Can I pass a parameter into a graph?

Hi CJ. Yes, you can pass a parameter into a graph. Here's an example:

@op
def upstream_op():
    return 5

@op
def downstream_op(my_input):
    return my_input


@graph
def my_graph(my_input):
    downstream_op(my_input)


@job
def my_job():
    my_graph(upstream_op())

Thanks Claire! Can I use parameters I generate on the fly? Such as integers or strings? I want to process the same graph multiple times, and pass a different value to the graph each time.

You could provide inputs via

run_config

. For example:

@op
def downstream_op(my_input):
    return my_input


@graph
def my_graph(my_input):
    downstream_op(my_input)


my_graph.execute_in_process(
    run_config={"inputs": {"my_input": {"value": 8}}}
)

right... but I want to do this in a loop - processing the same graph multiple times. I think it might be a conceptual issue, in that Dagster needs to process the graph before it runs it. So a runtime loop makes no sense.

If you want to process the same graph multiple times, maybe you can try something like this with Dynamic Outputs:

@op
def downstream_op(my_input):
    return my_input

@graph
def my_graph(my_input):
    downstream_op(my_input)

@op(out=DynamicOut())
def dynamic_outs():
    for i in range(5):
        yield DynamicOutput(i, mapping_key=str(i))

@job
def my_job():
    dynamic_outs().map(my_graph)

Dynamic outputs will allow you to return an arbitrary number of outputs from an upstream op, and then you can map these outputs to be inputs to your graph.

Sounds great Claire, I think you gave me the recipe I need. Thanks!