agent-lab-sdk 0.1.42.1__tar.gz → 0.1.44__tar.gz

{agent_lab_sdk-0.1.42.1 → agent_lab_sdk-0.1.44}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agent-lab-sdk
-Version: 0.1.42.1
+Version: 0.1.44
 Summary: SDK для работы с Agent Lab
 Author-email: Andrew Ohurtsov <andermirik@yandex.com>
 License: Proprietary and Confidential — All Rights Reserved

{agent_lab_sdk-0.1.42.1 → agent_lab_sdk-0.1.44}/agent_lab_sdk/langgraph/checkpoint/agw_saver.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import asyncio
+import threading
 import base64
 import logging
 import os
@@ -102,7 +103,10 @@ class AsyncAGWCheckpointSaver(BaseCheckpointSaver):
         super().__init__(serde=serde)
         self.base_url = base_url.rstrip("/")
         self.timeout = timeout
-        self.loop = asyncio.get_running_loop()
+        # Фоновый loop для sync-обёрток
+        self._bg_loop: asyncio.AbstractEventLoop | None = None
+        self._bg_thread: threading.Thread | None = None
+        self._loop_lock = threading.Lock()
 
         raw_attempts = os.getenv("AGW_HTTP_MAX_RETRIES")
         if raw_attempts is None:
@@ -175,6 +179,7 @@ class AsyncAGWCheckpointSaver(BaseCheckpointSaver):
 
         self._verify = verify
         self._client: httpx.AsyncClient | None = None
+        self._client_loop: asyncio.AbstractEventLoop | None = None
 
     def _create_client(self) -> httpx.AsyncClient:
         return httpx.AsyncClient(
@@ -185,14 +190,23 @@ class AsyncAGWCheckpointSaver(BaseCheckpointSaver):
             trust_env=True,
         )
 
-    def _ensure_client(self) -> httpx.AsyncClient:
-        client = self._client
-        if client is None or client.is_closed:
-            if client is not None and client.is_closed:
-                logger.debug("Recreating closed httpx.AsyncClient for AGW")
-            client = self._create_client()
-            self._client = client
-        return client
+    def _ensure_bg_loop(self) -> asyncio.AbstractEventLoop:
+        with self._loop_lock:
+            if self._bg_loop and self._bg_loop.is_running():
+                return self._bg_loop
+
+            loop = asyncio.new_event_loop()
+
+            def runner():
+                asyncio.set_event_loop(loop)
+                loop.run_forever()
+
+            t = threading.Thread(target=runner, name="agw-checkpoint-loop", daemon=True)
+            t.start()
+
+            self._bg_loop = loop
+            self._bg_thread = t
+            return loop
 
     def _compute_retry_delay(self, attempt: int) -> float:
         if attempt <= 0:
@@ -218,6 +232,14 @@ class AsyncAGWCheckpointSaver(BaseCheckpointSaver):
                 logger.debug("Failed to close AGW httpx.AsyncClient: %s", close_exc)
             finally:
                 self._client = None
+                self._client_loop = None
+        # останавливаем фоновый loop, если поднимали
+        if self._bg_loop is not None:
+            try:
+                self._bg_loop.call_soon_threadsafe(self._bg_loop.stop)
+            finally:
+                self._bg_loop = None
+                self._bg_thread = None
 
     # ----------------------- universal dump/load ---------------------
     def _safe_dump(self, obj: Any) -> Any:
@@ -528,6 +550,7 @@ class AsyncAGWCheckpointSaver(BaseCheckpointSaver):
         path: str,
         *,
         ok_statuses: Iterable[int] | None = None,
+        label_path: str | None = None,
         **kw,
     ) -> httpx.Response:
         if "json" in kw:
@@ -536,14 +559,26 @@ class AsyncAGWCheckpointSaver(BaseCheckpointSaver):
             logger.debug("AGW HTTP payload: %s", kw["data"].decode())
 
         ok_set = set(ok_statuses) if ok_statuses is not None else set()
+        metric_path = label_path or path
 
         attempt = 1
         while True:
-            client = self._ensure_client()
+            # клиент должен принадлежать текущему loop
+            current_loop = asyncio.get_running_loop()
+            client = self._client
+            if client is None or client.is_closed or self._client_loop is not current_loop:
+                if client is not None:
+                    try:
+                        await client.aclose()
+                    except Exception as e:
+                        logger.exception("ошибка при закрытии клиента", e)
+                client = self._create_client()
+                self._client = client
+                self._client_loop = current_loop
             try:
                 resp = await client.request(method, path, **kw)
             except httpx.RequestError as exc:
-                AGW_HTTP_ERROR.labels(method, path).inc()
+                AGW_HTTP_ERROR.labels(method, metric_path).inc()
                 logger.warning(
                     "AGW request %s %s failed on attempt %d/%d: %s",
                     method,
@@ -553,7 +588,7 @@ class AsyncAGWCheckpointSaver(BaseCheckpointSaver):
                     exc,
                 )
                 if attempt >= self.retry_max_attempts:
-                    AGW_HTTP_FINAL_ERROR.labels(method, path).inc()
+                    AGW_HTTP_FINAL_ERROR.labels(method, metric_path).inc()
                     if self._client is not None:
                         try:
                             await self._client.aclose()
@@ -564,6 +599,7 @@ class AsyncAGWCheckpointSaver(BaseCheckpointSaver):
                             )
                         finally:
                             self._client = None
+                            self._client_loop = None
                     raise
 
                 if self._client is not None:
@@ -576,6 +612,7 @@ class AsyncAGWCheckpointSaver(BaseCheckpointSaver):
                         )
                     finally:
                         self._client = None
+                        self._client_loop = None
                 delay = self._compute_retry_delay(attempt)
                 if delay > 0:
                     await asyncio.sleep(delay)
@@ -584,16 +621,16 @@ class AsyncAGWCheckpointSaver(BaseCheckpointSaver):
 
             status = resp.status_code
             if status < 400 or status in ok_set:
-                AGW_HTTP_SUCCESS.labels(method, path).inc()
+                AGW_HTTP_SUCCESS.labels(method, metric_path).inc()
                 return resp
 
-            AGW_HTTP_ERROR.labels(method, path).inc()
+            AGW_HTTP_ERROR.labels(method, metric_path).inc()
             if status in (404, 406):
-                AGW_HTTP_FINAL_ERROR.labels(method, path).inc()
+                AGW_HTTP_FINAL_ERROR.labels(method, metric_path).inc()
                 return resp
 
             if attempt >= self.retry_max_attempts:
-                AGW_HTTP_FINAL_ERROR.labels(method, path).inc()
+                AGW_HTTP_FINAL_ERROR.labels(method, metric_path).inc()
                 return resp
 
             try:
@@ -611,6 +648,7 @@ class AsyncAGWCheckpointSaver(BaseCheckpointSaver):
                     )
                 finally:
                     self._client = None
+                    self._client_loop = None
             delay = self._compute_retry_delay(attempt)
             if delay > 0:
                 await asyncio.sleep(delay)
@@ -657,13 +695,15 @@ class AsyncAGWCheckpointSaver(BaseCheckpointSaver):
         tid = api_cfg["threadId"]
 
         if cid:
+            path_template = "/checkpoint/{threadId}/{checkpointId}"
             path = f"/checkpoint/{tid}/{cid}"
             params = {"checkpointNs": api_cfg.get("checkpointNs", "")}
         else:
+            path_template = "/checkpoint/{threadId}"
             path = f"/checkpoint/{tid}"
             params = None
 
-        resp = await self._http("GET", path, params=params)
+        resp = await self._http("GET", path, params=params, label_path=path_template)
         logger.debug("AGW aget_tuple response: %s", resp.text)
 
         if not resp.text:
@@ -687,7 +727,12 @@ class AsyncAGWCheckpointSaver(BaseCheckpointSaver):
             "before": self._to_api_config(before) if before else None,
             "limit": limit,
         }
-        resp = await self._http("POST", "/checkpoint/list", json=payload)
+        resp = await self._http(
+            "POST",
+            "/checkpoint/list",
+            json=payload,
+            label_path="/checkpoint/list",
+        )
         logger.debug("AGW alist response: %s", resp.text)
         resp.raise_for_status()
         for item in resp.json():
@@ -706,7 +751,7 @@ class AsyncAGWCheckpointSaver(BaseCheckpointSaver):
             "metadata": self._enc_meta(get_checkpoint_metadata(cfg, metadata)),
             "newVersions": new_versions,
         }
-        resp = await self._http("POST", "/checkpoint", json=payload)
+        resp = await self._http("POST", "/checkpoint", json=payload, label_path="/checkpoint")
         logger.debug("AGW aput response: %s", resp.text)
         resp.raise_for_status()
         return resp.json()["config"]
@@ -725,19 +770,31 @@ class AsyncAGWCheckpointSaver(BaseCheckpointSaver):
             "taskId": task_id,
             "taskPath": task_path,
         }
-        resp = await self._http("POST", "/checkpoint/writes", json=payload)
+        resp = await self._http(
+            "POST",
+            "/checkpoint/writes",
+            json=payload,
+            label_path="/checkpoint/writes",
+        )
         logger.debug("AGW aput_writes response: %s", resp.text)
         resp.raise_for_status()
 
     async def adelete_thread(self, thread_id: str) -> None:
-        resp = await self._http("DELETE", f"/checkpoint/{thread_id}")
+        resp = await self._http(
+            "DELETE",
+            f"/checkpoint/{thread_id}",
+            label_path="/checkpoint/{threadId}",
+        )
         resp.raise_for_status()
 
     # =================================================================
     # sync-обёртки
     # =================================================================
     def _run(self, coro):
-        return asyncio.run_coroutine_threadsafe(coro, self.loop).result()
+        # sync-обёртки всегда выполняем в собственном loop в отдельном потоке
+        loop = self._ensure_bg_loop()
+        fut = asyncio.run_coroutine_threadsafe(coro, loop)
+        return fut.result()
 
     def list(
         self,
@@ -747,12 +804,14 @@ class AsyncAGWCheckpointSaver(BaseCheckpointSaver):
         before: RunnableConfig | None = None,
         limit: int | None = None,
     ) -> Iterator[CheckpointTuple]:
-        aiter_ = self.alist(cfg, filter=filter, before=before, limit=limit)
-        while True:
-            try:
-                yield self._run(anext(aiter_))
-            except StopAsyncIteration:
-                break
+        async def _collect():
+            out = []
+            async for item in self.alist(cfg, filter=filter, before=before, limit=limit):
+                out.append(item)
+            return out
+
+        for item in self._run(_collect()):
+            yield item
 
     def get_tuple(self, cfg: RunnableConfig) -> CheckpointTuple | None:
         return self._run(self.aget_tuple(cfg))

{agent_lab_sdk-0.1.42.1 → agent_lab_sdk-0.1.44}/agent_lab_sdk.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agent-lab-sdk
-Version: 0.1.42.1
+Version: 0.1.44
 Summary: SDK для работы с Agent Lab
 Author-email: Andrew Ohurtsov <andermirik@yandex.com>
 License: Proprietary and Confidential — All Rights Reserved

{agent_lab_sdk-0.1.42.1 → agent_lab_sdk-0.1.44}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "agent-lab-sdk"
-version = "0.1.42.1"
+version = "0.1.44"
 description = "SDK для работы с Agent Lab"
 readme = "README.md"
 license = { text = "Proprietary and Confidential — All Rights Reserved" }