langgraph-api 0.4.48__py3-none-any.whl → 0.5.1__py3-none-any.whl

langgraph_api/__init__.py

@@ -1 +1 @@
-__version__ = "0.4.48"
+__version__ = "0.5.1"

langgraph_api/api/assistants.py

@@ -1,3 +1,4 @@
+from functools import partial
 from typing import Any
 from uuid import uuid4
 
@@ -37,7 +38,7 @@ from langgraph_api.validation import (
     ConfigValidator,
 )
 from langgraph_runtime.checkpoint import Checkpointer
-from langgraph_runtime.database import connect
+from langgraph_runtime.database import connect as base_connect
 from langgraph_runtime.ops import Assistants
 from langgraph_runtime.retry import retry_db
 
@@ -45,6 +46,8 @@ logger = structlog.stdlib.get_logger(__name__)
 
 CrudAssistants = GrpcAssistants if FF_USE_CORE_API else Assistants
 
+connect = partial(base_connect, supports_core_api=FF_USE_CORE_API)
+
 EXCLUDED_CONFIG_SCHEMA = (
     "__pregel_checkpointer",
     "__pregel_store",
@@ -255,7 +258,7 @@ async def get_assistant_graph(
     assistant_id = get_assistant_id(str(request.path_params["assistant_id"]))
     validate_uuid(assistant_id, "Invalid assistant ID: must be a UUID")
     async with connect() as conn:
-        assistant_ = await Assistants.get(conn, assistant_id)
+        assistant_ = await CrudAssistants.get(conn, assistant_id)
         assistant = await fetchone(assistant_)
     config = json_loads(assistant["config"])
     configurable = config.setdefault("configurable", {})
@@ -312,43 +315,44 @@ async def get_assistant_subgraphs(
     assistant_id = request.path_params["assistant_id"]
     validate_uuid(assistant_id, "Invalid assistant ID: must be a UUID")
     async with connect() as conn:
-        assistant_ = await Assistants.get(conn, assistant_id)
+        assistant_ = await CrudAssistants.get(conn, assistant_id)
         assistant = await fetchone(assistant_)
-        config = json_loads(assistant["config"])
-        configurable = config.setdefault("configurable", {})
-        configurable.update(get_configurable_headers(request.headers))
-        async with get_graph(
-            assistant["graph_id"],
-            config,
-            checkpointer=Checkpointer(),
-            store=(await api_store.get_store()),
-        ) as graph:
-            namespace = request.path_params.get("namespace")
-
-            if isinstance(graph, BaseRemotePregel):
-                return ApiResponse(
-                    await graph.fetch_subgraphs(
+
+    config = json_loads(assistant["config"])
+    configurable = config.setdefault("configurable", {})
+    configurable.update(get_configurable_headers(request.headers))
+    async with get_graph(
+        assistant["graph_id"],
+        config,
+        checkpointer=Checkpointer(),
+        store=(await api_store.get_store()),
+    ) as graph:
+        namespace = request.path_params.get("namespace")
+
+        if isinstance(graph, BaseRemotePregel):
+            return ApiResponse(
+                await graph.fetch_subgraphs(
+                    namespace=namespace,
+                    recurse=request.query_params.get("recurse", "False")
+                    in ("true", "True"),
+                )
+            )
+
+        try:
+            return ApiResponse(
+                {
+                    ns: _graph_schemas(subgraph)
+                    async for ns, subgraph in graph.aget_subgraphs(
                         namespace=namespace,
                         recurse=request.query_params.get("recurse", "False")
                         in ("true", "True"),
                     )
-                )
-
-            try:
-                return ApiResponse(
-                    {
-                        ns: _graph_schemas(subgraph)
-                        async for ns, subgraph in graph.aget_subgraphs(
-                            namespace=namespace,
-                            recurse=request.query_params.get("recurse", "False")
-                            in ("true", "True"),
-                        )
-                    }
-                )
-            except NotImplementedError:
-                raise HTTPException(
-                    422, detail="The graph does not support visualization"
-                ) from None
+                }
+            )
+        except NotImplementedError:
+            raise HTTPException(
+                422, detail="The graph does not support visualization"
+            ) from None
 
 
 @retry_db
@@ -359,40 +363,40 @@ async def get_assistant_schemas(
     assistant_id = request.path_params["assistant_id"]
     validate_uuid(assistant_id, "Invalid assistant ID: must be a UUID")
     async with connect() as conn:
-        assistant_ = await Assistants.get(conn, assistant_id)
-        # TODO Implementa  cache so we can de-dent and release this connection.
+        assistant_ = await CrudAssistants.get(conn, assistant_id)
         assistant = await fetchone(assistant_)
-        config = json_loads(assistant["config"])
-        configurable = config.setdefault("configurable", {})
-        configurable.update(get_configurable_headers(request.headers))
-        async with get_graph(
-            assistant["graph_id"],
-            config,
-            checkpointer=Checkpointer(),
-            store=(await api_store.get_store()),
-        ) as graph:
-            if isinstance(graph, BaseRemotePregel):
-                schemas = await graph.fetch_state_schema()
-                return ApiResponse(
-                    {
-                        "graph_id": assistant["graph_id"],
-                        "input_schema": schemas.get("input"),
-                        "output_schema": schemas.get("output"),
-                        "state_schema": schemas.get("state"),
-                        "config_schema": schemas.get("config"),
-                        "context_schema": schemas.get("context"),
-                    }
-                )
-
-            schemas = _graph_schemas(graph)
 
+    config = json_loads(assistant["config"])
+    configurable = config.setdefault("configurable", {})
+    configurable.update(get_configurable_headers(request.headers))
+    async with get_graph(
+        assistant["graph_id"],
+        config,
+        checkpointer=Checkpointer(),
+        store=(await api_store.get_store()),
+    ) as graph:
+        if isinstance(graph, BaseRemotePregel):
+            schemas = await graph.fetch_state_schema()
             return ApiResponse(
                 {
                     "graph_id": assistant["graph_id"],
-                    **schemas,
+                    "input_schema": schemas.get("input"),
+                    "output_schema": schemas.get("output"),
+                    "state_schema": schemas.get("state"),
+                    "config_schema": schemas.get("config"),
+                    "context_schema": schemas.get("context"),
                 }
             )
 
+        schemas = _graph_schemas(graph)
+
+        return ApiResponse(
+            {
+                "graph_id": assistant["graph_id"],
+                **schemas,
+            }
+        )
+
 
 @retry_db
 async def patch_assistant(

langgraph_api/config.py

@@ -128,6 +128,45 @@ class StoreConfig(TypedDict, total=False):
     ttl: TTLConfig
 
 
+class SerdeConfig(TypedDict, total=False):
+    """Configuration for the built-in serde, which handles checkpointing of state.
+
+    If omitted, no serde is set up (the object store will still be present, however)."""
+
+    allowed_json_modules: list[list[str]] | Literal[True] | None
+    """Optional. List of allowed python modules to de-serialize custom objects from.
+    
+    If provided, only the specified modules will be allowed to be deserialized.
+    If omitted, no modules are allowed, and the object returned will simply be a json object OR
+    a deserialized langchain object.
+    
+    Example:
+    {...
+        "serde": {
+            "allowed_json_modules": [
+                ["my_agent", "my_file", "SomeType"],
+            ]
+        }
+    }
+
+    If you set this to True, any module will be allowed to be deserialized.
+
+    Example:
+    {...
+        "serde": {
+            "allowed_json_modules": true
+        }
+    }
+    
+    """
+    pickle_fallback: bool
+    """Optional. Whether to allow pickling as a fallback for deserialization.
+    
+    If True, pickling will be allowed as a fallback for deserialization.
+    If False, pickling will not be allowed as a fallback for deserialization.
+    Defaults to True if not configured."""
+
+
 class CheckpointerConfig(TypedDict, total=False):
     """Configuration for the built-in checkpointer, which handles checkpointing of state.
 
@@ -140,6 +179,8 @@ class CheckpointerConfig(TypedDict, total=False):
     If provided, the checkpointer will apply TTL settings according to the configuration.
     If omitted, no TTL behavior is configured.
     """
+    serde: SerdeConfig | None
+    """Optional. Defines the configuration for how checkpoints are serialized."""
 
 
 class SecurityConfig(TypedDict, total=False):
@@ -250,6 +291,13 @@ ALLOW_PRIVATE_NETWORK = env("ALLOW_PRIVATE_NETWORK", cast=bool, default=False)
 See https://developer.chrome.com/blog/private-network-access-update-2024-03
 """
 
+# gRPC client pool size for persistence server.
+GRPC_CLIENT_POOL_SIZE = env("GRPC_CLIENT_POOL_SIZE", cast=int, default=5)
+
+# Minimum payload size to use the dedicated thread pool for JSON parsing.
+# (Otherwise, the payload is parsed directly in the event loop.)
+JSON_THREAD_POOL_MINIMUM_SIZE_BYTES = 100 * 1024  # 100 KB
+
 HTTP_CONFIG = env("LANGGRAPH_HTTP", cast=_parse_schema(HttpConfig), default=None)
 STORE_CONFIG = env("LANGGRAPH_STORE", cast=_parse_schema(StoreConfig), default=None)
 
@@ -339,6 +387,11 @@ def _parse_thread_ttl(value: str | None) -> ThreadTTLConfig | None:
 CHECKPOINTER_CONFIG = env(
     "LANGGRAPH_CHECKPOINTER", cast=_parse_schema(CheckpointerConfig), default=None
 )
+SERDE: SerdeConfig | None = (
+    CHECKPOINTER_CONFIG["serde"]
+    if CHECKPOINTER_CONFIG and "serde" in CHECKPOINTER_CONFIG
+    else None
+)
 THREAD_TTL: ThreadTTLConfig | None = env(
     "LANGGRAPH_THREAD_TTL", cast=_parse_thread_ttl, default=None
 )
@@ -349,7 +402,6 @@ N_JOBS_PER_WORKER = env("N_JOBS_PER_WORKER", cast=int, default=10)
 BG_JOB_TIMEOUT_SECS = env("BG_JOB_TIMEOUT_SECS", cast=float, default=3600)
 
 FF_CRONS_ENABLED = env("FF_CRONS_ENABLED", cast=bool, default=True)
-FF_RICH_THREADS = env("FF_RICH_THREADS", cast=bool, default=True)
 FF_LOG_DROPPED_EVENTS = env("FF_LOG_DROPPED_EVENTS", cast=bool, default=False)
 FF_LOG_QUERY_AND_PARAMS = env("FF_LOG_QUERY_AND_PARAMS", cast=bool, default=False)
 

langgraph_api/grpc_ops/client.py

@@ -1,5 +1,6 @@
 """gRPC client wrapper for LangGraph persistence services."""
 
+import asyncio
 import os
 
 import structlog
@@ -10,6 +11,10 @@ from .generated.core_api_pb2_grpc import AdminStub, AssistantsStub, ThreadsStub
 logger = structlog.stdlib.get_logger(__name__)
 
 
+# Shared global client pool
+_client_pool: "GrpcClientPool | None" = None
+
+
 class GrpcClient:
     """gRPC client for LangGraph persistence services."""
 
@@ -90,3 +95,89 @@ class GrpcClient:
                 "Client not connected. Use async context manager or call connect() first."
             )
         return self._admin_stub
+
+
+class GrpcClientPool:
+    """Pool of gRPC clients for load distribution."""
+
+    def __init__(self, pool_size: int = 5, server_address: str | None = None):
+        self.pool_size = pool_size
+        self.server_address = server_address
+        self.clients: list[GrpcClient] = []
+        self._current_index = 0
+        self._init_lock = asyncio.Lock()
+        self._initialized = False
+
+    async def _initialize(self):
+        """Initialize the pool of clients."""
+        async with self._init_lock:
+            if self._initialized:
+                return
+
+            await logger.ainfo(
+                "Initializing gRPC client pool",
+                pool_size=self.pool_size,
+                server_address=self.server_address,
+            )
+
+            for _ in range(self.pool_size):
+                client = GrpcClient(server_address=self.server_address)
+                await client.connect()
+                self.clients.append(client)
+
+            self._initialized = True
+            await logger.ainfo(
+                f"gRPC client pool initialized with {self.pool_size} clients"
+            )
+
+    async def get_client(self) -> GrpcClient:
+        """Get next client using round-robin selection.
+
+        Round-robin without strict locking - slight races are acceptable
+        and result in good enough distribution under high load.
+        """
+        if not self._initialized:
+            await self._initialize()
+
+        idx = self._current_index % self.pool_size
+        self._current_index = idx + 1
+        return self.clients[idx]
+
+    async def close(self):
+        """Close all clients in the pool."""
+        if self._initialized:
+            await logger.ainfo(f"Closing gRPC client pool ({self.pool_size} clients)")
+            for client in self.clients:
+                await client.close()
+            self.clients.clear()
+            self._initialized = False
+
+
+async def get_shared_client() -> GrpcClient:
+    """Get a gRPC client from the shared pool.
+
+    Uses a pool of channels for better performance under high concurrency.
+    Each channel is a separate TCP connection that can handle ~100-200
+    concurrent streams effectively.
+
+    Returns:
+        A GrpcClient instance from the pool
+    """
+    global _client_pool
+    if _client_pool is None:
+        from langgraph_api import config
+
+        _client_pool = GrpcClientPool(
+            pool_size=config.GRPC_CLIENT_POOL_SIZE,
+            server_address=os.getenv("GRPC_SERVER_ADDRESS"),
+        )
+
+    return await _client_pool.get_client()
+
+
+async def close_shared_client():
+    """Close the shared gRPC client pool."""
+    global _client_pool
+    if _client_pool is not None:
+        await _client_pool.close()
+        _client_pool = None

langgraph_api/grpc_ops/generated/core_api_pb2_grpc.py

@@ -6,7 +6,7 @@ import warnings
 from . import core_api_pb2 as core__api__pb2
 from google.protobuf import empty_pb2 as google_dot_protobuf_dot_empty__pb2
 
-GRPC_GENERATED_VERSION = '1.76.0'
+GRPC_GENERATED_VERSION = '1.75.1'
 GRPC_VERSION = grpc.__version__
 _version_not_supported = False
 
@@ -19,7 +19,7 @@ except ImportError:
 if _version_not_supported:
     raise RuntimeError(
         f'The grpc package installed is at version {GRPC_VERSION},'
-        + ' but the generated code in core_api_pb2_grpc.py depends on'
+        + f' but the generated code in core_api_pb2_grpc.py depends on'
         + f' grpcio>={GRPC_GENERATED_VERSION}.'
         + f' Please upgrade your grpc module to grpcio>={GRPC_GENERATED_VERSION}'
         + f' or downgrade your generated code using grpcio-tools<={GRPC_VERSION}.'

langgraph_api/grpc_ops/ops.py

@@ -32,7 +32,7 @@ from langgraph_api.schema import (
 )
 from langgraph_api.serde import json_dumpb, json_loads
 
-from .client import GrpcClient
+from .client import get_shared_client
 from .generated import core_api_pb2 as pb
 
 GRPC_STATUS_TO_HTTP_STATUS = {
@@ -539,9 +539,8 @@ class Assistants(Authenticated):
             select=select,
         )
 
-        # Make the gRPC call
-        async with GrpcClient() as client:
-            response = await client.assistants.Search(request)
+        client = await get_shared_client()
+        response = await client.assistants.Search(request)
 
         # Convert response to expected format
         assistants = [
@@ -578,9 +577,8 @@ class Assistants(Authenticated):
             filters=auth_filters or {},
         )
 
-        # Make the gRPC call
-        async with GrpcClient() as client:
-            response = await client.assistants.Get(request)
+        client = await get_shared_client()
+        response = await client.assistants.Get(request)
 
         # Convert and yield the result
         assistant = proto_to_assistant(response)
@@ -637,9 +635,8 @@ class Assistants(Authenticated):
             metadata=dict_to_struct(metadata or {}),
         )
 
-        # Make the gRPC call
-        async with GrpcClient() as client:
-            response = await client.assistants.Create(request)
+        client = await get_shared_client()
+        response = await client.assistants.Create(request)
 
         # Convert and yield the result
         assistant = proto_to_assistant(response)
@@ -700,9 +697,8 @@ class Assistants(Authenticated):
         if context:
             request.context.CopyFrom(dict_to_struct(context))
 
-        # Make the gRPC call
-        async with GrpcClient() as client:
-            response = await client.assistants.Patch(request)
+        client = await get_shared_client()
+        response = await client.assistants.Patch(request)
 
         # Convert and yield the result
         assistant = proto_to_assistant(response)
@@ -730,9 +726,8 @@ class Assistants(Authenticated):
             filters=auth_filters or {},
         )
 
-        # Make the gRPC call
-        async with GrpcClient() as client:
-            await client.assistants.Delete(request)
+        client = await get_shared_client()
+        await client.assistants.Delete(request)
 
         # Return the deleted ID
         async def generate_result():
@@ -765,9 +760,8 @@ class Assistants(Authenticated):
             filters=auth_filters or {},
         )
 
-        # Make the gRPC call
-        async with GrpcClient() as client:
-            response = await client.assistants.SetLatest(request)
+        client = await get_shared_client()
+        response = await client.assistants.SetLatest(request)
 
         # Convert and yield the result
         assistant = proto_to_assistant(response)
@@ -803,9 +797,8 @@ class Assistants(Authenticated):
             offset=offset,
         )
 
-        # Make the gRPC call
-        async with GrpcClient() as client:
-            response = await client.assistants.GetVersions(request)
+        client = await get_shared_client()
+        response = await client.assistants.GetVersions(request)
 
         # Convert and yield the results
         async def generate_results():
@@ -849,9 +842,8 @@ class Assistants(Authenticated):
             metadata=dict_to_struct(metadata or {}),
         )
 
-        # Make the gRPC call
-        async with GrpcClient() as client:
-            response = await client.assistants.Count(request)
+        client = await get_shared_client()
+        response = await client.assistants.Count(request)
 
         return int(response.count)
 
@@ -913,24 +905,22 @@ class Threads(Authenticated):
         if ids:
             normalized_ids = [_normalize_uuid(thread_id) for thread_id in ids]
             threads: list[Thread] = []
-            async with GrpcClient() as client:
-                for thread_id in normalized_ids:
-                    request = pb.GetThreadRequest(
-                        thread_id=pb.UUID(value=_normalize_uuid(thread_id)),
-                        filters=auth_filters or {},
-                    )
-                    response = await client.threads.Get(request)
-                    thread = proto_to_thread(response)
-
-                    if status and thread["status"] != status:
-                        continue
-                    if metadata and not _json_contains(thread["metadata"], metadata):
-                        continue
-                    if values and not _json_contains(
-                        thread.get("values") or {}, values
-                    ):
-                        continue
-                    threads.append(thread)
+            client = await get_shared_client()
+            for thread_id in normalized_ids:
+                request = pb.GetThreadRequest(
+                    thread_id=pb.UUID(value=_normalize_uuid(thread_id)),
+                    filters=auth_filters or {},
+                )
+                response = await client.threads.Get(request)
+                thread = proto_to_thread(response)
+
+                if status and thread["status"] != status:
+                    continue
+                if metadata and not _json_contains(thread["metadata"], metadata):
+                    continue
+                if values and not _json_contains(thread.get("values") or {}, values):
+                    continue
+                threads.append(thread)
 
             total = len(threads)
             paginated = threads[offset : offset + limit]
@@ -964,10 +954,10 @@ class Threads(Authenticated):
         if select:
             request_kwargs["select"] = select
 
-        async with GrpcClient() as client:
-            response = await client.threads.Search(
-                pb.SearchThreadsRequest(**request_kwargs)
-            )
+        client = await get_shared_client()
+        response = await client.threads.Search(
+            pb.SearchThreadsRequest(**request_kwargs)
+        )
 
         threads = [proto_to_thread(thread) for thread in response.threads]
         cursor = offset + limit if len(threads) == limit else None
@@ -1014,10 +1004,8 @@ class Threads(Authenticated):
                 )
             request_kwargs["status"] = mapped_status
 
-        async with GrpcClient() as client:
-            response = await client.threads.Count(
-                pb.CountThreadsRequest(**request_kwargs)
-            )
+        client = await get_shared_client()
+        response = await client.threads.Count(pb.CountThreadsRequest(**request_kwargs))
 
         return int(response.count)
 
@@ -1035,8 +1023,8 @@ class Threads(Authenticated):
             thread_id=pb.UUID(value=_normalize_uuid(thread_id)),
             filters=auth_filters or {},
         )
-        async with GrpcClient() as client:
-            response = await client.threads.Get(request)
+        client = await get_shared_client()
+        response = await client.threads.Get(request)
 
         thread = proto_to_thread(response)
 
@@ -1077,8 +1065,8 @@ class Threads(Authenticated):
         if ttl_config is not None:
             request.ttl.CopyFrom(ttl_config)
 
-        async with GrpcClient() as client:
-            response = await client.threads.Create(request)
+        client = await get_shared_client()
+        response = await client.threads.Create(request)
         thread = proto_to_thread(response)
 
         async def generate_result():
@@ -1118,8 +1106,8 @@ class Threads(Authenticated):
         if ttl_config is not None:
             request.ttl.CopyFrom(ttl_config)
 
-        async with GrpcClient() as client:
-            response = await client.threads.Patch(request)
+        client = await get_shared_client()
+        response = await client.threads.Patch(request)
 
         thread = proto_to_thread(response)
 
@@ -1147,8 +1135,8 @@ class Threads(Authenticated):
             filters=auth_filters or {},
         )
 
-        async with GrpcClient() as client:
-            response = await client.threads.Delete(request)
+        client = await get_shared_client()
+        response = await client.threads.Delete(request)
 
         deleted_id = UUID(response.value)
 
@@ -1176,8 +1164,8 @@ class Threads(Authenticated):
             filters=auth_filters or {},
         )
 
-        async with GrpcClient() as client:
-            response = await client.threads.Copy(request)
+        client = await get_shared_client()
+        response = await client.threads.Copy(request)
 
         thread = proto_to_thread(response)
 

langgraph_api/js/package.json

@@ -18,7 +18,7 @@
     "@typescript/vfs": "^1.6.0",
     "dedent": "^1.5.3",
     "exit-hook": "^4.0.0",
-    "hono": "^4.10.2",
+    "hono": "^4.10.3",
     "p-queue": "^8.0.1",
     "p-retry": "^6.2.0",
     "tsx": "^4.19.3",

langgraph_api/js/src/graph.mts

@@ -62,6 +62,17 @@ export async function resolveGraph(
     return "compile" in graph && typeof graph.compile === "function";
   };
 
+  const isCompiledGraph = (
+    graph: GraphLike,
+  ): graph is CompiledGraph<string> => {
+    if (typeof graph !== "object" || graph == null) return false;
+    return (
+      "builder" in graph &&
+      typeof graph.builder === "object" &&
+      graph.builder != null
+    );
+  };
+
   const graph: GraphUnknown = await import(sourceFile).then(
     (module) => module[exportSymbol || "default"],
   );
@@ -73,6 +84,15 @@ export async function resolveGraph(
 
       const afterResolve = (graphLike: GraphLike): CompiledGraph<string> => {
         const graph = isGraph(graphLike) ? graphLike.compile() : graphLike;
+
+        // TODO: hack, remove once LangChain 1.x createAgent is fixed
+        // LangGraph API will assign it's checkpointer by setting it
+        // via `graph.checkpointer = ...` and `graph.store = ...`, and the 1.x `createAgent`
+        // hides the underlying `StateGraph` instance, so we need to access it directly.
+        if (!isCompiledGraph(graph) && "graph" in graph) {
+          return (graph as { graph: CompiledGraph<string> }).graph;
+        }
+
         return graph;
       };
 

langgraph_api/js/yarn.lock

@@ -971,10 +971,10 @@ has-flag@^4.0.0:
   resolved "https://registry.yarnpkg.com/has-flag/-/has-flag-4.0.0.tgz#944771fd9c81c81265c4d6941860da06bb59479b"
   integrity sha512-EykJT/Q1KjTWctppgIAgfSO0tKVuZUjhgMr17kqTumMl6Afv3EISleU7qZUzoXDFTAHTDC4NOoG/ZxU3EvlMPQ==
 
-hono@^4.10.2, hono@^4.5.4:
-  version "4.10.2"
-  resolved "https://registry.yarnpkg.com/hono/-/hono-4.10.2.tgz#a78d322f2c5fabebb1887f7ae1632a6524513e74"
-  integrity sha512-p6fyzl+mQo6uhESLxbF5WlBOAJMDh36PljwlKtP5V1v09NxlqGru3ShK+4wKhSuhuYf8qxMmrivHOa/M7q0sMg==
+hono@^4.10.3, hono@^4.5.4:
+  version "4.10.3"
+  resolved "https://registry.yarnpkg.com/hono/-/hono-4.10.3.tgz#4e4063eebaac2b735ca4c7455b4d7a6339afc251"
+  integrity sha512-2LOYWUbnhdxdL8MNbNg9XZig6k+cZXm5IjHn2Aviv7honhBMOHb+jxrKIeJRZJRmn+htUCKhaicxwXuUDlchRA==
 
 icss-utils@^5.0.0, icss-utils@^5.1.0:
   version "5.1.0"

langgraph_api/route.py

@@ -14,6 +14,7 @@ from starlette.responses import JSONResponse
 from starlette.routing import Route, compile_path, get_name
 from starlette.types import ASGIApp, Receive, Scope, Send
 
+from langgraph_api import config
 from langgraph_api.serde import json_dumpb
 from langgraph_api.utils import get_auth_ctx, with_user
 
@@ -58,10 +59,11 @@ class ApiResponse(JSONResponse):
 
 
 def _json_loads(content: bytearray, schema: SchemaType) -> typing.Any:
-    json = orjson.loads(content)
+    """Parse JSON and validate schema. Used by threadpool for large payloads."""
+    json_data = orjson.loads(content)
     if schema is not None:
-        schema.validate(json)
-    return json
+        schema.validate(json_data)
+    return json_data
 
 
 class ApiRequest(Request):
@@ -76,8 +78,16 @@ class ApiRequest(Request):
     async def json(self, schema: SchemaType = None) -> typing.Any:
         if not hasattr(self, "_json"):
             body = await self.body()
+
+            # Hybrid approach for optimal performance:
+            # - Small payloads: parse directly (fast, no queueing/thread pool limitations)
+            # - Large payloads: use dedicated thread pool (safer, doesn't block event loop)
             try:
-                self._json = await run_in_threadpool(_json_loads, body, schema)
+                self._json = (
+                    await run_in_threadpool(_json_loads, body, schema)
+                    if len(body) > config.JSON_THREAD_POOL_MINIMUM_SIZE_BYTES
+                    else _json_loads(body, schema)
+                )
             except orjson.JSONDecodeError as e:
                 raise HTTPException(
                     status_code=422, detail="Invalid JSON in request body"

langgraph_api/serde.py

@@ -3,7 +3,7 @@ import re
 import uuid
 from base64 import b64encode
 from collections import deque
-from collections.abc import Mapping
+from collections.abc import Callable, Mapping
 from datetime import timedelta, timezone
 from decimal import Decimal
 from ipaddress import (
@@ -16,7 +16,7 @@ from ipaddress import (
 )
 from pathlib import Path
 from re import Pattern
-from typing import Any, NamedTuple, cast
+from typing import Any, Literal, NamedTuple, cast
 from zoneinfo import ZoneInfo
 
 import cloudpickle
@@ -123,6 +123,14 @@ def _sanitise(o: Any) -> Any:
     if isinstance(o, Mapping):
         return {_sanitise(k): _sanitise(v) for k, v in o.items()}
     if isinstance(o, list | tuple | set):
+        if (
+            isinstance(o, tuple)
+            and hasattr(o, "_asdict")
+            and callable(o._asdict)
+            and hasattr(o, "_fields")
+            and isinstance(o._fields, tuple)
+        ):  # named tuple
+            return {f: _sanitise(ov) for f, ov in zip(o._fields, o, strict=True)}
         ctor = list if isinstance(o, list) else type(o)
         return ctor(_sanitise(x) for x in o)
     return o
@@ -158,18 +166,46 @@ async def ajson_loads(content: bytes | Fragment) -> Any:
 
 
 class Serializer(JsonPlusSerializer):
+    def __init__(
+        self,
+        __unpack_ext_hook__: Callable[[int, bytes], Any] | None = None,
+        pickle_fallback: bool | None = None,
+    ):
+        from langgraph_api.config import SERDE
+
+        allowed_json_modules: list[tuple[str, ...]] | Literal[True] | None = None
+        if SERDE and "allowed_json_modules" in SERDE:
+            allowed_ = SERDE["allowed_json_modules"]
+            if allowed_ is True:
+                allowed_json_modules = True
+            elif allowed_ is None:
+                allowed_json_modules = None
+            else:
+                allowed_json_modules = [tuple(x) for x in allowed_]
+        if pickle_fallback is None:
+            if SERDE and "pickle_fallback" in SERDE:
+                pickle_fallback = SERDE["pickle_fallback"]
+            else:
+                pickle_fallback = True
+
+        super().__init__(
+            allowed_json_modules=allowed_json_modules,
+            __unpack_ext_hook__=__unpack_ext_hook__,
+        )
+        self.pickle_fallback = pickle_fallback
+
     def dumps_typed(self, obj: Any) -> tuple[str, bytes]:
         try:
             return super().dumps_typed(obj)
         except TypeError:
             return "pickle", cloudpickle.dumps(obj)
 
-    def dumps(self, obj: Any) -> bytes:
-        # See comment above (in json_dumpb)
-        return super().dumps(obj).replace(rb"\\u0000", b"").replace(rb"\u0000", b"")
-
     def loads_typed(self, data: tuple[str, bytes]) -> Any:
         if data[0] == "pickle":
+            if not self.pickle_fallback:
+                raise ValueError(
+                    "Pickle fallback is disabled. Cannot deserialize pickled object."
+                )
             try:
                 return cloudpickle.loads(data[1])
             except Exception as e:
@@ -177,8 +213,16 @@ class Serializer(JsonPlusSerializer):
                     "Failed to unpickle object, replacing w None", exc_info=e
                 )
                 return None
-        return super().loads_typed(data)
-
-
-mpack_keys = {"method", "value"}
-SERIALIZER = Serializer()
+        try:
+            return super().loads_typed(data)
+        except Exception:
+            if data[0] == "json":
+                logger.exception(
+                    "Heads up! There was a deserialization error of an item stored using 'json'-type serialization."
+                    ' For security reasons, starting in langgraph-api version 0.5.0, we no longer serialize objects using the "json" type.'
+                    " If you would like to retain the ability to deserialize old checkpoints saved in this format, "
+                    'please set the "allowed_json_modules" option in your langgraph.json configuration to add the'
+                    " necessary module and type paths to an allow-list to be deserialized. You can alkso retain the"
+                    ' ability to insecurely deserialize custom types by setting it to "true".'
+                )
+            raise

{langgraph_api-0.4.48.dist-info → langgraph_api-0.5.1.dist-info}/METADATA

@@ -1,7 +1,7 @@
 Metadata-Version: 2.4
 Name: langgraph-api
-Version: 0.4.48
-Author-email: Nuno Campos <nuno@langchain.dev>, Will Fu-Hinthorn <will@langchain.dev>
+Version: 0.5.1
+Author-email: Will Fu-Hinthorn <will@langchain.dev>, Josh Rogers <josh@langchain.dev>, Parker Rule <parker@langchain.dev>
 License: Elastic-2.0
 License-File: LICENSE
 Requires-Python: >=3.11
@@ -12,10 +12,10 @@ Requires-Dist: grpcio<2.0.0,>=1.75.0
 Requires-Dist: httpx>=0.25.0
 Requires-Dist: jsonschema-rs<0.30,>=0.20.0
 Requires-Dist: langchain-core>=0.3.64
-Requires-Dist: langgraph-checkpoint>=2.0.23
-Requires-Dist: langgraph-runtime-inmem<0.15.0,>=0.14.0
+Requires-Dist: langgraph-checkpoint<4,>=3
+Requires-Dist: langgraph-runtime-inmem<0.16.0,>=0.15.0
 Requires-Dist: langgraph-sdk>=0.2.0
-Requires-Dist: langgraph>=0.4.0
+Requires-Dist: langgraph<2,>=0.4.10
 Requires-Dist: langsmith>=0.3.45
 Requires-Dist: opentelemetry-api>=1.37.0
 Requires-Dist: opentelemetry-exporter-otlp-proto-http>=1.37.0

{langgraph_api-0.4.48.dist-info → langgraph_api-0.5.1.dist-info}/RECORD

@@ -1,9 +1,9 @@
-langgraph_api/__init__.py,sha256=w0y38C_CURQVpRIfkitkwViZZChCbN89DFUhKKh3PA8,23
+langgraph_api/__init__.py,sha256=eZ1bOun1DDVV0YLOBW4wj2FP1ajReLjbIrGmzN7ASBw,22
 langgraph_api/asgi_transport.py,sha256=XtiLOu4WWsd-xizagBLzT5xUkxc9ZG9YqwvETBPjBFE,5161
 langgraph_api/asyncio.py,sha256=FEEkLm_N-15cbElo4vQ309MkDKBZuRqAYV8VJ1DocNw,9860
 langgraph_api/cli.py,sha256=aEI2pfztEEziIwUk2imiLkNVK1LapMp_3dxvcar1org,18341
 langgraph_api/command.py,sha256=Bh-rvuTLwdHCqFWryCjB1M8oWxPBwRBUjMNj_04KPxM,852
-langgraph_api/config.py,sha256=79efOb8cBNNT1geUfVsUPJFycjUBXnIxWwMkMsbl36A,15292
+langgraph_api/config.py,sha256=YK0tQ3ZbbI3GgqLcX4-Jayzt4iJX3gsFaa8IeTWEGXM,17047
 langgraph_api/cron_scheduler.py,sha256=25wYzEQrhPEivZrAPYOmzLPDOQa-aFogU37mTXc9TJk,2566
 langgraph_api/errors.py,sha256=zlnl3xXIwVG0oGNKKpXf1an9Rn_SBDHSyhe53hU6aLw,1858
 langgraph_api/executor_entrypoint.py,sha256=CaX813ygtf9CpOaBkfkQXJAHjFtmlScCkrOvTDmu4Aw,750
@@ -16,11 +16,11 @@ langgraph_api/logging.py,sha256=o5iVARqtFYKIcRrK2nk1ymcKEiVYKd_dHmhXLF2khFI,6090
 langgraph_api/metadata.py,sha256=Ah5x5TB8O1VAypzDa1UTrsptS1hjy9z-PuNF8WYl3VM,8597
 langgraph_api/patch.py,sha256=J0MmcfpZG15SUVaVcI0Z4x_c0-0rbbT7Pwh9fDAQOpA,1566
 langgraph_api/queue_entrypoint.py,sha256=z3ZUBl3CpnMm0KFPqCuGvSohPAmYQbhAdyRizSJSClM,8481
-langgraph_api/route.py,sha256=EBhELuJ1He-ZYcAnR5YTImcIeDtWthDae5CHELBxPkM,5056
+langgraph_api/route.py,sha256=wh2vMKksTpXJRQ_rLLrFXBSlG608fSMJguZATSWu0Y8,5593
 langgraph_api/schema.py,sha256=spZ_XPT4AMJfw2YatsdnMZZLzgB9Sm3YR8n0SlgGdJ8,8480
 langgraph_api/self_hosted_logs.py,sha256=9ljOz3KH3O1SwsD7eTKnreyJ80NbeR7nj7SuxBlrmCc,4422
 langgraph_api/self_hosted_metrics.py,sha256=3FFezxjU0Vs-bsH39f4Dcwn7fporTLHV9REQ3UQ315A,14004
-langgraph_api/serde.py,sha256=Jkww6ixP5o2YZmnXtM7ihuAYC6YSuNDNPvE-8ILoqVo,5499
+langgraph_api/serde.py,sha256=gNnTO98OVcIsfXJr60bG7GPioKUhTDZYqIWHc97Uyz4,7658
 langgraph_api/server.py,sha256=PExNHgem0tY_KkRFiFzj8m8Np6TrP4M0XJsEw6O2SAU,10112
 langgraph_api/sse.py,sha256=SLdtZmTdh5D8fbWrQjuY9HYLd2dg8Rmi6ZMmFMVc2iE,4204
 langgraph_api/state.py,sha256=AjkLbUQakIwK7oGzJ8oqubazRsXxG3vDMnRa0s0mzDM,4716
@@ -33,7 +33,7 @@ langgraph_api/webhook.py,sha256=SvSM1rdnNtiH4q3JQYmAqJUk2Sable5xAcwOLuRhtlo,1723
 langgraph_api/worker.py,sha256=HHgf590xElF7v02lgn0lG0iK2v2sENMjdx7TVFCvYXY,15399
 langgraph_api/api/__init__.py,sha256=wrnxz_204b2Vhv4-N0WpiPf-ZpDDlmIQkbh-TiXPnOo,5997
 langgraph_api/api/a2a.py,sha256=HIHZkLnIcM1u1FJti-L2NH-h1I9BZ_d-QW9z3gFonn8,53995
-langgraph_api/api/assistants.py,sha256=tRJse7Gr2BTeTZPljL05UvGkFiULpA-6hy03nBx9PF4,18177
+langgraph_api/api/assistants.py,sha256=OCup8rXk0HaqWqDhOLz59f1KSTS5fwxXrCaDgyopges,17981
 langgraph_api/api/mcp.py,sha256=qe10ZRMN3f-Hli-9TI8nbQyWvMeBb72YB1PZVbyqBQw,14418
 langgraph_api/api/meta.py,sha256=_jG61UKs0J_alsCDgIwCAx1rX5pYuUwKrmOEpWnzR1I,4817
 langgraph_api/api/openapi.py,sha256=If-z1ckXt-Yu5bwQytK1LWyX_T7G46UtLfixgEP8hwc,11959
@@ -50,12 +50,12 @@ langgraph_api/auth/langsmith/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NM
 langgraph_api/auth/langsmith/backend.py,sha256=JVf8-q1IvB5EeiLJge3cOtPvDg6qHzK_4cR-R8hPXXQ,3753
 langgraph_api/auth/langsmith/client.py,sha256=79kwCVeHU64nsHsxWipfZhf44lM6vfs2nlfTxlJF6LU,4142
 langgraph_api/grpc_ops/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-langgraph_api/grpc_ops/client.py,sha256=usSA1ZObwFzspgh3VYvhwRv9ZEOErcOzVc25JrsvYQU,2920
-langgraph_api/grpc_ops/ops.py,sha256=OHU1ikDMKJXRSTU7zWDXorEuB6o9czLErijh3mhDYh0,37867
+langgraph_api/grpc_ops/client.py,sha256=Qr07JVaJrMr3jWQKFHngyC3gqsj-VNLzahbnpj1vDO8,5753
+langgraph_api/grpc_ops/ops.py,sha256=6L7OiKcVQgAVjkrWiGO5o1pnkZ1Fcas1H3ULtyp0vbo,37442
 langgraph_api/grpc_ops/generated/__init__.py,sha256=dRiB_iGscPKdMpuLp9ueLwAmIfRaNjNXC64ABtb4cg8,135
 langgraph_api/grpc_ops/generated/core_api_pb2.py,sha256=l209i8cIazfs-zPTlt2jUg_og82oiDT4QMQCYAhU0P4,42262
 langgraph_api/grpc_ops/generated/core_api_pb2.pyi,sha256=6fnrjKRdN1-jJfHagLHhdlVog1cLkLoAc9fvTBzeFdM,49429
-langgraph_api/grpc_ops/generated/core_api_pb2_grpc.py,sha256=K_bHjM6BDzqIc1N8lc1SaxRLGFp1GUTpSiQEr5-70Oo,52466
+langgraph_api/grpc_ops/generated/core_api_pb2_grpc.py,sha256=Qav2DuCMUSmR8nP4-fVtUBbY0Vc42jqjCs3L4LdIl-0,52467
 langgraph_api/js/.gitignore,sha256=l5yI6G_V6F1600I1IjiUKn87f4uYIrBAYU1MOyBBhg4,59
 langgraph_api/js/.prettierrc,sha256=0es3ovvyNIqIw81rPQsdt1zCQcOdBqyR_DMbFE4Ifms,19
 langgraph_api/js/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -65,15 +65,15 @@ langgraph_api/js/client.http.mts,sha256=FeVM53vduTPCyMPaYs__kmB3iWcz0k0om811DG0J
 langgraph_api/js/client.mts,sha256=8T5wp_114c2wGPfktY77StTnejhYL3ZWBmLwaUvp5XU,32333
 langgraph_api/js/errors.py,sha256=Cm1TKWlUCwZReDC5AQ6SgNIVGD27Qov2xcgHyf8-GXo,361
 langgraph_api/js/global.d.ts,sha256=j4GhgtQSZ5_cHzjSPcHgMJ8tfBThxrH-pUOrrJGteOU,196
-langgraph_api/js/package.json,sha256=besBq5s3c370nNWhHXy8ZxD0X350Xbzx4FNUDxhG-Pk,1330
+langgraph_api/js/package.json,sha256=i6LBt4R-bi70FuMC3lIkZAuWZVJnlDE_CpI0mnArrOA,1330
 langgraph_api/js/remote.py,sha256=gBk273R7esmXg8aR6InxasNFc5E6Qju2bv2DhmmGJyU,38676
 langgraph_api/js/schema.py,sha256=M4fLtr50O1jck8H1hm_0W4cZOGYGdkrB7riLyCes4oY,438
 langgraph_api/js/sse.py,sha256=hHkbncnYnXNIbHhAWneGWYkHp4UhhhGB7-MYtDrY264,4141
 langgraph_api/js/traceblock.mts,sha256=QtGSN5VpzmGqDfbArrGXkMiONY94pMQ5CgzetT_bKYg,761
 langgraph_api/js/tsconfig.json,sha256=imCYqVnqFpaBoZPx8k1nO4slHIWBFsSlmCYhO73cpBs,341
 langgraph_api/js/ui.py,sha256=l9regrvKIxLOjH5SIYE2nhr8QCKLK1Q_1pZgxdL71X4,2488
-langgraph_api/js/yarn.lock,sha256=eWe1iuI634accFo7tumAcG8I7gLrcYKe7OyX0TMfK_s,83943
-langgraph_api/js/src/graph.mts,sha256=9zTQNdtanI_CFnOwNRoamoCVHHQHGbNlbm91aRxDeOc,2675
+langgraph_api/js/yarn.lock,sha256=CV3hl-TyqWBKPLYvnWcn42-95nxVlDJRvUuPcC-CQuo,83943
+langgraph_api/js/src/graph.mts,sha256=etZd27NaoVevyitJ-LAUue0HeR7V3F2YNeSGwWHm13s,3417
 langgraph_api/js/src/load.hooks.mjs,sha256=xNVHq75W0Lk6MUKl1pQYrx-wtQ8_neiUyI6SO-k0ecM,2235
 langgraph_api/js/src/preload.mjs,sha256=8m3bYkf9iZLCQzKAYAdU8snxUwAG3dVLwGvAjfGfgIc,959
 langgraph_api/js/src/utils/files.mts,sha256=nU09Y8lN8SYsg0x2ffmbIW8LEDBl-SWkmxsoXunFU0M,219
@@ -110,8 +110,8 @@ langgraph_runtime/store.py,sha256=7mowndlsIroGHv3NpTSOZDJR0lCuaYMBoTnTrewjslw,11
 LICENSE,sha256=ZPwVR73Biwm3sK6vR54djCrhaRiM4cAD2zvOQZV8Xis,3859
 logging.json,sha256=3RNjSADZmDq38eHePMm1CbP6qZ71AmpBtLwCmKU9Zgo,379
 openapi.json,sha256=Oi2tU1b8PsXb-6XNHafQvcZv934vLNQhBNPYXr9e2nU,172620
-langgraph_api-0.4.48.dist-info/METADATA,sha256=UDqSGmsHJIXF0DrV_u8dFni__0-xszEswI232NY8l6M,4149
-langgraph_api-0.4.48.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-langgraph_api-0.4.48.dist-info/entry_points.txt,sha256=hGedv8n7cgi41PypMfinwS_HfCwA7xJIfS0jAp8htV8,78
-langgraph_api-0.4.48.dist-info/licenses/LICENSE,sha256=ZPwVR73Biwm3sK6vR54djCrhaRiM4cAD2zvOQZV8Xis,3859
-langgraph_api-0.4.48.dist-info/RECORD,,
+langgraph_api-0.5.1.dist-info/METADATA,sha256=8VfO9GgGcyYdDTov627-7xBHu-GD23j2SMVAORR4zoM,4186
+langgraph_api-0.5.1.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+langgraph_api-0.5.1.dist-info/entry_points.txt,sha256=hGedv8n7cgi41PypMfinwS_HfCwA7xJIfS0jAp8htV8,78
+langgraph_api-0.5.1.dist-info/licenses/LICENSE,sha256=ZPwVR73Biwm3sK6vR54djCrhaRiM4cAD2zvOQZV8Xis,3859
+langgraph_api-0.5.1.dist-info/RECORD,,