Graph Lifecycle

Understanding the lifecycle of a single execution run is critical for debugging.

The Steps of Execution

graph TD
    Input[Input Data] --> Init[Initialize State]
    Init --> Entry[Entry Node]
    Entry --> Exec{Execute Logic}
    Exec --> Update[Update State]
    Update --> Check{Check Edges}
    Check -- "Next Node" --> Entry
    Check -- "END" --> Terminate[Finish]

1. Initialize: The graph receives input (e.g., {"messages": ["Hi"]}). This initializes the starting State.
2. Enter Entry Point: The graph identifies the first node.
3. Execute Node: The function runs. It returns a state update (e.g., {"key": "new_value"}).
4. Update State: The runtime merges the update into the global state.
5. Scan Edges: The runtime looks at outgoing edges from the current node.
6. Transition: The system moves to the next node.

Persistence (Checkpoints)

If you configured a Checkpointer (like Postgres), LangGraph saves the State to the database after Step 4 (every state update). This means if the server crashes at Step 5, you can resume exactly where you left off.