Retries were previously entirely in-memory: the attempt counter and backoff sleep happened inside run_with_retry/arun_with_retry without any checkpoint writes. If the process crashed mid-retry, the state was lost and the task would restart from attempt 0 on resume, effectively ignoring max_attempts across restarts. This change adds a RETRY special write channel. On each retryable failure, before sleeping, the retry loop persists (attempt_count, next_retry_timestamp) to the checkpoint via put_writes. On resume, the scratchpad restores these values so the retry loop continues from the correct attempt and honors the remaining backoff time. Key changes: - New RETRY constant in _constants.py and checkpoint serde/types - RETRY added to WRITES_IDX_MAP for checkpoint storage - PregelScratchpad gains retry_attempt/retry_ts fields - _scratchpad() in _algo.py extracts RETRY from pending_writes - _match_writes skips RETRY (like ERROR/INTERRUPT/RESUME) - output_writes returns early for RETRY (no stream output) - run_with_retry/arun_with_retry accept put_writes callback, restore attempt state from scratchpad, persist before sleep - Runner passes put_writes to retry functions https://claude.ai/code/session_01768mrKrVCCtXdJgRNFdWrb
LangGraph Checkpoint
This library defines the base interface for LangGraph checkpointers. Checkpointers provide a persistence layer for LangGraph. They allow you to interact with and manage the graph's state. When you use a graph with a checkpointer, the checkpointer saves a checkpoint of the graph state at every superstep, enabling several powerful capabilities like human-in-the-loop, "memory" between interactions and more.
Key concepts
Checkpoint
Checkpoint is a snapshot of the graph state at a given point in time. Checkpoint tuple refers to an object containing checkpoint and the associated config, metadata and pending writes.
Thread
Threads enable the checkpointing of multiple different runs, making them essential for multi-tenant chat applications and other scenarios where maintaining separate states is necessary. A thread is a unique ID assigned to a series of checkpoints saved by a checkpointer. When using a checkpointer, you must specify a thread_id and optionally checkpoint_id when running the graph.
thread_idis simply the ID of a thread. This is always required.checkpoint_idcan optionally be passed. This identifier refers to a specific checkpoint within a thread. This can be used to kick off a run of a graph from some point halfway through a thread.
You must pass these when invoking the graph as part of the configurable part of the config, e.g.
{"configurable": {"thread_id": "1"}} # valid config
{"configurable": {"thread_id": "1", "checkpoint_id": "0c62ca34-ac19-445d-bbb0-5b4984975b2a"}} # also valid config
Serde
langgraph_checkpoint also defines protocol for serialization/deserialization (serde) and provides an default implementation (langgraph.checkpoint.serde.jsonplus.JsonPlusSerializer) that handles a wide variety of types, including LangChain and LangGraph primitives, datetimes, enums and more.
Pending writes
When a graph node fails mid-execution at a given superstep, LangGraph stores pending checkpoint writes from any other nodes that completed successfully at that superstep, so that whenever we resume graph execution from that superstep we don't re-run the successful nodes.
Interface
Each checkpointer should conform to langgraph.checkpoint.base.BaseCheckpointSaver interface and must implement the following methods:
.put- Store a checkpoint with its configuration and metadata..put_writes- Store intermediate writes linked to a checkpoint (i.e. pending writes)..get_tuple- Fetch a checkpoint tuple using for a given configuration (thread_idandcheckpoint_id)..list- List checkpoints that match a given configuration and filter criteria..delete_thread()- Delete all checkpoints and writes associated with a thread..get_next_version()- Generate the next version ID for a channel.
If the checkpointer will be used with asynchronous graph execution (i.e. executing the graph via .ainvoke, .astream, .abatch), checkpointer must implement asynchronous versions of the above methods (.aput, .aput_writes, .aget_tuple, .alist). Similarly, the checkpointer must implement .adelete_thread() if asynchronous thread cleanup is desired. The base class provides a default implementation of .get_next_version() that generates an integer sequence starting from 1, but this method should be overridden for custom versioning schemes.
Usage
from langgraph.checkpoint.memory import InMemorySaver
write_config = {"configurable": {"thread_id": "1", "checkpoint_ns": ""}}
read_config = {"configurable": {"thread_id": "1"}}
checkpointer = InMemorySaver()
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))