Spark - Application Execution Configuration

Card Puncher Data Processing


Computer - Capacity Planning (Sizing) in Spark to run an app

ie how to calculate:

  • Num-executors - The number of concurrent tasks (executor) that can be executed.
  • Executor-memory - The amount of memory allocated to each executor.
  • Executor-cores - The number of cores allocated to each executor.


Cluster Info

via Ambari or any admin, we can get the cluster composition

For instance:

  • 8 D4v2 nodes
  • Memory = 8 x 25 Gb since each D4v2 has 25GB of YARN memory.
Total YARN memory = nodes * YARN memory per node
Total YARN memory = 8 nodes * 25GB = 200GB
  • How many apps are running - for instance 2 apps on your cluster, including the one you are going to run.



For instance: 6GB of executor-memory are needed to load the data

set  executor-memory = 6GB


Setting cores per executor to larger than 4 may cause garbage collection problems.

set executor-cores = 4


The num-executors parameter is determined by taking the minimum of the memory constraint and the CPU constraint divided by the # of apps running on Spark.

  • Calculate memory constraint – The memory constraint is calculated as the total YARN memory divided by the memory per executor.
Memory constraint = (total YARN memory / executor memory) / # of apps   
Memory constraint = (200GB / 6GB) / 2   
Memory constraint = 16 (rounded)
  • Calculate CPU constraint - The CPU constraint is calculated as the total yarn cores divided by the number of cores per executor.
YARN cores = nodes in cluster * # of cores per node * 2   
YARN cores = 8 nodes * 8 cores per D14 * 2 = 128
CPU constraint = (total YARN cores / # of cores per executor) / # of apps
CPU constraint = (128 / 4) / 2
CPU constraint = 16
  • Set num-executors
num-executors = Min (memory constraint, CPU constraint)
num-executors = Min (16, 16)
num-executors = 16

Documentation / Reference

Discover More
Spark Program
Spark - Application

An application is an instance of a driver created via the initialization of a spark context (RDD) or a spark session (Data Set) This instance can be created via: a whole script (called batch mode)...

Share this page:
Follow us:
Task Runner