Understanding workflow when multiple nodes are connected to a node

[clay-jona]

I have a graph that looks like

My expectation is that node a/b will be run first, c will be run second, and then d will be run.

But the actual run is that a & d runs, and then b & c & d also runs separately. So the final node calls are a, b, d, c, d (note 2 runs of d). I can confirm this happens in both synchronous and asynchronous cases.

But when I try a graph like below, d is run only once, not twice.

Could someone help me understand this discrepancy?

async def main():
  from pydantic import BaseModel
  from langgraph.graph import StateGraph

  class State(BaseModel):
      a: int = 0
      b: int = 0
      c: float = 0

  graph = StateGraph(State)
  graph.add_node("A", lambda state: {"a": state.a + 1})
  graph.add_node("B", lambda state: {"b": state.b + 1})
  graph.add_node("C", lambda state: {"b": state.b + 10})
  graph.add_node("D", lambda state: {"c": state.a / state.b})

  graph.add_edge("A", "D")
  graph.add_edge("B", "C")
  graph.add_edge("C", "D")

  graph.set_entry_point("A")
  graph.set_entry_point("B")
  graph.set_finish_point("D")

  graph = graph.compile()
  output = await graph.ainvoke({"a": 1, "b": 2})
  print("output:", output)

import asyncio
asyncio.run(main())