Files
langgraph/libs/checkpoint-postgres
Sydney Runkle d120f127ca refactor(delta-channel): plain SELECT WHERE replaces recursive CTE
The recursive CTE was bottlenecked by a JSON-expression join
(`bl.version = checkpoint->'channel_versions'->>bl.channel`) that the
planner could not index, producing an O(ancestors x blobs) nested-loop.
At depth 1000 it ran ~275 ms and removed ~2M filter rows; the recursion
itself was 2.4 ms.

Switch to three plain indexed SELECTs per delta channel
(checkpoints, checkpoint_writes, checkpoint_blobs); a pure helper on
BasePostgresSaver walks the parent chain and assembles
DeltaChannelWrites. Sync (__init__.py) and async (aio.py) each own
their three-roundtrip I/O wrappers.

Bench numbers (notes/delta_channel_query_bench.md): 3x at depth 50,
15x at depth 200, ~100x at depth 1000. Plain over-fetches sibling rows
when the thread branches but still wins at every realistic depth on
both local and remote postgres.

Multi-channel coalescing dropped — reconstruction is per-channel now.
Same shape as InMemorySaver. Can come back as a SQL-level
optimization later if needed.
2026-04-23 08:21:00 -04:00
..
2026-04-22 17:47:53 -04:00

LangGraph Checkpoint Postgres

Implementation of LangGraph CheckpointSaver that uses Postgres.

Dependencies

By default langgraph-checkpoint-postgres installs psycopg (Psycopg 3) without any extras. However, you can choose a specific installation that best suits your needs here (for example, psycopg[binary]).

Security

Important

Set LANGGRAPH_STRICT_MSGPACK=true or pass an explicit allowed_msgpack_modules list when creating your checkpointer. This restricts checkpoint deserialization to known-safe types, preventing code execution if the database is compromised. See the langgraph-checkpoint README for details.

Usage

Important

When using Postgres checkpointers for the first time, make sure to call .setup() method on them to create required tables. See example below.

Important

When manually creating Postgres connections and passing them to PostgresSaver or AsyncPostgresSaver, make sure to include autocommit=True and row_factory=dict_row (from psycopg.rows import dict_row). See a full example in this how-to guide.

Why these parameters are required:

  • autocommit=True: Required for the .setup() method to properly commit the checkpoint tables to the database. Without this, table creation may not be persisted.
  • row_factory=dict_row: Required because the PostgresSaver implementation accesses database rows using dictionary-style syntax (e.g., row["column_name"]). The default tuple_row factory returns tuples that only support index-based access (e.g., row[0]), which will cause TypeError exceptions when the checkpointer tries to access columns by name.

Example of incorrect usage:

# ❌ This will fail with TypeError during checkpointer operations
with psycopg.connect(DB_URI) as conn:  # Missing autocommit=True and row_factory=dict_row
    checkpointer = PostgresSaver(conn)
    checkpointer.setup()  # May not persist tables properly
    # Any operation that reads from database will fail with:
    # TypeError: tuple indices must be integers or slices, not str
from langgraph.checkpoint.postgres import PostgresSaver

write_config = {"configurable": {"thread_id": "1", "checkpoint_ns": ""}}
read_config = {"configurable": {"thread_id": "1"}}

DB_URI = "postgres://postgres:postgres@localhost:5432/postgres?sslmode=disable"
with PostgresSaver.from_conn_string(DB_URI) as checkpointer:
    # call .setup() the first time you're using the checkpointer
    checkpointer.setup()
    checkpoint = {
        "v": 4,
        "ts": "2024-07-31T20:14:19.804150+00:00",
        "id": "1ef4f797-8335-6428-8001-8a1503f9b875",
        "channel_values": {
            "my_key": "meow",
            "node": "node"
        },
        "channel_versions": {
            "__start__": 2,
            "my_key": 3,
            "start:node": 3,
            "node": 3
        },
        "versions_seen": {
            "__input__": {},
            "__start__": {
            "__start__": 1
            },
            "node": {
            "start:node": 2
            }
        },
    }

    # store checkpoint
    checkpointer.put(write_config, checkpoint, {}, {})

    # load checkpoint
    checkpointer.get(read_config)

    # list checkpoints
    list(checkpointer.list(read_config))

Async

from langgraph.checkpoint.postgres.aio import AsyncPostgresSaver

async with AsyncPostgresSaver.from_conn_string(DB_URI) as checkpointer:
    checkpoint = {
        "v": 4,
        "ts": "2024-07-31T20:14:19.804150+00:00",
        "id": "1ef4f797-8335-6428-8001-8a1503f9b875",
        "channel_values": {
            "my_key": "meow",
            "node": "node"
        },
        "channel_versions": {
            "__start__": 2,
            "my_key": 3,
            "start:node": 3,
            "node": 3
        },
        "versions_seen": {
            "__input__": {},
            "__start__": {
            "__start__": 1
            },
            "node": {
            "start:node": 2
            }
        },
    }

    # store checkpoint
    await checkpointer.aput(write_config, checkpoint, {}, {})

    # load checkpoint
    await checkpointer.aget(read_config)

    # list checkpoints
    [c async for c in checkpointer.alist(read_config)]