Compare commits

..
Author SHA1 Message Date
Elior Nataf Lackritz 3be96b7d73 refactor(langgraph): drop the deferred fork-snapshot queue
Now that a forced snapshot mints a version for a never-written channel,
the first checkpoint a forked run writes seals every delta channel, so
_delta_channels_awaiting_fork_snapshot never outlived it. Seed
_delta_channels_forced_snapshot directly at loop construction instead.

Also asserts that forking by invoke leaves the abandoned branch's reads
intact, and trims comments and test docstrings.
2026-09-23 10:50:20 -04:00
Elior Nataf Lackritz 302ce795a5 fix(langgraph): seal a fork whose delta channel has no value yet
A DeltaChannel that was never written on the branch being forked has no
value to snapshot and no entry in channel_versions, so create_checkpoint
skipped it and the fork's first checkpoint recorded no boundary at all.
The walk then ran past the fork into the shared base and collected the
abandoned branch's writes, the same failure this branch already fixes for
channels that do have a value.

Two shapes leaked. A run forking off a checkpoint older than the channel's
first value and never writing that channel returned ['in-1'] where the
plain-channel oracle returned []. A bulk update writing the delta key only
in its second superstep returned ['in-1', 's2'] against ['s2'].

No new blob type is needed. _DeltaSnapshot already carries the value and
is already serialized by every saver, and from_checkpoint turns MISSING
into typ(), so _DeltaSnapshot(typ()) reconstructs to the same empty value
the channel would have had. What was missing is a version: without one,
put drops the blob as not-a-new-version, so mint a first one.

Deferring the seal to a later superstep does not work. That superstep
reconstructs through the still-unsealed checkpoint and would only bake the
corrupted value into its own snapshot.

Checked that minting a version does not fire nodes that subscribe to the
channel: a raw Pregel node subscribed directly to the delta channel stays
silent across the fork.

Reported by the Open SWE review bot on #8548.
2026-09-23 10:42:44 -04:00
Elior Nataf Lackritz 0552703e21 test(langgraph): compare read-back checkpoints in the immutability saver
MemorySaverAssertImmutable recorded the checkpoint object handed to put,
then compared it against one read back through get. Those two are not the
same shape: channel_values are stored per (channel, version), so a channel
a step did not write is refilled from the blob its inherited version still
points at.

Every channel except DeltaChannel writes its value into channel_values on
every checkpoint, so the two agreed by accident. A DeltaChannel stores
nothing except at a snapshot, so once one snapshots and a later step does
not write it, the saver reports a checkpoint that changed after it was
written when nothing was mutated.

Reproducible on main with no fork involved: a delta channel with
snapshot_frequency=1 written by the first node and left alone by the next
two trips the assertion. Existing delta tests miss it only because they
all use snapshot_frequency=1000.

Record what the saver reads back instead. Comparing read-back against
read-back still catches a checkpoint whose stored data really changed.
2026-09-23 10:42:44 -04:00
Elior Nataf Lackritz c4df893d54 fix(langgraph): seal a fork on the first checkpoint it writes
The as_node INPUT, END and __copy__ paths write a checkpoint and return
before create_checkpoint_plan_for_update_state_api runs, so a bulk update
whose first superstep took one of them left the branch unsealed. Only
INPUT actually leaked: END absorbs the base's already-run task writes, so
its delta and plain channels agree.

Sealing on a later superstep does not help. By then that superstep has
reconstructed its value by walking through the unsealed checkpoint into
the shared base, so it snapshots an already-corrupted list. The fork's
first checkpoint is the one that has to carry the blob, which is what
create_fork_checkpoint does.

That snapshot was still being dropped by put: these paths apply writes to
the input channel, not the delta channel, so nothing bumped the delta
channel's version and it never entered new_versions. Pass get_next_version
for the manual bump, the same reason exit mode needs it, and derive
new_versions from the returned checkpoint.

fork_pending tracks what is still owed, mirroring
_delta_channels_awaiting_fork_snapshot in _loop.py.

Caught by the Open SWE review bot on #8548.
2026-09-23 10:42:44 -04:00
Elior Nataf Lackritz 02c91dccf7 fix(langgraph): only fork on the first superstep of a bulk update
perform_superstep returns the config of the checkpoint it just wrote and
bulk_update_state feeds that back in, so from the second superstep on the
incoming config always names a checkpoint whether or not the caller
addressed one. Deriving the fork flag from it made every superstep after
the first force-snapshot every available DeltaChannel and reset its
cadence, storing the whole growing value once per superstep.

Resolve the flag once from the caller's config and pass it explicitly,
true only for the first superstep. The clear-tasks recursion carries it
through, since the checkpoint written there has no delta snapshot and so
leaves a fork unsealed.

Caught by the Open SWE review bot on #8548.
2026-09-23 10:42:44 -04:00
dffb68192d fix(langgraph): don't replay an abandoned branch into a DeltaChannel fork
Addressing an older checkpoint creates a fork: the shared base ends up
with two children and keeps the checkpoint_writes of the branch the fork
abandons. Nothing records which child consumed which write, so the
DeltaChannel ancestor walk collected the abandoned branch's writes too.
Live execution was correct; only the reconstruction after a reload was
wrong, and it was wrong on every saver.

Fixed on the write side, so no saver changes are needed. A run launched
against an explicitly addressed checkpoint forces every DeltaChannel to
snapshot into its first checkpoint, terminating the walk inside the fork
instead of at the shared base. This mirrors the existing force-snapshot
for Overwrite writes, hence the rename to _delta_channels_forced_snapshot.
update_state against an older checkpoint takes the same path, for the
same reason is_fresh_thread already does.

A channel with no value at the fork base cannot carry a snapshot blob
yet, so the request stays queued until the first superstep that gives it
one. Cost is one snapshot per addressed run, not per superstep.

Fixes #8443

Co-Authored-By: AnnaSuSu <64579968+AnnaSuSu@users.noreply.github.com>
Co-Authored-By: UditDewan <194863456+UditDewan@users.noreply.github.com>
2026-09-23 10:42:44 -04:00
8 changed files with 523 additions and 67 deletions
+1 -1
View File
@@ -1 +1 @@
__version__ = "0.4.32"
__version__ = "0.4.31"
+7 -29
View File
@@ -12,7 +12,6 @@ from collections.abc import Callable, Mapping, Sequence
from contextlib import contextmanager
from dataclasses import asdict, dataclass, field
from datetime import datetime, timezone
from functools import partial
from typing import Protocol, TypeVar
import click
@@ -1691,18 +1690,11 @@ OPT_HOST_URL = click.option(
)
OPT_AGENT_ID = click.option(
"--agent-id",
envvar="LANGSMITH_AGENT_ID",
show_envvar=True,
help="Logical agent ID (requires agent mode enabled for the tenant).",
"--agent-id", help="Logical agent ID (requires agent mode enabled for the tenant)."
)
OPT_AGENT_ENVIRONMENT = partial(
click.option,
"--agent-environment",
"environment",
envvar="LANGSMITH_AGENT_ENVIRONMENT",
show_envvar=True,
OPT_AGENT_ENVIRONMENT = click.option(
"--environment",
type=click.Choice(["development", "staging", "production"]),
help="Agent environment (requires agent mode enabled for the tenant).",
)
@@ -1806,9 +1798,7 @@ def _deploy_base_options(
OPT_HOST_API_KEY,
OPT_HOST_DEPLOYMENT_NAME,
OPT_AGENT_ID,
OPT_AGENT_ENVIRONMENT()
if include_docker_args
else OPT_AGENT_ENVIRONMENT(type=str),
OPT_AGENT_ENVIRONMENT,
click.option(
"--deployment-id",
help=(
@@ -1940,12 +1930,6 @@ def deploy(ctx: click.Context, **_: object):
# otherwise, we return None here and click will proceed to actually run the subcommand (list or delete)
if ctx.invoked_subcommand is not None:
return
environment_param = next(
param for param in _deploy_cmd.params if param.name == "environment"
)
ctx.params["environment"] = environment_param.type_cast_value(
ctx, ctx.params["environment"]
)
if (
ctx.params.get("agent_id") is not None
or ctx.params.get("environment") is not None
@@ -1998,14 +1982,13 @@ def _deploy_cmd(
validate_deploy_commands(install_command, build_command)
agent = None
if agent_id is not None or environment is not None:
em.note("Note: --agent-id and --agent-environment flags are in private beta")
if not agent_id or not agent_id.strip() or not environment:
raise click.UsageError(
"--agent-id and --agent-environment are required together."
"--agent-id and --environment are required together."
)
if name is not None or deployment_id is not None:
raise click.UsageError(
"--agent-id and --agent-environment cannot be combined with --name or --deployment-id."
"--agent-id and --environment cannot be combined with --name or --deployment-id."
)
agent = {"agent_id": agent_id, "environment": environment}
if not config.exists():
@@ -2141,7 +2124,7 @@ def _deploy_cmd(
@OPT_HOST_API_KEY
@OPT_HOST_URL
@OPT_AGENT_ID
@OPT_AGENT_ENVIRONMENT()
@OPT_AGENT_ENVIRONMENT
@click.option(
"--name-contains",
default="",
@@ -2155,11 +2138,6 @@ def deploy_list(
agent_id: str | None,
environment: str | None,
) -> None:
if agent_id is not None or environment is not None:
click.secho(
"Note: --agent-id and --agent-environment flags are in private beta",
fg="yellow",
)
if agent_id is not None and not agent_id.strip():
raise click.UsageError("--agent-id must not be empty.")
filters = {}
@@ -58,7 +58,7 @@ AGENT_ARGS = [
"deploy",
"--agent-id",
"customer-support",
"--agent-environment",
"--environment",
"staging",
"--remote",
"--no-wait",
+56 -10
View File
@@ -80,12 +80,14 @@ def get_updated_channels_from_tasks(
def get_delta_channels_from_all_channels(
channels: Mapping[str, BaseChannel],
*,
include_unavailable: bool = False,
) -> set[str]:
"""DeltaChannels to snapshot on the first update_state of a fresh thread."""
"""DeltaChannels to snapshot on the first update_state of a fresh thread or fork."""
return {
k
for k, ch in channels.items()
if isinstance(ch, DeltaChannel) and ch.is_available()
if isinstance(ch, DeltaChannel) and (include_unavailable or ch.is_available())
}
@@ -122,15 +124,22 @@ def create_checkpoint_plan_for_update_state_api(
parents: dict[str, Any],
saved_metadata: Mapping[str, Any] | None,
is_fresh_thread: bool,
is_fork: bool,
) -> tuple[set[str], dict[str, Any]]:
"""Return ``(channels_to_snapshot, metadata)`` for an update_state head."""
"""Return ``(channels_to_snapshot, metadata)`` for an update_state head.
A fork snapshots everything, like a fresh thread: its base also holds the
writes of the branch it abandons, so the ancestor walk must stop here.
"""
metadata: dict[str, Any] = {
"source": "update",
"step": step,
"parents": parents,
}
if is_fresh_thread:
return get_delta_channels_from_all_channels(channels), metadata
if is_fresh_thread or is_fork:
return get_delta_channels_from_all_channels(
channels, include_unavailable=is_fork
), metadata
new_counters = create_metadata_for_update_state_api(
channels,
@@ -146,6 +155,34 @@ def create_checkpoint_plan_for_update_state_api(
return channels_to_snapshot, metadata
def create_fork_checkpoint(
checkpoint: Checkpoint,
channels: Mapping[str, BaseChannel],
step: int,
*,
is_fork: bool,
get_next_version: GetNextVersion,
) -> Checkpoint:
"""``create_checkpoint`` for the update_state paths that skip the plan.
The fork has to be sealed by its first checkpoint: any later superstep
has already rebuilt its delta channels through the shared base. These
paths never write the delta channel, so its version must be bumped here
or ``put`` drops the blob; derive ``new_versions`` from the result.
"""
if not is_fork:
return create_checkpoint(checkpoint, channels, step)
return create_checkpoint(
checkpoint,
channels,
step,
get_next_version=get_next_version,
channels_to_snapshot=get_delta_channels_from_all_channels(
channels, include_unavailable=True
),
)
def create_checkpoint(
checkpoint: Checkpoint,
channels: Mapping[str, BaseChannel] | None,
@@ -174,14 +211,23 @@ def create_checkpoint(
values = {}
channel_versions = dict(checkpoint["channel_versions"])
for k in channels:
if k not in channel_versions:
continue
ch = channels[k]
if k not in channel_versions:
# A forced snapshot of a never-written channel still has to
# land to stop the ancestor walk, and `put` only stores blobs
# for versioned channels.
if k in channels_to_snapshot and get_next_version is not None:
channel_versions[k] = get_next_version(None, None)
values[k] = _DeltaSnapshot(
ch.get() if ch.is_available() else ch.typ()
)
continue
if k in channels_to_snapshot:
# Callers force a full snapshot blob here: exit mode when a
# delta channel reaches its snapshot cadence, and update_state
# on a fresh thread (no ancestor to replay writes from). The
# manual version-bump below only applies to the exit-mode case.
# delta channel reaches its snapshot cadence, update_state on
# a fresh thread (no ancestor to replay writes from), and a
# fork. The manual version-bump below only applies to the
# exit-mode case.
#
# In exit mode, the snapshot decision is deferred to exit
# time (intermediate steps have do_checkpoint=False). The
+22 -11
View File
@@ -222,10 +222,16 @@ class PregelLoop:
# under the saver's `ORDER BY task_id, idx` sorting.
_exit_delta_writes: list[tuple[int, str, str, Any]] | None = None
# Delta channels that saw an Overwrite since the last checkpoint. These
# channels must snapshot after live update applies overwrite semantics so
# sparse replay starts from the same post-overwrite value.
_delta_channels_with_overwrite: set[str]
# Delta channels that must snapshot at the next checkpoint, whatever their
# cadence counters say:
# * an Overwrite arrived since the last checkpoint, so sparse replay has to
# start from the post-overwrite value;
# * this run forked off an explicitly addressed checkpoint. That base also
# holds the writes of the branch the fork abandons, and nothing records
# which child consumed which, so the ancestor walk must stop inside the
# fork. Any addressed checkpoint counts, because telling a real fork
# apart would mean trusting the base's `pending_writes` to be complete.
_delta_channels_forced_snapshot: set[str]
# The checkpoint_config that points at the parent loaded at `__enter__`
# (or the synthetic-empty checkpoint, on first run). We capture it
@@ -369,6 +375,13 @@ class PregelLoop:
if self.config[CONF].get(CONFIG_KEY_CHECKPOINT_NS)
else ()
)
# Value, not key presence like `is_replaying`: subgraph task configs
# always carry an explicit `None` checkpoint_id.
self._delta_channels_forced_snapshot = (
{k for k, spec in specs.items() if isinstance(spec, DeltaChannel)}
if self.checkpoint_config[CONF].get(CONFIG_KEY_CHECKPOINT_ID)
else set()
)
self.prev_checkpoint_config = None
runtime = self.config[CONF].get(CONFIG_KEY_RUNTIME)
self.control = runtime.control if isinstance(runtime, Runtime) else None
@@ -683,7 +696,7 @@ class PregelLoop:
def after_tick(self) -> None:
# finish superstep
writes = [w for t in self.tasks.values() for w in t.writes]
self._delta_channels_with_overwrite.update(
self._delta_channels_forced_snapshot.update(
ch
for ch, v in writes
if isinstance(self.specs.get(ch), DeltaChannel) and _get_overwrite(v)[0]
@@ -991,7 +1004,7 @@ class PregelLoop:
manager=None,
updated_channels=updated_channels,
)
self._delta_channels_with_overwrite.update(
self._delta_channels_forced_snapshot.update(
c
for c, v in input_writes
if isinstance(self.specs.get(c), DeltaChannel) and _get_overwrite(v)[0]
@@ -1136,7 +1149,7 @@ class PregelLoop:
# create new checkpoint
channels_to_snapshot = (
delta_channels_to_snapshot(self.channels, new_counters)
| self._delta_channels_with_overwrite
| self._delta_channels_forced_snapshot
if do_checkpoint
else set()
)
@@ -1154,7 +1167,7 @@ class PregelLoop:
for k in channels_to_snapshot:
new_counters[k] = (0, 0)
if do_checkpoint:
self._delta_channels_with_overwrite.difference_update(channels_to_snapshot)
self._delta_channels_forced_snapshot.difference_update(channels_to_snapshot)
non_zero = {k: v for k, v in new_counters.items() if v != (0, 0)}
if non_zero:
self.checkpoint_metadata["counters_since_delta_snapshot"] = non_zero
@@ -1239,7 +1252,7 @@ class PregelLoop:
)
channels_to_snapshot = (
delta_channels_to_snapshot(self.channels, counters)
| self._delta_channels_with_overwrite
| self._delta_channels_forced_snapshot
)
pending = [
@@ -1684,7 +1697,6 @@ class SyncPregelLoop(PregelLoop, AbstractContextManager):
)
self._delta_write_futs = []
self._error_handler_write_futs = []
self._delta_channels_with_overwrite = set()
self._exit_delta_writes = (
[] if self.durability == "exit" and self.checkpointer is not None else None
)
@@ -1942,7 +1954,6 @@ class AsyncPregelLoop(PregelLoop, AbstractAsyncContextManager):
)
self._delta_write_futs = []
self._error_handler_write_futs = []
self._delta_channels_with_overwrite = set()
self._exit_delta_writes = (
[] if self.durability == "exit" and self.checkpointer is not None else None
)
+85 -12
View File
@@ -108,6 +108,7 @@ from langgraph.callbacks import (
get_sync_graph_callback_manager_for_config,
)
from langgraph.channels.base import BaseChannel
from langgraph.channels.delta import DeltaChannel
from langgraph.channels.topic import Topic
from langgraph.config import get_config
from langgraph.constants import END
@@ -133,6 +134,7 @@ from langgraph.pregel._checkpoint import (
copy_checkpoint,
create_checkpoint,
create_checkpoint_plan_for_update_state_api,
create_fork_checkpoint,
empty_checkpoint,
get_updated_channels_from_tasks,
)
@@ -1637,8 +1639,21 @@ class Pregel(
else:
raise ValueError(f"Subgraph {recast} not found")
# Read once from the caller's config: every later superstep receives
# the config of the checkpoint just written, which always names one.
# Cleared by the first checkpoint that carries the snapshots, which
# `__copy__` does not write.
fork_pending: set[str] = (
{k for k, v in self.channels.items() if isinstance(v, DeltaChannel)}
if config[CONF].get(CONFIG_KEY_CHECKPOINT_ID)
else set()
)
def perform_superstep(
input_config: RunnableConfig, updates: Sequence[StateUpdate]
input_config: RunnableConfig,
updates: Sequence[StateUpdate],
*,
is_fork: bool,
) -> RunnableConfig:
# get last checkpoint
config = ensure_config(self.config, input_config)
@@ -1726,9 +1741,17 @@ class Pregel(
self.trigger_to_nodes,
)
# save checkpoint
next_checkpoint = create_fork_checkpoint(
checkpoint,
channels,
step,
is_fork=is_fork,
get_next_version=checkpointer.get_next_version,
)
fork_pending.difference_update(next_checkpoint["channel_values"])
next_config = checkpointer.put(
checkpoint_config,
create_checkpoint(checkpoint, channels, step),
next_checkpoint,
{
"source": "update",
"step": step + 1,
@@ -1736,7 +1759,7 @@ class Pregel(
},
get_new_channel_versions(
checkpoint_previous_versions,
checkpoint["channel_versions"],
next_checkpoint["channel_versions"],
),
)
return patch_checkpoint_map(
@@ -1765,9 +1788,17 @@ class Pregel(
if saved and saved.metadata.get("step") is not None
else -1
)
next_checkpoint = create_fork_checkpoint(
checkpoint,
channels,
next_step,
is_fork=is_fork,
get_next_version=checkpointer.get_next_version,
)
fork_pending.difference_update(next_checkpoint["channel_values"])
next_config = checkpointer.put(
checkpoint_config,
create_checkpoint(checkpoint, channels, next_step),
next_checkpoint,
{
"source": "input",
"step": next_step,
@@ -1777,7 +1808,7 @@ class Pregel(
},
get_new_channel_versions(
checkpoint_previous_versions,
checkpoint["channel_versions"],
next_checkpoint["channel_versions"],
),
)
@@ -1873,6 +1904,7 @@ class Pregel(
return perform_superstep(
patch_checkpoint_map(next_config, saved.metadata),
[item for lst in user_group_by.values() for item in lst],
is_fork=is_fork,
)
return patch_checkpoint_map(next_config, saved.metadata)
@@ -2020,6 +2052,7 @@ class Pregel(
parents=saved.metadata.get("parents", {}) if saved else {},
saved_metadata=saved.metadata if saved else None,
is_fresh_thread=saved is None,
is_fork=is_fork,
)
)
checkpoint = create_checkpoint(
@@ -2032,6 +2065,8 @@ class Pregel(
else None,
channels_to_snapshot=channels_to_snapshot,
)
if is_fork:
fork_pending.difference_update(checkpoint["channel_values"])
next_config = checkpointer.put(
checkpoint_config,
checkpoint,
@@ -2050,7 +2085,9 @@ class Pregel(
config, {CONFIG_KEY_THREAD_ID: str(config[CONF][CONFIG_KEY_THREAD_ID])}
)
for superstep in supersteps:
current_config = perform_superstep(current_config, superstep)
current_config = perform_superstep(
current_config, superstep, is_fork=bool(fork_pending)
)
return current_config
async def abulk_update_state(
@@ -2103,8 +2140,21 @@ class Pregel(
else:
raise ValueError(f"Subgraph {recast} not found")
# Read once from the caller's config: every later superstep receives
# the config of the checkpoint just written, which always names one.
# Cleared by the first checkpoint that carries the snapshots, which
# `__copy__` does not write.
fork_pending: set[str] = (
{k for k, v in self.channels.items() if isinstance(v, DeltaChannel)}
if config[CONF].get(CONFIG_KEY_CHECKPOINT_ID)
else set()
)
async def aperform_superstep(
input_config: RunnableConfig, updates: Sequence[StateUpdate]
input_config: RunnableConfig,
updates: Sequence[StateUpdate],
*,
is_fork: bool,
) -> RunnableConfig:
# get last checkpoint
config = ensure_config(self.config, input_config)
@@ -2190,16 +2240,25 @@ class Pregel(
self.trigger_to_nodes,
)
# save checkpoint
next_checkpoint = create_fork_checkpoint(
checkpoint,
channels,
step,
is_fork=is_fork,
get_next_version=checkpointer.get_next_version,
)
fork_pending.difference_update(next_checkpoint["channel_values"])
next_config = await checkpointer.aput(
checkpoint_config,
create_checkpoint(checkpoint, channels, step),
next_checkpoint,
{
"source": "update",
"step": step + 1,
"parents": saved.metadata.get("parents", {}) if saved else {},
},
get_new_channel_versions(
checkpoint_previous_versions, checkpoint["channel_versions"]
checkpoint_previous_versions,
next_checkpoint["channel_versions"],
),
)
return patch_checkpoint_map(
@@ -2228,9 +2287,17 @@ class Pregel(
if saved and saved.metadata.get("step") is not None
else -1
)
next_checkpoint = create_fork_checkpoint(
checkpoint,
channels,
next_step,
is_fork=is_fork,
get_next_version=checkpointer.get_next_version,
)
fork_pending.difference_update(next_checkpoint["channel_values"])
next_config = await checkpointer.aput(
checkpoint_config,
create_checkpoint(checkpoint, channels, next_step),
next_checkpoint,
{
"source": "input",
"step": next_step,
@@ -2240,7 +2307,7 @@ class Pregel(
},
get_new_channel_versions(
checkpoint_previous_versions,
checkpoint["channel_versions"],
next_checkpoint["channel_versions"],
),
)
@@ -2335,6 +2402,7 @@ class Pregel(
return await aperform_superstep(
patch_checkpoint_map(next_config, saved.metadata),
[item for lst in user_group_by.values() for item in lst],
is_fork=is_fork,
)
return patch_checkpoint_map(
@@ -2480,6 +2548,7 @@ class Pregel(
parents=saved.metadata.get("parents", {}) if saved else {},
saved_metadata=saved.metadata if saved else None,
is_fresh_thread=saved is None,
is_fork=is_fork,
)
)
checkpoint = create_checkpoint(
@@ -2492,6 +2561,8 @@ class Pregel(
else None,
channels_to_snapshot=channels_to_snapshot,
)
if is_fork:
fork_pending.difference_update(checkpoint["channel_values"])
next_config = await checkpointer.aput(
checkpoint_config,
checkpoint,
@@ -2509,7 +2580,9 @@ class Pregel(
config, {CONFIG_KEY_THREAD_ID: str(config[CONF][CONFIG_KEY_THREAD_ID])}
)
for superstep in supersteps:
current_config = await aperform_superstep(current_config, superstep)
current_config = await aperform_superstep(
current_config, superstep, is_fork=bool(fork_pending)
)
return current_config
def update_state(
+5 -3
View File
@@ -85,11 +85,13 @@ class MemorySaverAssertImmutable(InMemorySaver):
)
== saved
), config["configurable"]["checkpoint_ns"]
next_config = super().put(config, checkpoint, metadata, new_versions)
# Read back, not the object handed in: a DeltaChannel a step did not
# write is refilled on read from the blob its inherited version points at.
self.storage_for_copies[thread_id][checkpoint_ns][checkpoint["id"]] = (
self.serde.dumps_typed(checkpoint)
self.serde.dumps_typed(super().get(next_config))
)
# call super to write checkpoint
return super().put(config, checkpoint, metadata, new_versions)
return next_config
class MemorySaverNoPending(InMemorySaver):
@@ -0,0 +1,346 @@
"""Forking a thread must not replay the abandoned branch into the fork.
Every graph carries a ``DeltaChannel`` and a plain reducer channel fed the same
values; the plain channel needs no replay, so it is the oracle.
"""
from collections.abc import Sequence
from operator import add
from typing import Annotated, Any
import pytest
from langchain_core.runnables import RunnableConfig
from langgraph.checkpoint.base import BaseCheckpointSaver
from langgraph.checkpoint.serde.types import _DeltaSnapshot
from typing_extensions import TypedDict
from langgraph._internal._constants import INPUT
from langgraph.channels.delta import DeltaChannel
from langgraph.graph import END, StateGraph
from langgraph.types import Durability, StateSnapshot, StateUpdate
pytestmark = pytest.mark.anyio
def _append(current: list | None, writes: Sequence[Any]) -> list:
out = list(current or [])
for write in writes:
out.extend(write if isinstance(write, list) else [write])
return out
class _State(TypedDict):
log: Annotated[list, DeltaChannel(_append, snapshot_frequency=1000)]
plain: Annotated[list, add]
other: Annotated[list, add]
def _build(checkpointer: BaseCheckpointSaver, tag: str) -> Any:
def node(state: _State) -> dict:
return {"log": [f"{tag}-out"], "plain": [f"{tag}-out"]}
builder = StateGraph(_State)
builder.add_node("n", node)
builder.set_entry_point("n")
builder.set_finish_point("n")
return builder.compile(checkpointer=checkpointer)
def _build_without_delta_writes(checkpointer: BaseCheckpointSaver, tag: str) -> Any:
def node(state: _State) -> dict:
return {"other": [f"{tag}-other"]}
builder = StateGraph(_State)
builder.add_node("n", node)
builder.set_entry_point("n")
builder.set_finish_point("n")
return builder.compile(checkpointer=checkpointer)
def _thread(thread_id: str) -> RunnableConfig:
return {"configurable": {"thread_id": thread_id}}
def _at(config: RunnableConfig, snapshot: StateSnapshot) -> RunnableConfig:
return {
"configurable": {
**config["configurable"],
"checkpoint_ns": "",
"checkpoint_id": snapshot.config["configurable"]["checkpoint_id"],
}
}
def _input(marker: str) -> dict:
return {"log": [marker], "plain": [marker]}
def _snapshotted_checkpoints(
checkpointer: BaseCheckpointSaver, config: RunnableConfig
) -> list[str]:
return [
tuple_.config["configurable"]["checkpoint_id"]
for tuple_ in checkpointer.list(config)
if isinstance(tuple_.checkpoint["channel_values"].get("log"), _DeltaSnapshot)
]
def _assert_fork_is_clean(state: StateSnapshot, abandoned: str) -> None:
assert state.values["log"] == state.values["plain"], (
f"delta channel diverged from the plain channel: "
f"{state.values['log']} != {state.values['plain']}"
)
assert abandoned not in state.values["log"], (
f"{abandoned!r} belongs to the branch the fork replaced, "
f"but was replayed into {state.values['log']}"
)
def test_fork_by_invoke(
sync_checkpointer: BaseCheckpointSaver, durability: Durability
) -> None:
config = _thread("t")
_build(sync_checkpointer, "first").invoke(
_input("in-1"), config, durability=durability
)
graph = _build(sync_checkpointer, "second")
graph.invoke(_input("in-2"), config, durability=durability)
abandoned_head = graph.get_state(config)
base = next(
snapshot
for snapshot in graph.get_state_history(config)
if "in-2" not in snapshot.values["log"]
)
_build(sync_checkpointer, "third").invoke(
_input("in-3"), _at(config, base), durability=durability
)
state = graph.get_state(config)
_assert_fork_is_clean(state, "in-2")
assert state.values["log"] == [*base.values["log"], "in-3", "third-out"]
abandoned = graph.get_state(abandoned_head.config).values
assert abandoned["log"] == abandoned["plain"] == abandoned_head.values["log"]
async def test_afork_by_invoke(
async_checkpointer: BaseCheckpointSaver, durability: Durability
) -> None:
config = _thread("t")
await _build(async_checkpointer, "first").ainvoke(
_input("in-1"), config, durability=durability
)
graph = _build(async_checkpointer, "second")
await graph.ainvoke(_input("in-2"), config, durability=durability)
abandoned_head = await graph.aget_state(config)
base = await anext(
snapshot
async for snapshot in graph.aget_state_history(config)
if "in-2" not in snapshot.values["log"]
)
await _build(async_checkpointer, "third").ainvoke(
_input("in-3"), _at(config, base), durability=durability
)
state = await graph.aget_state(config)
_assert_fork_is_clean(state, "in-2")
assert state.values["log"] == [*base.values["log"], "in-3", "third-out"]
abandoned = (await graph.aget_state(abandoned_head.config)).values
assert abandoned["log"] == abandoned["plain"] == abandoned_head.values["log"]
def test_fork_off_checkpoint_before_first_input(
sync_checkpointer: BaseCheckpointSaver, durability: Durability
) -> None:
config = _thread("t")
graph = _build(sync_checkpointer, "first")
graph.invoke(_input("in-1"), config, durability=durability)
root = list(graph.get_state_history(config))[-1]
assert root.values["log"] == []
_build(sync_checkpointer, "third").invoke(
_input("in-9"), _at(config, root), durability=durability
)
state = graph.get_state(config)
_assert_fork_is_clean(state, "in-1")
assert state.values["log"] == ["in-9", "third-out"]
async def test_afork_off_checkpoint_before_first_input(
async_checkpointer: BaseCheckpointSaver, durability: Durability
) -> None:
config = _thread("t")
graph = _build(async_checkpointer, "first")
await graph.ainvoke(_input("in-1"), config, durability=durability)
root = [snapshot async for snapshot in graph.aget_state_history(config)][-1]
assert root.values["log"] == []
await _build(async_checkpointer, "third").ainvoke(
_input("in-9"), _at(config, root), durability=durability
)
state = await graph.aget_state(config)
_assert_fork_is_clean(state, "in-1")
assert state.values["log"] == ["in-9", "third-out"]
def test_fork_by_update_state(sync_checkpointer: BaseCheckpointSaver) -> None:
config = _thread("t")
_build(sync_checkpointer, "first").invoke(_input("in-1"), config)
graph = _build(sync_checkpointer, "second")
graph.invoke(_input("in-2"), config)
base = next(
snapshot
for snapshot in graph.get_state_history(config)
if "in-2" not in snapshot.values["log"]
)
forked = graph.update_state(_at(config, base), _input("patched"))
state = graph.get_state(forked)
_assert_fork_is_clean(state, "in-2")
assert state.values["log"] == [*base.values["log"], "patched"]
async def test_afork_by_update_state(
async_checkpointer: BaseCheckpointSaver,
) -> None:
config = _thread("t")
await _build(async_checkpointer, "first").ainvoke(_input("in-1"), config)
graph = _build(async_checkpointer, "second")
await graph.ainvoke(_input("in-2"), config)
base = await anext(
snapshot
async for snapshot in graph.aget_state_history(config)
if "in-2" not in snapshot.values["log"]
)
forked = await graph.aupdate_state(_at(config, base), _input("patched"))
state = await graph.aget_state(forked)
_assert_fork_is_clean(state, "in-2")
assert state.values["log"] == [*base.values["log"], "patched"]
def test_unaddressed_run_keeps_snapshot_cadence(
sync_checkpointer: BaseCheckpointSaver, durability: Durability
) -> None:
config = _thread("t")
graph = _build(sync_checkpointer, "first")
graph.invoke(_input("in-1"), config, durability=durability)
graph.invoke(_input("in-2"), config, durability=durability)
assert not _snapshotted_checkpoints(sync_checkpointer, config)
def test_fork_before_first_value_when_fork_never_writes_the_channel(
sync_checkpointer: BaseCheckpointSaver, durability: Durability
) -> None:
config = _thread("t")
graph = _build(sync_checkpointer, "first")
graph.invoke(_input("in-1"), config, durability=durability)
root = list(graph.get_state_history(config))[-1]
assert root.values["log"] == []
_build_without_delta_writes(sync_checkpointer, "third").invoke(
{"other": ["in-9"]}, _at(config, root), durability=durability
)
state = graph.get_state(config)
_assert_fork_is_clean(state, "in-1")
assert state.values["log"] == []
async def test_afork_before_first_value_when_fork_never_writes_the_channel(
async_checkpointer: BaseCheckpointSaver, durability: Durability
) -> None:
config = _thread("t")
graph = _build(async_checkpointer, "first")
await graph.ainvoke(_input("in-1"), config, durability=durability)
root = [snapshot async for snapshot in graph.aget_state_history(config)][-1]
assert root.values["log"] == []
await _build_without_delta_writes(async_checkpointer, "third").ainvoke(
{"other": ["in-9"]}, _at(config, root), durability=durability
)
state = await graph.aget_state(config)
_assert_fork_is_clean(state, "in-1")
assert state.values["log"] == []
def test_fork_before_first_value_by_bulk_update(
sync_checkpointer: BaseCheckpointSaver,
) -> None:
config = _thread("t")
graph = _build(sync_checkpointer, "first")
graph.invoke(_input("in-1"), config)
root = list(graph.get_state_history(config))[-1]
assert root.values["log"] == []
forked = graph.bulk_update_state(
_at(config, root),
[
[StateUpdate({"other": ["s1"]}, "n")],
[StateUpdate(_input("s2"), "n")],
],
)
state = graph.get_state(forked)
_assert_fork_is_clean(state, "in-1")
assert state.values["log"] == ["s2"]
@pytest.mark.parametrize("first_as_node", [INPUT, END, "__copy__"])
def test_fork_by_bulk_update_whose_first_superstep_skips_the_plan(
sync_checkpointer: BaseCheckpointSaver, first_as_node: str
) -> None:
config = _thread("t")
_build(sync_checkpointer, "first").invoke(_input("in-1"), config)
graph = _build(sync_checkpointer, "second")
graph.invoke(_input("in-2"), config)
base = next(
snapshot
for snapshot in graph.get_state_history(config)
if "in-2" not in snapshot.values["log"]
)
first = (
StateUpdate(_input("first-step"), first_as_node)
if first_as_node == INPUT
else StateUpdate(None, first_as_node)
)
forked = graph.bulk_update_state(
_at(config, base),
[[first], [StateUpdate(_input("second-step"), "n")]],
)
state = graph.get_state(forked)
assert state.values["log"] == state.values["plain"], (
f"delta channel diverged from the plain channel: "
f"{state.values['log']} != {state.values['plain']}"
)
def test_unaddressed_bulk_update_keeps_snapshot_cadence(
sync_checkpointer: BaseCheckpointSaver,
) -> None:
config = _thread("t")
graph = _build(sync_checkpointer, "first")
graph.invoke(_input("in-1"), config)
graph.bulk_update_state(
config,
[[StateUpdate(_input(f"u{i}"), "n")] for i in range(4)],
)
assert not _snapshotted_checkpoints(sync_checkpointer, config)