Configuring app runs individually¶

By default, the run configuration and instance of an app step are set once, on the app decorator. Every run of that step then uses the same values. When fanning out with the foreach decorator, this is not always what you want: the workflow often already knows something about each scenario that should influence how it runs.

For this, an AppRunConfig can carry its own RunConfiguration and instance_id, in addition to the input and options.

Choosing the execution class per run¶

The following example gives large scenarios a bigger execution class:

import nextmv

from nextpipe import AppRunConfig, app, foreach, needs, step

@foreach()
@step
def create_scenarios(data: dict):
    """Create scenarios, sizing the run to the scenario at hand."""
    return [
        AppRunConfig(
            name=f"scenario-{scenario['id']}",
            input=scenario,
            run_configuration=nextmv.RunConfiguration(
                execution_class="8c16gb12h" if scenario["large"] else "4c8gb12h",
            ),
        )
        for scenario in split(data)
    ]

@needs(predecessors=[create_scenarios])
@app(app_id="solver-app")
@step
def solve():
    """Run external solver for each scenario."""
    pass

Choosing the instance per run¶

In the same way, instance_id sends a run to a specific instance, e.g., one tuned for large scenarios. Scenarios that leave it unset run on the instance given to the decorator:

AppRunConfig(
    input=scenario,
    instance_id="tuned-for-large" if scenario["large"] else None,
)

How the values are combined¶

Both fields are optional. If set, they replace the corresponding argument of the app decorator for that run, rather than merging with it. For run_configuration this means the whole configuration is replaced — no individual fields are inherited from the decorator. If a field is left unset on the AppRunConfig, the value from the decorator is used unchanged.