`graph.add_edge([parents], child)` triggers child multiple times instead of once after all parents complete

OSS Product Help LangGraph
1

Description

When creating a node with multiple parents using the list syntax in graph.add_edge,
I expected the child node to run once after all parent nodes have completed.

Example:

graph.add_edge(START, "step_1")
graph.add_edge(START, "step_2")
graph.add_edge("step_2", "step_3")
graph.add_edge(["step_1", "step_3"], "step_4")
graph.add_edge("step_4", END)

Expected Behavior

step_4 should execute only once, after both step_1 and step_3 have finished.

Actual Behavior

step_4 is executed twice:

  • once after step_1 completes

  • again after step_3 completes

This results in multiple executions of the same node (e.g., LLM called twice with slightly different results),
even though the intent was to wait for all parents and run only once.

Why this matters

For workflows that aggregate results from multiple branches (e.g., synthesis step),
we need the child node to run once with the combined inputs, not multiple times.

Question

  • Is the current behavior (multiple triggers) by design?

  • If so, could we add an “all parents must complete” semantics for list-style edges?

  • Or should this require an explicit join node in user code?

2

hi @ssssssss

could you share all node definitions too?