Running PySpark in a Virtual Environment

For many PySpark applications, it is sufficient to use --py-files to specify dependencies. However, there are times when --py-files is inconvenient, such as the following scenarios:

• A large PySpark application has many dependencies, including transitive dependencies.

• A large application needs a Python package that requires C code to be compiled before installation.

• You want to run different versions of Python for different applications.

For these situations, you can create a virtual environment as an isolated Python runtime environment. HDP supports VirtualEnv for PySpark in both local and distributed environments, easing the transition from a local environment to a distributed environment.

	Note
	This feature is currently only supported in YARN mode.