fix(parallelism): mix sequence_id into parallel sub-app keys

Author: anon

burr/core/parallelism.py

@@ -518,7 +518,7 @@ def _create_task(key: str, action: Action, substate: State) -> SubGraphTask:
         def _tasks() -> Generator[SubGraphTask, None, None]:
             for i, action in enumerate(self.actions(state, context, inputs)):
                 for j, substate in enumerate(self.states(state, context, inputs)):
-                    key = f"{i}-{j}"  # this is a stable hash for now but will not handle caching
+                    key = f"{context.sequence_id}-{i}-{j}"
                     yield _create_task(key, action, substate)
 
         async def _atasks() -> AsyncGenerator[SubGraphTask, None]:
@@ -528,7 +528,7 @@ async def _atasks() -> AsyncGenerator[SubGraphTask, None]:
             states = await async_utils.arealize(state_generator)
             for i, action in enumerate(actions):
                 for j, substate in enumerate(states):
-                    key = f"{i}-{j}"
+                    key = f"{context.sequence_id}-{i}-{j}"
                     yield _create_task(key, action, substate)
 
         return _atasks() if self.is_async() else _tasks()

tests/core/test_parallelism.py

@@ -45,7 +45,12 @@
     _cascade_adapter,
     map_reduce_action,
 )
-from burr.core.persistence import BaseStateLoader, BaseStateSaver, PersistedStateData
+from burr.core.persistence import (
+    BaseStateLoader,
+    BaseStateSaver,
+    InMemoryPersister,
+    PersistedStateData,
+)
 from burr.tracking.base import SyncTrackingClient
 from burr.visibility import ActionSpan
 
@@ -1227,3 +1232,69 @@ def reads(self) -> list[str]:
     assert task.state_initializer is not None
     assert task.tracker is not None
     assert task.state_persister is task.state_initializer  # This ensures they're the same
+
+
+def test_map_states_no_stale_replay_on_repeated_invocation():
+    """Regression test for #761.
+
+    When the parent app cascades a state initializer (via ``initialize_from``)
+    and a MapStates action is invoked more than once, each invocation must
+    spawn fresh sub-applications. Before the fix, sub-app ids were keyed only
+    on ``(i, j)``, so they collided across invocations and the cascaded
+    initializer replayed the prior call's persisted sub-state.
+    """
+
+    @old_action(reads=["round"], writes=["output_number"])
+    def emit_round(state: State) -> State:
+        return state.update(output_number=state["round"])
+
+    @old_action(reads=["round"], writes=["round"])
+    def bump(state: State) -> State:
+        return state.update(round=state["round"] + 1)
+
+    class Fan(MapStates):
+        def action(self, state: State, inputs: Dict[str, Any]):
+            return emit_round
+
+        def states(
+            self, state: State, context: ApplicationContext, inputs: Dict[str, Any]
+        ) -> Generator[State, None, None]:
+            for _ in range(3):
+                yield state
+
+        def reduce(self, state: State, states: Generator[State, None, None]) -> State:
+            new_state = state
+            for output_state in states:
+                new_state = new_state.append(outputs=output_state["output_number"])
+            return new_state
+
+        @property
+        def reads(self) -> list[str]:
+            return ["round"]
+
+        @property
+        def writes(self) -> list[str]:
+            return ["outputs"]
+
+    persister = InMemoryPersister()
+    app = (
+        ApplicationBuilder()
+        .with_actions(fan=Fan(), bump=bump)
+        .with_transitions(("fan", "bump"), ("bump", "fan"))
+        .with_state_persister(persister)
+        .with_identifiers(app_id="parent-app")
+        .initialize_from(
+            persister,
+            resume_at_next_action=True,
+            default_state={"round": 1, "outputs": []},
+            default_entrypoint="fan",
+        )
+        .build()
+    )
+
+    app.run(halt_after=["fan"])  # first fan invocation, round=1
+    app.run(halt_after=["fan"])  # bump runs, then second fan invocation, round=2
+
+    # Each fan invocation contributes 3 outputs. Fresh execution -> [1,1,1,2,2,2].
+    # Buggy behavior replays the first invocation's persisted sub-state -> [1,1,1,1,1,1].
+    assert list(app.state["outputs"]) == [1, 1, 1, 2, 2, 2]