Implementation notes#

Graph of actors#

An Ewoks task graph is converted to a graph of actors before execution.

Source#

A task is wrapped by an EwoksPythonActor followed by some binder actors, depending on the type of link between source and target. For each link in the task graph, the final actor of the source is a NameMapperActor which is connected to the target.

Unconditional link

        graph LR;
EwoksPythonActor-->NameMapperActor;
    

Conditional link with one condition

        graph LR;
EwoksPythonActor-->ConditionalActor;
ConditionalActor-->NameMapperActor;
    

On-Error conditional link

        graph LR;
EwoksPythonActor-->NameMapperActor;
    

Target#

        graph LR;
InputMergeActor-->EwoksPythonActor;
    

The NameMapperActor of a source for a specific target is connected to the InputMergeActor before that target.

Start actor#

The StartActor is connected to all tasks that are the start of a task scheduling thread.

Stop actor#

Tasks without successors are connected to the StopActor. This actor manages a dictionary of outputs which is considered to be the “output of the graph”.

Error handler#

All actors that do not have an _on_error_ conditional link are connected to the ErrorHandler which directly connected to the StopActor.

Custom actors#

These actors are not provided by pypushflow

  • EwoksythonActor: like PythonActor but it passes node name and attributes to the next actor.

  • InputMergeActor: like Joinactor (merges the input data dictionaries) triggers the downstream actors when all required input has been provided at least once. Only one non-required input passed.

  • ConditionalActor: triggers downstream actors when all conditions are fulfilled.

  • NameMapperActor: before triggering the next task it applies filtering and name mapping to the input data.