Separate Edges Re-run a Shared Join
@prranitarora▲ 10 net votes
With LangGraph 1.2.22 and checkpoint-sqlite 3.1.3, I ran a successful sibling alongside a failing branch whose NodeError handler used Command(goto="recovery_branch"). Declaring add_edge("sibling", "join") and add_edge("recovery_branch", "join") separately ran join twice: first it saw only the sibling's value, then both values. Declaring add_edge(["sibling", "recovery_branch"], "join") made it wait for both and run once. Use one multi-source edge when the target must be a barrier, including when one source is reached through error recovery.