langgraph-api 0.2.130__py3-none-any.whl → 0.2.134__py3-none-any.whl

langgraph_api/http.py

@@ -72,7 +72,7 @@ class JsonHttpClient:
 
 
 _http_client: JsonHttpClient
-_loopback_client: JsonHttpClient = None
+_loopback_client: JsonHttpClient | None = None
 
 
 async def start_http_client() -> None:
@@ -113,16 +113,16 @@ def get_loopback_client() -> JsonHttpClient:
     return _loopback_client
 
 
-def is_retriable_error(exception: Exception) -> bool:
+def is_retriable_error(exception: BaseException) -> bool:
     # httpx error hierarchy: https://www.python-httpx.org/exceptions/
     # Retry all timeout related errors
     if isinstance(exception, httpx.TimeoutException | httpx.NetworkError):
         return True
     # Seems to just apply to HttpStatusError but doesn't hurt to check all
     if isinstance(exception, httpx.HTTPError):
-        return getattr(exception, "response", None) is not None and (
-            exception.response.status_code >= 500
-            or exception.response.status_code == 429
+        response = getattr(exception, "response", None)
+        return response is not None and (
+            response.status_code >= 500 or response.status_code == 429
         )
     return False
 
@@ -149,7 +149,7 @@ async def http_request(
     request_timeout: float | None = 30,
     raise_error: bool = True,
     client: JsonHttpClient | None = None,
-) -> httpx.Response:
+) -> None:
     """Make an HTTP request with retries.
 
     Args:
@@ -173,7 +173,10 @@ async def http_request(
 
     content = None
     if body is not None:
-        content = body
+        if isinstance(body, str):
+            content = body.encode("utf-8")
+        else:
+            content = body
     elif json is not None:
         content = json_dumpb(json)
 

langgraph_api/http_metrics.py

@@ -98,7 +98,10 @@ class HTTPMetricsCollector:
         hist_data["count"] += 1
 
     def get_metrics(
-        self, project_id: str, revision_id: str, format: str = "prometheus"
+        self,
+        project_id: str | None,
+        revision_id: str | None,
+        format: str = "prometheus",
     ) -> dict | list[str]:
         if format == "json":
             return {

langgraph_api/js/build.mts

@@ -17,7 +17,14 @@ const __dirname = new URL(".", import.meta.url).pathname;
 
 async function main() {
   const specs = Object.entries(
-    z.record(z.string()).parse(JSON.parse(process.env.LANGSERVE_GRAPHS)),
+    z
+      .record(
+        z.union([
+          z.string(),
+          z.object({ path: z.string(), description: z.string().nullish() }),
+        ]),
+      )
+      .parse(JSON.parse(process.env.LANGSERVE_GRAPHS)),
   ).filter(([_, spec]) => filterValidExportPath(spec));
 
   let GRAPH_SCHEMAS: Record<string, Record<string, GraphSchema> | false> = {};
@@ -49,7 +56,9 @@ async function main() {
       await Promise.all(
         specs.map(async ([graphId, rawSpec]) => {
           console.info(`[${graphId}]: Checking for source file existence`);
-          const { resolved, ...spec } = await resolveGraph(rawSpec, {
+          const importPath =
+            typeof rawSpec === "string" ? rawSpec : rawSpec.path;
+          const { resolved, ...spec } = await resolveGraph(importPath, {
             onlyFilePresence: true,
           });
 

langgraph_api/js/client.http.mts

@@ -104,6 +104,8 @@ async function registerHttp(appPath: string, options: { cwd: string }) {
 }
 
 async function main() {
+  logger.info("Starting HTTP loop", { pid: process.pid });
+
   const http = z
     .object({
       app: z.string().optional(),

langgraph_api/js/client.mts

@@ -950,6 +950,8 @@ async function* getStateHistoryRequest(
 const __dirname = new URL(".", import.meta.url).pathname;
 
 async function main() {
+  logger.info("Starting graph loop", { pid: process.pid });
+
   const app = new Hono();
 
   GRAPH_OPTIONS = {
@@ -959,7 +961,12 @@ async function main() {
 
   const specs = Object.entries(
     z
-      .record(z.string())
+      .record(
+        z.union([
+          z.string(),
+          z.object({ path: z.string(), description: z.string().nullish() }),
+        ]),
+      )
       .parse(JSON.parse(process.env.LANGSERVE_GRAPHS ?? "{}")),
   ).filter(([_, spec]) => filterValidExportPath(spec));
 
@@ -978,7 +985,8 @@ async function main() {
   await Promise.all(
     specs.map(async ([graphId, rawSpec]) => {
       logger.info(`Resolving graph ${graphId}`);
-      const { resolved, ...spec } = await resolveGraph(rawSpec);
+      const importPath = typeof rawSpec === "string" ? rawSpec : rawSpec.path;
+      const { resolved, ...spec } = await resolveGraph(importPath);
 
       GRAPH_RESOLVED[graphId] = resolved;
       GRAPH_SPEC[graphId] = spec;
@@ -1140,7 +1148,9 @@ async function getNodesExecutedRequest(
 ) {
   const value = nodesExecuted;
   nodesExecuted = 0;
-  logger.debug(`Returning ${value} nodes executed. Reset nodes executed to ${nodesExecuted}.`); 
+  logger.debug(
+    `Returning ${value} nodes executed. Reset nodes executed to ${nodesExecuted}.`,
+  );
   return { nodesExecuted: value };
 }
 patchFetch();

langgraph_api/js/package.json

@@ -11,8 +11,8 @@
     "@hono/zod-validator": "^0.2.2",
     "@langchain/core": "^0.3.59",
     "@langchain/langgraph": "^0.2.65",
-    "@langchain/langgraph-api": "~0.0.56",
-    "@langchain/langgraph-ui": "~0.0.56",
+    "@langchain/langgraph-api": "~0.0.59",
+    "@langchain/langgraph-ui": "~0.0.59",
     "@langchain/langgraph-checkpoint": "~0.0.18",
     "@types/json-schema": "^7.0.15",
     "@typescript/vfs": "^1.6.0",

langgraph_api/js/remote.py

@@ -106,6 +106,10 @@ async def _client_stream(method: str, data: dict[str, Any]):
                 yield sse.data
 
 
+class NoopModel(BaseModel):
+    pass
+
+
 async def _client_invoke(method: str, data: dict[str, Any]):
     graph_id = data.get("graph_id")
     res = await _client.post(
@@ -177,13 +181,10 @@ class RemotePregel(BaseRemotePregel):
         nodes: list[Any] = response.pop("nodes")
         edges: list[Any] = response.pop("edges")
 
-        class NoopModel(BaseModel):
-            pass
-
         return DrawableGraph(
             {
                 data["id"]: Node(
-                    data["id"], data["id"], NoopModel(), data.get("metadata")
+                    data["id"], data["id"], NoopModel, data.get("metadata")
                 )
                 for data in nodes
             },
@@ -244,9 +245,9 @@ class RemotePregel(BaseRemotePregel):
                 )
             return tuple(result)
 
-        return StateSnapshot(
+        return StateSnapshot(  # type: ignore[missing-argument]
             item.get("values"),
-            item.get("next"),
+            cast(tuple, item.get("next", ())),
             item.get("config"),
             item.get("metadata"),
             item.get("createdAt"),
@@ -352,7 +353,7 @@ class RemotePregel(BaseRemotePregel):
         return result["nodesExecuted"]
 
 
-async def run_js_process(paths_str: str, watch: bool = False):
+async def run_js_process(paths_str: str | None, watch: bool = False):
     # check if tsx is available
     tsx_path = shutil.which("tsx")
     if tsx_path is None:
@@ -360,7 +361,7 @@ async def run_js_process(paths_str: str, watch: bool = False):
             "tsx not found in PATH. Please upgrade to latest LangGraph CLI to support running JS graphs."
         )
     attempt = 0
-    while not asyncio.current_task().cancelled():
+    while (current_task := asyncio.current_task()) and not current_task.cancelled():
         client_file = os.path.join(os.path.dirname(__file__), "client.mts")
         client_preload_file = os.path.join(
             os.path.dirname(__file__), "src", "preload.mjs"
@@ -408,7 +409,9 @@ async def run_js_process(paths_str: str, watch: bool = False):
                 attempt += 1
 
 
-async def run_js_http_process(paths_str: str, http_config: dict, watch: bool = False):
+async def run_js_http_process(
+    paths_str: str | None, http_config: dict, watch: bool = False
+):
     # check if tsx is available
     tsx_path = shutil.which("tsx")
     if tsx_path is None:
@@ -417,7 +420,7 @@ async def run_js_http_process(paths_str: str, http_config: dict, watch: bool = F
         )
 
     attempt = 0
-    while not asyncio.current_task().cancelled():
+    while (current_task := asyncio.current_task()) and not current_task.cancelled():
         client_file = os.path.join(os.path.dirname(__file__), "client.http.mts")
         args = ("tsx", "watch", client_file) if watch else ("tsx", client_file)
         pid = None
@@ -795,7 +798,9 @@ async def wait_until_js_ready():
             ) as checkpointer_client,
         ):
             attempt = 0
-            while not asyncio.current_task().cancelled():
+            while (
+                current_task := asyncio.current_task()
+            ) and not current_task.cancelled():
                 try:
                     res = await graph_client.get("/ok")
                     res.raise_for_status()
@@ -909,7 +914,7 @@ class CustomJsAuthBackend(AuthenticationBackend):
 
 
 async def handle_js_auth_event(
-    ctx: Auth.types.AuthContext | None,
+    ctx: Auth.types.AuthContext,
     value: dict,
 ) -> Auth.types.FilterType | None:
     if hasattr(ctx.user, "dict") and callable(ctx.user.dict):

langgraph_api/js/src/preload.mjs

@@ -11,7 +11,15 @@ const cwd = process.cwd();
 // be working fine as well.
 const firstGraphFile =
   Object.values(graphs)
-    .flatMap((i) => i.split(":").at(0))
+    .map((i) => {
+      if (typeof i === "string") {
+        return i.split(":").at(0);
+      } else if (i && typeof i === "object" && i.path) {
+        return i.path.split(":").at(0);
+      }
+      return null;
+    })
+    .filter(Boolean)
     .at(0) || "index.mts";
 
 // enforce API @langchain/langgraph resolution

langgraph_api/js/src/utils/files.mts

@@ -1,4 +1,7 @@
-export function filterValidExportPath(path: string | undefined) {
+export function filterValidExportPath(
+  path: string | { path: string } | undefined,
+) {
   if (!path) return false;
-  return !path.split(":")[0].endsWith(".py");
+  const p = typeof path === "string" ? path : path.path;
+  return !p.split(":")[0].endsWith(".py");
 }

langgraph_api/js/sse.py

@@ -91,7 +91,7 @@ class SSEDecoder:
 
             sse = StreamPart(
                 event=self._event,
-                data=orjson.loads(self._data) if self._data else None,
+                data=orjson.loads(self._data) if self._data else None,  # type: ignore[invalid-argument-type]
             )
 
             # NOTE: as per the SSE spec, do not reset last_event_id.

langgraph_api/js/yarn.lock

@@ -203,15 +203,15 @@
     zod "^3.25.32"
     zod-to-json-schema "^3.22.3"
 
-"@langchain/langgraph-api@~0.0.56":
-  version "0.0.56"
-  resolved "https://registry.yarnpkg.com/@langchain/langgraph-api/-/langgraph-api-0.0.56.tgz#9e0a6e5dd9af2f6ec259a32ca9081b19457a0825"
-  integrity sha512-BhrbMKSc3f4CXXDXgvHR8mE2bsgD/G6dDzr+yDkEC7a9eeFI7C8bPhkGmegFJ+q7i6GWtqlQOYSCWHqQFZxynQ==
+"@langchain/langgraph-api@~0.0.59":
+  version "0.0.59"
+  resolved "https://registry.yarnpkg.com/@langchain/langgraph-api/-/langgraph-api-0.0.59.tgz#a3d69b8cc68ceebd8a8d86b77abd03924ddcd02c"
+  integrity sha512-pUt3yKB2z1nsXdhqpRQgVefYg5MdJVsqH64gTNzSb/JYJBPIUj2h8XttRx9+3mHtVqCzvL2bb+HVIterNFOKtw==
   dependencies:
     "@babel/code-frame" "^7.26.2"
     "@hono/node-server" "^1.12.0"
     "@hono/zod-validator" "^0.2.2"
-    "@langchain/langgraph-ui" "0.0.56"
+    "@langchain/langgraph-ui" "0.0.59"
     "@types/json-schema" "^7.0.15"
     "@typescript/vfs" "^1.6.0"
     dedent "^1.5.3"
@@ -256,10 +256,10 @@
     p-retry "4"
     uuid "^9.0.0"
 
-"@langchain/langgraph-ui@0.0.56", "@langchain/langgraph-ui@~0.0.56":
-  version "0.0.56"
-  resolved "https://registry.yarnpkg.com/@langchain/langgraph-ui/-/langgraph-ui-0.0.56.tgz#b566afa4ad940fc6bd942205a00b3afac7fd7363"
-  integrity sha512-cRU+fMCz1NAOjRLcreNa64Gap89yio0p8baRigehtL1SNsXJEaZ2xdGjbnjaWIl8rTrriDak4oisx07whwcBjQ==
+"@langchain/langgraph-ui@0.0.59", "@langchain/langgraph-ui@~0.0.59":
+  version "0.0.59"
+  resolved "https://registry.yarnpkg.com/@langchain/langgraph-ui/-/langgraph-ui-0.0.59.tgz#41988eae48c7520c5ebfa9bcdda65ddadfe1aab9"
+  integrity sha512-x6Jqt7TZRfHGU0MyxVbz7ugucT/oJQm9kM+DlGdmfAEm2JrNqgy1W85E589xJCJjpuRiDLpP7NiqSwnp8lCEqg==
   dependencies:
     "@commander-js/extra-typings" "^13.0.0"
     commander "^13.0.0"

langgraph_api/logging.py

@@ -19,8 +19,8 @@ LOG_LEVEL = log_env("LOG_LEVEL", cast=str, default="INFO")
 logging.getLogger().setLevel(LOG_LEVEL.upper())
 logging.getLogger("psycopg").setLevel(logging.WARNING)
 
-worker_config: contextvars.ContextVar[dict[str, typing.Any] | None] = (
-    contextvars.ContextVar("worker_config", default=None)
+worker_config = contextvars.ContextVar[dict[str, typing.Any] | None](
+    "worker_config", default=None
 )
 
 # custom processors
@@ -165,6 +165,6 @@ if not structlog.is_configured():
             structlog.stdlib.ProcessorFormatter.wrap_for_formatter,
         ],
         logger_factory=structlog.stdlib.LoggerFactory(),
-        wrapper_class=structlog.stdlib.BoundLogger,
+        wrapper_class=structlog.stdlib.BoundLogger,  # type: ignore[invalid-argument-type]
         cache_logger_on_first_use=True,
     )

langgraph_api/middleware/http_logger.py

@@ -84,6 +84,7 @@ class AccessLoggerMiddleware:
 
             if method and route and status:
                 HTTP_METRICS_COLLECTOR.record_request(method, route, status, latency)
+            qs = scope.get("query_string")
             self.logger.log(
                 _get_level(status),
                 f"{method} {path} {status} {latency}ms",
@@ -93,7 +94,7 @@ class AccessLoggerMiddleware:
                 latency_ms=latency,
                 route=route,
                 path_params=scope.get("path_params"),
-                query_string=scope.get("query_string").decode(),
+                query_string=qs.decode() if qs else "",
                 proto=scope.get("http_version"),
                 req_header=_headers_to_dict(scope.get("headers")),
                 res_header=_headers_to_dict(info["response"].get("headers")),

langgraph_api/models/run.py

@@ -3,11 +3,11 @@ import time
 import urllib.parse
 import uuid
 from collections.abc import Mapping, Sequence
-from typing import Any, NamedTuple
+from typing import Any, NamedTuple, cast
 from uuid import UUID
 
 import orjson
-from langgraph.checkpoint.base.id import uuid6
+import structlog
 from starlette.authentication import BaseUser
 from starlette.exceptions import HTTPException
 from typing_extensions import TypedDict
@@ -27,7 +27,10 @@ from langgraph_api.schema import (
 )
 from langgraph_api.utils import AsyncConnectionProto, get_auth_ctx
 from langgraph_api.utils.headers import should_include_header
-from langgraph_runtime.ops import Runs, logger
+from langgraph_api.utils.uuids import uuid7
+from langgraph_runtime.ops import Runs
+
+logger = structlog.stdlib.get_logger(__name__)
 
 
 class LangSmithTracer(TypedDict, total=False):
@@ -183,7 +186,7 @@ LANGSMITH_PROJECT = "langsmith-project"
 DEFAULT_RUN_HEADERS_EXCLUDE = {"x-api-key", "x-tenant-id", "x-service-key"}
 
 
-def get_configurable_headers(headers: dict[str, str]) -> dict[str, str]:
+def get_configurable_headers(headers: Mapping[str, str]) -> dict[str, str]:
     """Extract headers that should be added to run configuration.
 
     This function handles special cases like langsmith-trace and baggage headers,
@@ -249,7 +252,7 @@ async def create_valid_run(
     request_id = headers.get("x-request-id")  # Will be null in the crons scheduler.
     (
         assistant_id,
-        thread_id,
+        thread_id_,
         checkpoint_id,
         run_id,
     ) = _get_ids(
@@ -258,7 +261,7 @@ async def create_valid_run(
         run_id=run_id,
     )
     if (
-        thread_id is None
+        thread_id_ is None
         and (command := payload.get("command"))
         and command.get("resume")
     ):
@@ -266,7 +269,9 @@ async def create_valid_run(
             status_code=400,
             detail="You must provide a thread_id when resuming.",
         )
-    temporary = thread_id is None and payload.get("on_completion", "delete") == "delete"
+    temporary = (
+        thread_id_ is None and payload.get("on_completion", "delete") == "delete"
+    )
     stream_resumable = payload.get("stream_resumable", False)
     stream_mode, multitask_strategy, prevent_insert_if_inflight = assign_defaults(
         payload
@@ -296,7 +301,7 @@ async def create_valid_run(
     configurable.update(get_configurable_headers(headers))
     ctx = get_auth_ctx()
     if ctx:
-        user = ctx.user
+        user = cast(BaseUser | None, ctx.user)
         user_id = get_user_id(user)
         configurable["langgraph_auth_user"] = user
         configurable["langgraph_auth_user_id"] = user_id
@@ -336,7 +341,7 @@ async def create_valid_run(
         metadata=payload.get("metadata"),
         status="pending",
         user_id=user_id,
-        thread_id=thread_id,
+        thread_id=thread_id_,
         run_id=run_id,
         multitask_strategy=multitask_strategy,
         prevent_insert_if_inflight=prevent_insert_if_inflight,
@@ -362,7 +367,7 @@ async def create_valid_run(
         logger.info(
             "Created run",
             run_id=str(run_id),
-            thread_id=str(thread_id),
+            thread_id=str(thread_id_),
             assistant_id=str(assistant_id),
             multitask_strategy=multitask_strategy,
             stream_mode=stream_mode,
@@ -383,7 +388,7 @@ async def create_valid_run(
                 await Runs.cancel(
                     conn,
                     [run["run_id"] for run in inflight_runs],
-                    thread_id=thread_id,
+                    thread_id=thread_id_,
                     action=multitask_strategy,
                 )
             except HTTPException:
@@ -415,15 +420,15 @@ def _get_ids(
     assistant_id = get_assistant_id(payload["assistant_id"])
 
     # ensure UUID validity defaults
-    assistant_id, thread_id, checkpoint_id = ensure_ids(
+    assistant_id, thread_id_, checkpoint_id = ensure_ids(
         assistant_id, thread_id, payload
     )
 
-    run_id = run_id or uuid6()
+    run_id = run_id or uuid7()
 
     return _Ids(
         assistant_id,
-        thread_id,
+        thread_id_,
         checkpoint_id,
         run_id,
     )

langgraph_api/patch.py

@@ -41,8 +41,8 @@ async def StreamingResponse_stream_response(self, send: Send) -> None:
 
 # patch StreamingResponse.stream_response
 
-StreamingResponse.stream_response = StreamingResponse_stream_response
+StreamingResponse.stream_response = StreamingResponse_stream_response  # type: ignore[invalid-assignment]
 
 # patch Response.render
 
-Response.render = Response_render
+Response.render = Response_render  # type: ignore[invalid-assignment]

langgraph_api/queue_entrypoint.py

@@ -18,9 +18,9 @@ import os
 import pathlib
 import signal
 from contextlib import asynccontextmanager
+from typing import cast
 
 import structlog
-import uvloop
 
 from langgraph_runtime.lifespan import lifespan
 from langgraph_runtime.metrics import get_metrics
@@ -36,21 +36,22 @@ async def health_and_metrics_server():
     class HealthAndMetricsHandler(http.server.SimpleHTTPRequestHandler):
         def log_message(self, format, *args):
             # Skip logging for /ok and /metrics endpoints
-            if self.path in ["/ok", "/metrics"]:
+            if getattr(self, "path", None) in ["/ok", "/metrics"]:
                 return
             # Log other requests normally
             super().log_message(format, *args)
 
         def do_GET(self):
-            if self.path == "/ok":
+            path = getattr(self, "path", None)
+            if path == "/ok":
                 self.send_response(200)
                 self.send_header("Content-Type", "application/json")
                 self.send_header("Content-Length", ok_len)
                 self.end_headers()
                 self.wfile.write(ok)
-            elif self.path == "/metrics":
+            elif path == "/metrics":
                 metrics = get_metrics()
-                worker_metrics = metrics["workers"]
+                worker_metrics = cast(dict[str, int], metrics["workers"])
                 workers_max = worker_metrics["max"]
                 workers_active = worker_metrics["active"]
                 workers_available = worker_metrics["available"]
@@ -93,20 +94,27 @@ async def health_and_metrics_server():
             httpd.shutdown()
 
 
-async def entrypoint():
+async def entrypoint(
+    grpc_port: int | None = None, entrypoint_name: str = "python-queue"
+):
     from langgraph_api import logging as lg_logging
     from langgraph_api.api import user_router
 
-    lg_logging.set_logging_context({"entrypoint": "python-queue"})
+    lg_logging.set_logging_context({"entrypoint": entrypoint_name})
     tasks: set[asyncio.Task] = set()
     tasks.add(asyncio.create_task(health_and_metrics_server()))
 
     original_lifespan = user_router.router.lifespan_context if user_router else None
 
     @asynccontextmanager
-    async def combined_lifespan(app, with_cron_scheduler=False, taskset=None):
+    async def combined_lifespan(
+        app, with_cron_scheduler=False, grpc_port=None, taskset=None
+    ):
         async with lifespan(
-            app, with_cron_scheduler=with_cron_scheduler, taskset=taskset
+            app,
+            with_cron_scheduler=with_cron_scheduler,
+            grpc_port=grpc_port,
+            taskset=taskset,
         ):
             if original_lifespan:
                 async with original_lifespan(app):
@@ -114,11 +122,13 @@ async def entrypoint():
             else:
                 yield
 
-    async with combined_lifespan(None, with_cron_scheduler=False, taskset=tasks):
+    async with combined_lifespan(
+        None, with_cron_scheduler=False, grpc_port=grpc_port, taskset=tasks
+    ):
         await asyncio.gather(*tasks)
 
 
-async def main():
+async def main(grpc_port: int | None = None, entrypoint_name: str = "python-queue"):
     """Run the queue entrypoint and shut down gracefully on SIGTERM/SIGINT."""
     loop = asyncio.get_running_loop()
     stop_event = asyncio.Event()
@@ -132,11 +142,9 @@ async def main():
     except (NotImplementedError, RuntimeError):
         signal.signal(signal.SIGTERM, lambda *_: _handle_signal())
 
-    from langgraph_api import config
-
-    config.IS_QUEUE_ENTRYPOINT = True
-
-    entry_task = asyncio.create_task(entrypoint())
+    entry_task = asyncio.create_task(
+        entrypoint(grpc_port=grpc_port, entrypoint_name=entrypoint_name)
+    )
     await stop_event.wait()
 
     logger.warning("Cancelling queue entrypoint task")
@@ -146,10 +154,17 @@ async def main():
 
 
 if __name__ == "__main__":
+    from langgraph_api import config
+
+    config.IS_QUEUE_ENTRYPOINT = True
     with open(pathlib.Path(__file__).parent.parent / "logging.json") as file:
         loaded_config = json.load(file)
         logging.config.dictConfig(loaded_config)
+    try:
+        import uvloop  # type: ignore[unresolved-import]
 
-    uvloop.install()
-
+        uvloop.install()
+    except ImportError:
+        pass
+    # run the entrypoint
     asyncio.run(main())