0
votes

Hi i am launch my Spark application with the spark submit script as such

spark-submit --master spark://Maatari-xxxxxxx.local:7077 --class EstimatorApp /Users/sul.maatari/IdeaProjects/Workshit/target/scala-2.11/Workshit-assembly-1.0.jar  --d
eploy-mode cluster --executor-memory 15G num-executors 2

I have a spark standalone cluster deployed on two nodes (my 2 laptops). The cluster is running fine. By default it set 15G for the workers and 8 cores for the executors. Now i am experiencing the following strange behavior. Although i am explicity setting the memory and this can also be seen in the environmement variable of the sparconf UI, in the Cluster UI it says that my application is limited to 1024MB for the executor memory. This makes me think of the default 1G parameter. I wonder why that it.

Cluster UI Environment in the SparkConf UI

My application indeed fail because of the memory issue. I know that i need a lot of memory for that application.

One last point of confusion is the Driver program. Why given that i am on cluster mode, spark submit does not return immediately ? I though that given that the driver is executed on the cluster, the client i.e. submit application should return immediately. This further suggest me that something is not right with my conf and how things are being executed.

Can anyone help diagnose that ?

1
Spark default driver and executor memory is 1g. I am guessing somehow your spark.executorEnv. settings are not taken by spark. Try to set the --driver-memory to other values and test. Also, set them through the program or call toDebugString on conf and see the values to debug more.moon
Although not addressing the memory issue you have a separate mini problem: the num-executors 2 should be --num-executors 2WestCoastProjects

1 Answers

1
votes

Two possibilities:

  • given that your command line has the --num-executors mis-specified: it may be that Spark "gives up" on the other setting as well.
  • how much memory does your laptop have? Most of us use mac's .. and then you would not be able to run it with more than about 8GB in my experience.