Update docstring on scheduling

src/chatbot/graph.py

@@ -1,5 +1,6 @@
 """Example chatbot that incorporates user memories."""
 
+import uuid
 from dataclasses import dataclass
 from datetime import datetime, timezone
 
@@ -47,8 +48,26 @@ async def schedule_memories(state: ChatState, config: RunnableConfig) -> None:
     """Prompt the bot to respond to the user, incorporating memories (if provided)."""
     configurable = ChatConfigurable.from_runnable_config(config)
     memory_client = get_client()
+    mem_thread_id = str(
+        uuid.uuid5(
+            uuid.NAMESPACE_URL,
+            configurable.mem_assistant_id + config["configurable"]["thread_id"],
+        )
+    )
+    await memory_client.threads.create(thread_id=mem_thread_id, if_exists="do_nothing")
     await memory_client.runs.create(
-        thread_id=config["configurable"]["thread_id"],
+        # Generate a thread so we can run the memory service on a separate
+        # but consistent thread. This lets us cancel scheduled runs if
+        # a new message arrives to our chatbot before the memory service
+        # begins processing.
+        thread_id=mem_thread_id,
+        # Rollback & cancel any scheduled runs for the target thread
+        # that haven't completed
+        multitask_strategy="rollback",
+        # This lets us "debounce" repeated requests to the memory graph
+        # if the user is actively engaging in a conversation
+        after_seconds=configurable.delay_seconds,
+        # Specify the graph and/or graph configuration to handle the memory processing
         assistant_id=configurable.mem_assistant_id,
         input={
             # the service dedupes messages by ID, so we can send the full convo each time
@@ -57,13 +76,10 @@ async def schedule_memories(state: ChatState, config: RunnableConfig) -> None:
         },
         config={
             "configurable": {
+                # Ensure the memory service knows where to save the extracted memories
                 "user_id": configurable.user_id,
             },
         },
-        multitask_strategy="enqueue",
-        # This lets us "debounce" repeated requests to the memory graph
-        # if the user is actively engaging in a conversation
-        after_seconds=configurable.delay_seconds,
     )