Spark applications run as a collection of multiple processes. Each application consists of a process for the main program (the driver program), and one or more executor processes that run Spark tasks.

These processes are multithreaded. Figure 14  illustrates the general Spark execution model.

Figure 14: Spark execution model

Spark relies on a separate Cluster Manager to allocate and manage resources for these processes in multinode environments. Spark supports several cluster managers:

• Spark Standalone - A simple cluster manager included with Spark
• Apache Mesos - A general-purpose cluster manager, invented at the UC Berkeley AMPLab, which invented Spark
• Apache Hadoop YARN - A resource manager developed for the Apache Hadoop project
• Kubernetes – A system for automating deployment, scaling, and management of containerized applications

This reference architecture uses Kubernetes as the cluster manager.