Hi, Few things for closer examination:
* is yarn master URL accepted in 1.3? I thought it was only in later releases. Since you're seeing the issue it seems it does work. * I've never seen specifying confs using a single string. Can you check in the Web ui they're applied? * what about this <custom> in class? Can you publish the application logs from the driver and executors? Jacek On 2 Jun 2016 10:33 a.m., "Prabeesh K." <prabsma...@gmail.com> wrote: > Hi I am using the below command to run a spark job and I get an error like > "Container preempted by scheduler" > > I am not sure if it's related to the wrong usage of Memory: > > nohup ~/spark1.3/bin/spark-submit \ --num-executors 50 \ --master yarn \ > --deploy-mode cluster \ --queue adhoc \ --driver-memory 18G \ > --executor-memory 12G \ --class main.ru.<custom>.bigdata.externalchurn.Main > \ --conf "spark.task.maxFailures=100" \ --conf > "spark.yarn.max.executor.failures=10000" \ --conf "spark.executor.cores=1" > \ --conf "spark.akka.frameSize=50" \ --conf > "spark.storage.memoryFraction=0.5" \ --conf > "spark.driver.maxResultSize=10G" \ > ~/external-flow/externalChurn-1.0-SNAPSHOT-shaded.jar \ > prepareTraining=true \ prepareTrainingMNP=true \ prepareMap=false \ > bouldozerMode=true \ &> ~/external-flow/run.log & echo "STARTED" tail -f > ~/external-flow/run.log > > Thanks, > > > > >