AsyncShallowRedisSaver._aload_pending_sends fails like AsyncRedisSaver prior to 0.0.7 version

[khangtrn]

Summary

The bug is similar to langchain-ai/langgraph#5074 but for AsyncShallowRedisSaver

Environment

Python version: 3.12
langgraph: 0.4.8
langgraph-checkpoint: 2.1.1
langgraph-checkpoint-redis: 0.0.7
langgraph-prebuilt: 0.6.2
langgraph-sdk: 0.2.0

Error log

025-08-01T04:44:09.505418295Z     async for mode, chunk in graph.astream(input=inp, stream_mode=["updates", "values"], config={"thread_id": thread_id}):
2025-08-01T04:44:09.505435671Z   File "/usr/local/lib/python3.12/site-packages/langgraph/pregel/__init__.py", line 2596, in astream
2025-08-01T04:44:09.505441129Z     async with AsyncPregelLoop(
2025-08-01T04:44:09.505452337Z                ^^^^^^^^^^^^^^^^
2025-08-01T04:44:09.505457470Z   File "/usr/local/lib/python3.12/site-packages/langgraph/pregel/loop.py", line 1339, in __aenter__
2025-08-01T04:44:09.505462789Z     saved = await self.checkpointer.aget_tuple(self.checkpoint_config)
2025-08-01T04:44:09.505467959Z             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2025-08-01T04:44:09.505480746Z   File "/usr/local/lib/python3.12/site-packages/langgraph/checkpoint/redis/ashallow.py", line 461, in aget_tuple
2025-08-01T04:44:09.505515351Z     pending_sends = await self._aload_pending_sends(
2025-08-01T04:44:09.505543051Z                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2025-08-01T04:44:09.505565466Z   File "/usr/local/lib/python3.12/site-packages/langgraph/checkpoint/redis/ashallow.py", line 701, in _aload_pending_sends
2025-08-01T04:44:09.505569244Z     return [(doc.type, doc.blob) for doc in sorted_writes]
2025-08-01T04:44:09.505572263Z                        ^^^^^^^^
2025-08-01T04:44:09.505575539Z AttributeError: 'Document' object has no attribute 'blob'

[abrookins]

@claude Triage this

[abrookins]

@khangtrn Thanks for reporting! We'll check this out.

[abrookins]

That's not confusing at all

[abrookins]

It looks like AsyncRedisSaver is the only checkpointer class that uses a defensive guard against a missing 'blob'. But why would the blob be null? 🤔 I need to clear up whether not getting that value back from the query is itself a deeper concern. If that's normal, then we should reproduce the same defensive check in the other checkpointers (normal sync plus both shallow classes).