNING or ERRORs - the executors just seem to stop logging. I am running Spark 1.5.2 on YARN. On Dec 7, 2015, at 1:20 PM, Ted Yu <yuzhih...@gmail.com<mailto:yuzhih...@gmail.com>> wrote: bq. complete a shuffle stage due to lost executors Have you taken a look at the log for the lost executor(s) ? Which release of Spark are you using ? Cheers On Mon, Dec 7, 2015 at 10:12 AM, <ross.cramb...@thomsonreuters.com<mailto:ross.cramb...@thomsonreuters.com>> wrote: I have pyspark app loading a large-ish (100GB) dataframe from JSON files and it turns out there are a number of duplicate JSON objects in the source data. I am trying to find the best way to remove these duplicates before using the dataframe. With both df.dropDuplicates() and df.sqlContext.sql(‘’’SELECT DISTINCT *…’’’) the application is not able to complete a shuffle stage due to lost executors. Is there a more efficient way to remove these duplicate rows? If not, what settings can I tweak to help this succeed? I have tried both increasing and decreasing the number of default shuffle partitions (to 100 and 500, respectively) but neither changes the behavior. --------------------------------------------------------------------- To unsubscribe, e-mail: user-unsubscr...@spark.apache.org<mailto:user-unsubscr...@spark.apache.org> For additional commands, e-mail: user-h...@spark.apache.org<mailto:user-h...@spark.apache.org>