Parallel nodes without a branch

OSS Product Help LangGraph
1

Hello,

I got a workflow with the following logic:

  • Node 1 makes an action and depending on the result may trigger Node 2,Node 3,Node 4 and wi
  • Node 2,Node 3 launch subgraph async tasks and should not do anything after that simply end do not go back into the graph

-Node 4 does another action that may retrigger Node 1

My question is should i route Node 2 and Node 3 to END ?
I m actually currently debugging it and im left under the impression this ends the whole graph .Ending should only be decided in Node 4.
Not sure about that .

Thanks !

2

Hi @Cgrdj

CAUTION: this is my hypothetical answer based on theory :slight_smile: If you shared your code, I would say more accurately regarding the issue you are facing.

Route Node 2 and Node 3 to END (or give them no outgoing edges). This ends only those branches. The overall graph finishes only when there are no runnable tasks left. Node 4 can continue independently and re-trigger Node 1 as designed.

  • What END means: END is a special sink node. Adding an edge to END means “no further scheduling for this branch.” The run completes when there are no pending Sends and no nodes triggered by channel updates - i.e., when all active branches have finished.

  • Parallel branches: If Node 1 fans out to Node 2, Node 3, and Node 4 (via conditional edges or returning multiple Send(...) packets), those branches execute independently. Routing Node 2/3 to END won’t cancel Node 4.

  • Looping control: If Node 4 may loop back to Node 1, give Node 4 an edge to Node 1 (or a conditional mapping that sometimes returns Node 1 and sometimes END). Termination is then decided by Node 4’s path (and any other still-active branches), not by Node 2/3.

Example


from langgraph.graph import StateGraph, START, END

builder = StateGraph(State)

builder.add_node("node1", node1)
builder.add_node("node2", node2)
builder.add_node("node3", node3)
builder.add_node("node4", node4)
# Node 1 can fan out to multiple nodes
builder.add_conditional_edges("node1", route, ["node2", "node3", "node4"]) # or return [Send(...), ...]

# Node 2 and Node 3 should finish their branches
builder.add_edge("node2", END)
builder.add_edge("node3", END)
# Node 4 may loop back to Node 1
builder.add_edge("node4", "node1")
If your run appears to “end early”

  • Ensure Node 1 really fans out (e.g., returns multiple Send(...) or a conditional mapping to multiple next nodes).

  • Verify Node 4’s logic doesn’t map to END in the case you’re testing.

  • Confirm Node 2/3 aren’t the only triggered branches (i.e., that Node 4 was actually scheduled in that step).