Support for multiple result nodes

[soxofaan]

follow up of Open-EO/openeo-processes#401

python client needs some kind of support for building process graphs with multiple result nodes

[jdries]

It looks like we can create a valid process graph from multiple datacubes, as long as we use the same flattener.

    week = s2cube.aggregate_temporal_period(period="week",reducer="median")
    month = s2cube.aggregate_temporal_period(period="month",reducer="median")

    cube1=week.save_result("netCDF")
    cube2 = month.save_result("netCDF")

    from openeo.internal.graph_building import GraphFlattener
    flattener = GraphFlattener()
    dict1 = flattener.flatten(cube1.result_node())

    #only one node can be marked as 'result':True, so set all others to False
    for v in dict1.values():
        v["result"] = False

    secondary_result = cube2.result_node()

    #by reusing the flattener, the previously flattened nodes are also included
    dict2 = flattener.flatten(secondary_result)

    final_graph = {"process_graph":dict2}
    import json
    merged = json.dumps(final_graph, indent=2)
    print(merged)

    connection = openeo.connect("openeo.dataspace.copernicus.eu")
    connection.create_job(final_graph,title="job with 2 save results")

[jdries]

so way forward is either updating create_job or a variant of that method that accepts multiple datacubes?

or
cube.execute_batch(args..., extra_cubes=[extra_cube1])
or
connection.execute_batch([cube1,cube2]), other args...)

[soxofaan]

In the python client we already have a base class _ProcessGraphAbstraction, which is further implemented by DataCube, VectorCube, MlModel. So I was thinking along the lines of creating another subclass of that, say ProcessGraph or MultiResult, which you can pass one or more cubes and work with that. E.g.

results = MultiResult([cube_as_tiff, cube_as_netcdf, vector_cube])

results.execute_batch(...)

(all naming up for discussion of course)