mirror of
https://github.com/langchain-ai/langgraph.git
synced 2026-09-05 17:27:47 +02:00
e095d9c41a
DeltaChannel reconstructs its value by replaying ancestor writes through the reducer. Every saver ordered a checkpoint's writes by (task_id, idx), but live execution applies them in task-path order: apply_writes sorts a super-step's tasks by task_path_str(task.path[:3]) before calling channel.update. task_id is a hash of the path, so the two orders are unrelated, and two or more tasks writing one DeltaChannel in a single super-step replayed in an arbitrary permutation. Reducers are only required to be batching-invariant, not order-invariant, so the permutation changes the value: get_state disagreed with what invoke returned, and continuing the thread persisted the reordered replay as the base for later writes. Replay now orders by (task_path, task_id, idx), following the precedent already set for the Send channel by SELECT_PENDING_SENDS_SQL. InMemorySaver and the postgres savers already persisted task_path and only needed the sort key; sqlite accepted task_path on put_writes and dropped it, so the writes table gains the column plus a probe-guarded ALTER for databases created by earlier versions. Writes stored without a task_path sort first within their checkpoint, which is where live execution applies the task-less input writes that carry "". The graph-level regression tests run against the full async_checkpointer matrix (memory, sqlite, postgres in three pool modes) because each saver reconstructs delta channels through its own override; 20 of the 25 cases fail on main, and the 5 that pass are the sequential control. Co-authored-by: ErenAta16 <149434812+ErenAta16@users.noreply.github.com> Co-authored-by: ragnarok268 <58264829+ragnarok268@users.noreply.github.com>