AsyncRedisSaver aget_tuple Incorrect Checkpoint ID

[ghulvey-yardi]

The aget_tuple function in AsyncRedisSaver is not returning the correct checkpoint_id unless it is specified in the RunnableConfig. If you call aget_tuple with a runnable config that contains only thread_id and checkpoint_ns the returned checkpoint_id will always be None. RedisSaver does not have this problem, checkpoint_id appears to always be set.

Replication

import asyncio
import uuid
from langgraph.checkpoint.base import empty_checkpoint
from langgraph.checkpoint.redis import RedisSaver
from langgraph.checkpoint.redis.aio import AsyncRedisSaver

def sync_checkpointer():
    runnable_config = {
        "configurable": {"thread_id": str(uuid.uuid4()), "checkpoint_ns": ""}
    }
    with RedisSaver.from_conn_string("redis://localhost:6379") as checkpointer:
        checkpointer.setup()
        for run in range(3):
            checkpointer.put(
                {
                    "configurable": {
                        "thread_id": runnable_config["configurable"]["thread_id"],
                        "checkpoint_id": str(run),
                        "checkpoint_ns": runnable_config["configurable"][
                            "checkpoint_ns"
                        ],
                    }
                },
                empty_checkpoint(),
                {
                    "source": "loop",
                    "step": run,
                    "writes": {},
                },
                {},
            )
            get_tuple = checkpointer.get_tuple(runnable_config)
            print(
                f"Get tuple ({run}): {get_tuple.config['configurable']['checkpoint_id']}"
            )


async def async_checkpointer():
    runnable_config = {
        "configurable": {"thread_id": str(uuid.uuid4()), "checkpoint_ns": ""}
    }
    async with AsyncRedisSaver.from_conn_string(
        "redis://localhost:6379"
    ) as checkpointer:
        await checkpointer.asetup()
        for run in range(3):
            await checkpointer.aput(
                {
                    "configurable": {
                        "thread_id": runnable_config["configurable"]["thread_id"],
                        "checkpoint_id": str(run),
                        "checkpoint_ns": runnable_config["configurable"][
                            "checkpoint_ns"
                        ],
                    }
                },
                empty_checkpoint(),
                {
                    "source": "loop",
                    "step": run,
                    "writes": {},
                },
                {},
            )
            get_tuple = await checkpointer.aget_tuple(runnable_config)
            print(
                f"Get tuple ({run}): {get_tuple.config['configurable']['checkpoint_id']}"
            )


if __name__ == "__main__":
    print("Sync checkpointer:")
    sync_checkpointer()
    print("=======================================")
    print("Async checkpointer:")
    asyncio.run(async_checkpointer())

Replication Output

Sync checkpointer:
Get tuple (0): 0
Get tuple (1): 1
Get tuple (2): 2
=======================================
Async checkpointer:
Get tuple (0): None
Get tuple (1): None
Get tuple (2): None

[tylerhutcherson]

@claude take a look at this please

[claude]

Claude encountered an error —— View job

---

I'll analyze this and get back to you.

[claude]

Claude finished @tylerhutcherson's task —— View job • claude/issue-64-20250623_221640 • Create PR ➔

---

AsyncRedisSaver aget_tuple Checkpoint ID Fix Complete ✅

I've successfully identified and fixed the bug where AsyncRedisSaver.aget_tuple() was returning None for checkpoint_id while the sync version worked correctly.

Root Cause

The issue was in langgraph/checkpoint/redis/aio.py at two locations:

• Line 353: Used original checkpoint_id (None) instead of doc_checkpoint_id in config construction
• Line 364: Used original checkpoint_id instead of doc_checkpoint_id in _aload_pending_writes call

Fix Applied

Fixed both locations to use doc_checkpoint_id (the actual retrieved checkpoint ID from Redis) instead of checkpoint_id (the input parameter which may be None).

Changes Made:

1. Fixed langgraph/checkpoint/redis/aio.py:

   • Line 353: Changed "checkpoint_id": checkpoint_id to "checkpoint_id": doc_checkpoint_id
   • Line 364: Changed checkpoint_id or EMPTY_ID_SENTINEL to doc_checkpoint_id

2. Added comprehensive test coverage in tests/test_async_aget_tuple_checkpoint_id.py:

   • Test that reproduces the original bug scenario
   • Test with explicit checkpoint_id (should continue working)
   • Test with non-existent thread (should return None)

The fix ensures that AsyncRedisSaver.aget_tuple() now behaves identically to the sync version, returning the correct checkpoint ID from the retrieved Redis document instead of None.

---