Issue: draw_all_possible_flows only displays one branch instead of 3

[alex-stoica]

I have a workflow that creates three branches and then combines them later with some logging. However, when I attempt to visualize the workflow using draw_all_possible_flows, I only see a single LetterRequest coming out of the dispatch step instead of the expected three.

Here is the relevant part of my workflow

@step
async def dispatch(self, ctx: Context, ev: StartEvent) -> LetterRequest | LetterRequest | LetterRequest | GatherEvent:
    await ctx.set("separator", ev.get("separator", ""))
    await ctx.set("workflow_start", format_time(datetime.now()))
    ctx.send_event(LetterRequest(branch="A"))
    ctx.send_event(LetterRequest(branch="B"))
    ctx.send_event(LetterRequest(branch="C"))
    return GatherEvent()

# ...
draw_all_possible_flows(branch_workflow, "branch_workflow_visualization.html")

Is this the expected behavior of draw_all_possible_flows, or am I missing something? Should I be structuring the dispatch step differently to ensure all branches are properly visualized?

Thanks!

[logan-markewich]

This is expected @alex-stoica -- The drawing util is only drawing steps and which events trigger those steps.

In this case, you have one step (dispatch), which can emit one or more LetterRequest events. And each LetterRequest event will trigger a run of process_letter for that instance of that event.

So, the drawing is indeed accurate. And this is totally a fine way to structure a workflow (I think we even show this pattern in the docs)

[alex-stoica]

Thanks @logan-markewich, I see the rationale now - show only the steps + event types that trigger them, not all the individual instances.
I agree that in many cases this higher level overview is sufficient. However, do you think it might be beneficial to offer an option or an alternate function to display each individual event? It might help understanding the internal workings of multi-branch steps

[logan-markewich]

There's no way for us to know what a user wrote in a workflow (like send_event()) without actually running that workflow. This is the difference between draw_all_possible_flows and draw_most_recent_execution.

Although now that I think about it, I'm not actually sure what draw_most_recent_execution would draw in this case