Priority of task inputs

Details

A node in a workflow can get inputs from three different sources:

1. Via the data_mapping link attribute of an incoming link (see Link attributes)

2. Via the parameters CLI argument (or inputs for Python) when executing/submitting the workflow (see ewoks execute reference)

3. Via the default_inputs node attribute of the node itself (see Node attributes)

If the same input is specified by these different sources, ewoks applies the following priorities:

• Data mapping takes precedence over workflow parameters and default inputs

• If the input is not specified by the data mapping, the workflow parameters takes precedence over the default inputs

• If the input is not specified by the data mapping nor the workflow parameters, default inputs are used as last resort

TL;DR

Data mapping > Workflow parameters > Node default inputs

Example

Consider the following workflow made of SumTask nodes from ewokscore (SumTask takes two inputs a and b and sums them):

wf = {
    "graph": {"id": "Sum"},
    "nodes": [
        {
            "id": "sum_node",
            "task_type": "class",
            "task_identifier": "ewokscore.tests.examples.tasks.sumtask.SumTask",
            "default_inputs": [{"name": "a", "value": 1}, {"name": "b", "value": 2}],
        },
    ],
    "links": [],
}

In this case, we have given two default_inputs to our sum_node:

• a = 1

• b = 2

If we execute the workflow with no parameters/inputs:

>>> execute_graph(graph=wf)
{'result': 3}

There is no incoming link with data mapping, no workflow inputs, the two default inputs are summed to get 3.

—

If we specify a value for b when executing the workflow:

>>> execute_graph(
...     graph=wf,
...     inputs=[
...         {
...             "name": "b",
...             "value": 20,
...             "id": "sum_node",
...         }
...     ],
... )
{'result': 21}

a is still 1 but the default value of b was replaced by the workflow input value (20), changing to result to 21.

—

If we change our workflow to include a SumTask before our sum_node with an incoming data mapping that sets b value to its result (in this case 200):

new_wf = {
    "graph": {"id": "Sum"},
    "nodes": [
        {
            "id": "first_node",
            "task_type": "class",
            "task_identifier": "ewokscore.tests.examples.tasks.sumtask.SumTask",
            "default_inputs": [{"name": "a", "value": 100}, {"name": "b", "value": 100}],
        },
        {
            "id": "sum_node",
            "task_type": "class",
            "task_identifier": "ewokscore.tests.examples.tasks.sumtask.SumTask",
            "default_inputs": [{"name": "a", "value": 1}, {"name": "b", "value": 2}],
        },
    ],
    "links": [
        {
            "source": "first_node",
            "target": "sum_node",
            "data_mapping": [{"source_output": "result", "target_input": "b"}],
        }
    ],
}

Then, if we execute as before (with the workflow input):

>>> execute_graph(
...     graph=new_wf,
...     inputs=[
...         {
...             "name": "b",
...             "value": 20,
...             "id": "sum_node",
...         }
...     ],
... )
{'result': 201}

We see that both the default input and the workflow input for b of sum_node are ignored, the data mapping taking precedence to change the result to 201.