Evaluator

Evaluators and Tasks

The Evaluator is the runtime environment for Tasks. On one Evaluator, there is either no or one Task executing at any given point in time. Different or multiple executions of the same Tasks can be executed in sequence on an Evaluator. The Evaluator and Task lifecycle are decoupled: Whenever a Task finishes, the Driver receives the CompletedTask event, which contains a reference to the Evaluator the Task executed on. It is then up to the Driver to decide whether to return the Evaluator to the resource manager or to make other use of it, e.g. by submitting another task.

Evaluators and Contexts

Contexts are REEF's form of state management inside of the Evaluator. See the Context section for more information.

Evaluators and the Resource Manager

On typical resource managers, an Evaluator is a process executing inside a container. Depending on the resource manager, that process may or may not be guarded by a resource or security isolation layer.

This also means that the Evaluator, not the Task, is the unit of resource consumption: while an Evaluator is occupying a Container, that Container is "used" from the perspective of the Resource Manager.That is true even if the Evaluator is idle from the perspective of the Driver, i.e. when no Task is running on it.