crewplus 0.2.70__py3-none-any.whl → 0.2.72__py3-none-any.whl

crewplus/services/tracing_manager.py

@@ -1,6 +1,6 @@
 # File: crewplus/services/tracing_manager.py
 
-from typing import Any, Optional, List, Protocol
+from typing import Any, Optional, List, Protocol, Dict
 import os
 import logging
 
@@ -9,11 +9,14 @@ import logging
 try:
     from langfuse.langchain import CallbackHandler as LangfuseCallbackHandler
     from ..callbacks.async_langfuse_handler import AsyncLangfuseCallbackHandler
+    from ..utils.tracing_util import get_langfuse_handler, get_async_langfuse_handler
     LANGFUSE_AVAILABLE = True
 except ImportError:
     LANGFUSE_AVAILABLE = False
     LangfuseCallbackHandler = None
     AsyncLangfuseCallbackHandler = None
+    get_langfuse_handler = None
+    get_async_langfuse_handler = None
 
 class TracingContext(Protocol):
     """
@@ -102,11 +105,11 @@ class TracingManager:
         if enable_langfuse:
             try:
                 # Create both sync and async handlers. We'll pick one at runtime.
-                sync_handler = LangfuseCallbackHandler()
+                sync_handler = get_langfuse_handler()
                 self._sync_handlers.append(sync_handler)
 
                 if AsyncLangfuseCallbackHandler:
-                    async_handler = AsyncLangfuseCallbackHandler()
+                    async_handler = get_async_langfuse_handler()
                     self._async_handlers.append(async_handler)
                 
                 self.context.logger.info(f"Langfuse tracing enabled for {self.context.get_model_identifier()}")
@@ -115,7 +118,7 @@ class TracingManager:
         else:
             self.context.logger.info("Langfuse is not enabled, skipping handler initialization.")
     
-    def _add_callbacks_to_config(self, config: Optional[dict], handlers: List[Any]) -> dict:
+    def add_callbacks_to_config(self, config: Optional[dict], handlers: List[Any]) -> dict:
         """A generic helper to add a list of handlers to a config object."""
         if config is None:
             config = {}
@@ -154,10 +157,26 @@ class TracingManager:
         
         return config
 
-    def add_sync_callbacks_to_config(self, config: Optional[dict]) -> dict:
-        """Adds synchronous tracing handlers to the request configuration."""
-        return self._add_callbacks_to_config(config, self._sync_handlers)
+    def add_sync_callbacks_to_config(self, config: Optional[dict], handlers: Optional[List[Any]] = None) -> dict:
+        """
+        Adds synchronous tracing handlers to the request configuration.
+
+        Args:
+            config: The configuration dictionary to which callbacks will be added.
+            handlers: An optional list of handlers to add. If not provided,
+                      the manager's default synchronous handlers are used.
+        """
+        handlers_to_add = self._sync_handlers if handlers is None else handlers
+        return self.add_callbacks_to_config(config, handlers_to_add)
 
-    def add_async_callbacks_to_config(self, config: Optional[dict]) -> dict:
-        """Adds asynchronous tracing handlers to the request configuration."""
-        return self._add_callbacks_to_config(config, self._async_handlers)
+    def add_async_callbacks_to_config(self, config: Optional[dict], handlers: Optional[List[Any]] = None) -> dict:
+        """
+        Adds asynchronous tracing handlers to the request configuration.
+
+        Args:
+            config: The configuration dictionary to which callbacks will be added.
+            handlers: An optional list of handlers to add. If not provided,
+                      the manager's default asynchronous handlers are used.
+        """
+        handlers_to_add = self._async_handlers if handlers is None else handlers
+        return self.add_callbacks_to_config(config, handlers_to_add)

crewplus/utils/tracing_util.py

@@ -0,0 +1,51 @@
+from typing import Dict, Any, Optional
+from langchain_core.runnables import RunnableConfig
+from langfuse.langchain import CallbackHandler as LangfuseCallbackHandler
+from ..callbacks.async_langfuse_handler import AsyncLangfuseCallbackHandler
+
+# Singleton holder for the Langfuse handler to avoid multiple instances per run
+_LANGFUSE_HANDLER: Optional[LangfuseCallbackHandler] = None
+_ASYNC_LANGFUSE_HANDLER: Optional[AsyncLangfuseCallbackHandler] = None
+
+def _get_langfuse_handler() -> LangfuseCallbackHandler:
+    global _LANGFUSE_HANDLER
+    if _LANGFUSE_HANDLER is None:
+        _LANGFUSE_HANDLER = LangfuseCallbackHandler()
+    return _LANGFUSE_HANDLER
+
+def get_langfuse_handler() -> LangfuseCallbackHandler:
+    return _get_langfuse_handler()
+
+def get_async_langfuse_handler() -> "AsyncLangfuseCallbackHandler":
+    """Returns a singleton instance of the async Langfuse handler."""
+    global _ASYNC_LANGFUSE_HANDLER
+    if _ASYNC_LANGFUSE_HANDLER is None:
+        _ASYNC_LANGFUSE_HANDLER = AsyncLangfuseCallbackHandler()
+    return _ASYNC_LANGFUSE_HANDLER
+
+def prepare_trace_config(context: Dict[str, Any]) -> RunnableConfig:
+    """
+    Prepares a minimal RunnableConfig for tracing, primarily for Langfuse.
+
+    - Creates a new config containing only tracing-related information.
+    - Extracts 'trace_metadata' from the context's 'configurable' dict
+      and uses it as the 'metadata' for the new trace config.
+    - Adds a singleton Langfuse callback handler.
+    """
+    # The full config is passed in the 'config' key of the context
+    # Start with a copy of the existing config from the graph to preserve its state
+    run_config = context.get("config", {}).copy()
+
+    # Extract trace_metadata from the 'configurable' part of the full config
+    trace_metadata = run_config.get("trace_metadata", {})
+    if not trace_metadata:
+        trace_metadata = run_config.get("configurable", {}).get("trace_metadata", {})
+
+    # If trace_metadata exists, merge all its fields into the main metadata key
+    if trace_metadata and isinstance(trace_metadata, dict):
+        if "metadata" not in run_config:
+            run_config["metadata"] = {}
+        run_config["metadata"].update(trace_metadata)
+
+    return run_config
+    

crewplus/vectorstores/milvus/vdb_service.py

@@ -5,14 +5,15 @@
 # @Last Modified time: 2025-10-09
 
 import logging
-from typing import List, Dict, Union, Optional
+from typing import List, Dict, Union, Optional, Any
 from langchain_milvus import Milvus
 from langchain_core.embeddings import Embeddings
 from langchain_openai import AzureOpenAIEmbeddings
-from pymilvus import MilvusClient, AsyncMilvusClient
+from pymilvus import MilvusClient, AsyncMilvusClient, connections
 import time
 import asyncio
 import uuid
+from collections import defaultdict
 
 from ...services.init_services import get_model_balancer
 from .schema_milvus import SchemaMilvus, DEFAULT_SCHEMA
@@ -96,6 +97,7 @@ class VDBService(object):
     _async_client: Optional[AsyncMilvusClient] = None
     _instances: Dict[str, Milvus] = {}
     _async_instances: Dict[str, Milvus] = {}
+    _async_instance_locks: Dict[str, asyncio.Lock] = defaultdict(asyncio.Lock)
 
     schema: str
     embedding_function: Embeddings
@@ -168,6 +170,10 @@ class VDBService(object):
         # lazy-initialize async milvus
         # self._async_client = self._initialize_async_milvus_client(provider)
 
+        # Do not initialize the async client here.
+        # It must be lazily initialized within an async context.
+        self._async_client: Optional[AsyncMilvusClient] = None
+        
         self.schema = schema
         self.index_params = self.settings.get("index_params")
         
@@ -256,12 +262,16 @@ class VDBService(object):
 
     async def aget_async_vector_client(self) -> AsyncMilvusClient:
         """
-        Asynchronously returns the active AsyncMilvusClient instance, initializing it if necessary.
-
-        Returns:
-            AsyncMilvusClient: The initialized async client for interacting with the vector database.
+        Lazily initializes and returns the AsyncMilvusClient.
+        This ensures the client is created within the running event loop.
         """
-        return await self._get_or_create_async_client()
+        if self._async_client is None:
+            self.logger.info("Lazily initializing AsyncMilvusClient...")
+            client_args = self._get_milvus_client_args(self._provider)
+            # Use the dedicated async alias
+            client_args['alias'] = self.async_alias
+            self._async_client = AsyncMilvusClient(**client_args)
+        return self._async_client
 
     def get_vector_field(self, collection_name: str) -> str:
         """
@@ -370,6 +380,7 @@ class VDBService(object):
         Asynchronously checks if a collection exists and creates it if it doesn't.
         """
         try:
+            # Call the new lazy initializer for the async client
             client = await self.aget_async_vector_client()
             if check_existence and not await client.has_collection(collection_name):
                 self.logger.info(f"Collection '{collection_name}' does not exist. Creating it.")
@@ -498,130 +509,117 @@ class VDBService(object):
 
     async def _get_or_create_async_client(self) -> AsyncMilvusClient:
         """
-        Lazily initializes and returns the AsyncMilvusClient.
-        This runs the blocking constructor in a separate thread, but also creates
-        a temporary event loop inside that thread to satisfy the client's
-        initialization requirements.
+        Lazily initializes the AsyncMilvusClient.
+        Based on grpcio source, the client MUST be initialized in a thread
+        with a running event loop. Therefore, we initialize it directly in the
+        main async context. The synchronous __init__ is fast enough not to
+        block the event loop meaningfully.
         """
         if self._async_client is None:
-            self.logger.info("Lazily initializing AsyncMilvusClient...")
-            
-            def _create_with_loop():
-                # This function runs in a separate thread via asyncio.to_thread
-                try:
-                    # Check if an event loop exists in this new thread
-                    asyncio.get_running_loop()
-                except RuntimeError:  # 'RuntimeError: There is no current event loop...'
-                    # If not, create and set a new one
-                    loop = asyncio.new_event_loop()
-                    asyncio.set_event_loop(loop)
-
-                # Now, with an event loop present in this thread, initialize the client.
-                # This is still a blocking call, but it's contained in the thread.
-                provider = self.settings.get("vector_store", {}).get("provider")
-                return self._initialize_async_milvus_client(provider)
-
-            self._async_client = await asyncio.to_thread(_create_with_loop)
+            self.logger.info("Lazily initializing AsyncMilvusClient directly in the main event loop...")
+            provider = self.settings.get("vector_store", {}).get("provider")
+            # This is a synchronous call, but it's lightweight and must run here.
+            self._async_client = self._initialize_async_milvus_client(provider)
             
         return self._async_client
 
     async def aget_vector_store(self, collection_name: str, embeddings: Embeddings = None, metric_type: str = "IP") -> Milvus:
         """
         Asynchronously gets a vector store instance, creating it if it doesn't exist.
+        This version is optimized to handle high concurrency using a lock.
         """
         if not collection_name:
             self.logger.error("aget_vector_store called with no collection_name.")
             raise ValueError("collection_name must be provided.")
 
-        # Check for a cached instance. If found, return it immediately.
-        if collection_name in self._async_instances:
-            self.logger.info(f"Returning existing async vector store instance for collection: {collection_name}")
-            return self._async_instances[collection_name]
-
-        self.logger.info(f"Creating new async vector store instance for collection: {collection_name}")
-        if embeddings is None:
-            embeddings = self.get_embeddings()
+        lock = self._async_instance_locks[collection_name]
+        async with lock:
+            if collection_name in self._async_instances:
+                self.logger.info(f"Returning existing async vector store instance for collection: {collection_name} (post-lock)")
+                return self._async_instances[collection_name]
 
-        await self._aensure_collection_exists(collection_name, embeddings, check_existence=True)
+            self.logger.info(f"Creating new async vector store instance for collection: {collection_name}")
+            if embeddings is None:
+                embeddings = self.get_embeddings()
 
-        # try:
-        #     self.logger.info(f"Testing embedding function for collection '{collection_name}'...")
-        #     await embeddings.aembed_query("validation_test_string")
-        #     self.logger.info("Embedding function is valid.")
-        # except Exception as e:
-        #     self.logger.error(
-        #         f"The provided embedding function is invalid and failed with error: {e}. "
-        #         f"Cannot create a vector store for collection '{collection_name}'."
-        #     )
-        #     raise RuntimeError(f"Invalid embedding function provided.") from e
+            # CRITICAL: Ensure the shared async client is initialized *under the lock*
+            # before any operation that might use it.
+            await self._get_or_create_async_client()
 
-        index_params = self.index_params or {
-            "metric_type": metric_type,
-            "index_type": "AUTOINDEX",
-            "params": {}
-        }
-        
-        # Create a dedicated connection_args for the async path with the correct alias
-        async_conn_args = self.connection_args.copy()
-        async_conn_args['alias'] = self.async_alias
-        
-        # For async operations, we MUST instantiate the Milvus object using the SYNCHRONOUS alias
-        # because its __init__ method is synchronous. This is now done in a separate thread.
-        vdb = await self._acreate_milvus_instance_with_retry(
-            collection_name=collection_name,
-            embeddings=embeddings,
-            index_params=index_params,
-            connection_args=async_conn_args  # Pass the async-specific connection args
-        )
+            await self._aensure_collection_exists(collection_name, embeddings, check_existence=True)
 
-        # After successful synchronous initialization, we hot-swap the alias on the
-        # ASYNCHRONOUS client to ensure future async operations use the correct connection.
-        self.logger.info(f"Swapping to async alias for instance of collection {collection_name}")
-        # DO NOT get the async client here, get it outside this function
-        #await self._get_or_create_async_client()
-        vdb.aclient._using = self.async_alias
+            vdb = await self._acreate_milvus_instance_with_retry(
+                collection_name=collection_name,
+                embeddings=embeddings,
+                metric_type=metric_type
+            )
 
-        self._async_instances[collection_name] = vdb
+            self.logger.info(f"Swapping to async alias for instance of collection {collection_name}")
+            vdb.aclient._using = self.async_alias
 
-        return vdb
+            self._async_instances[collection_name] = vdb
+            return vdb
 
-    async def _acreate_milvus_instance_with_retry(self, collection_name: str, embeddings: Embeddings, index_params: dict, connection_args: Optional[dict] = None) -> Milvus:
+    async def _acreate_milvus_instance_with_retry(
+        self,
+        embeddings: Embeddings,
+        collection_name: str,
+        metric_type: str = "IP",
+    ) -> Milvus:
         """
-        Asynchronously creates a Milvus instance with a retry mechanism, running the synchronous
-        constructor in a separate thread to avoid blocking the event loop.
+        Asynchronously creates a Milvus instance with retry logic, ensuring the connection
+        is established in the target thread.
         """
-        retries = 2
-        conn_args = connection_args if connection_args is not None else self.connection_args
-        
-        def _create_instance():
-            # This synchronous function will be run in a thread
-            return Milvus(
-                embedding_function=embeddings,
-                collection_name=collection_name,
-                connection_args=conn_args,
-                index_params=index_params
-            )
-
-        self.logger.info(f"Creating Milvus instance for collection '{collection_name}' in a separate thread...")
-        self.logger.info(f"Connection args: {conn_args}")
+        retries = 3
+        last_exception = None
 
-        for attempt in range(retries + 1):
+        for attempt in range(retries):
             try:
-                # Run the blocking constructor in a separate thread
-                vdb = await asyncio.to_thread(_create_instance)
-                
-                self.logger.info(f"Successfully connected to Milvus for collection '{collection_name}' on attempt {attempt + 1}.")
-                return vdb  # Return on success
+                conn_args = self.connection_args.copy()
+                # Langchain's Milvus class will use the alias to find the connection.
+                conn_args["alias"] = self.sync_alias
+
+                def _create_instance_in_thread():
+                    # --- START: CRITICAL FIX ---
+                    # Manually connect within the thread before creating the Milvus instance.
+                    # This ensures pymilvus registers the connection details for the current thread.
+                    try:
+                        connections.connect(**conn_args)
+                        self.logger.info(f"Successfully connected to Milvus with alias '{self.sync_alias}' in thread.")
+                    except Exception as e:
+                        self.logger.error(f"Failed to manually connect in thread: {e}")
+                        raise
+
+                    # Now, creating the Milvus instance will find the existing connection via the alias.
+                    instance = Milvus(
+                        embedding_function=embeddings,
+                        collection_name=collection_name,
+                        connection_args=conn_args, # Pass args for completeness
+                        # metric_type=metric_type,  # <-- CRITICAL FIX: REMOVE THIS LINE
+                        consistency_level="Strong",
+                        # --- START: CRITICAL FIX ---
+                        # Pass self.index_params to the Milvus constructor here
+                        index_params=self.index_params,
+                        # --- END: CRITICAL FIX ---
+                    )
+                    return instance
+                    # --- END: CRITICAL FIX ---
+
+                self.logger.info(f"Attempt {attempt + 1}/{retries}: Creating Milvus instance for collection '{collection_name}' in a separate thread...")
+                vdb = await asyncio.to_thread(_create_instance_in_thread)
+                self.logger.info("Successfully created Milvus instance.")
+                return vdb
+
             except Exception as e:
+                last_exception = e
                 self.logger.warning(
-                    f"Attempt {attempt + 1}/{retries + 1} to connect to Milvus for collection '{collection_name}' failed: {e}"
+                    f"Attempt {attempt + 1}/{retries} failed to create Milvus instance: {e}. Retrying in {2 ** attempt}s..."
                 )
-                if attempt < retries:
-                    self.logger.info("Retrying in 3 seconds...")
-                    await asyncio.sleep(3) # Use async sleep
-                else:
-                    self.logger.error(f"Failed to connect to Milvus for collection '{collection_name}' after {retries + 1} attempts.")
-                    raise RuntimeError(f"Could not connect to Milvus after {retries + 1} attempts.") from e
+                await asyncio.sleep(2 ** attempt)
+        
+        raise RuntimeError(
+            f"Failed to create Milvus instance after {retries} retries."
+        ) from last_exception
 
     def _create_milvus_instance_with_retry(self, collection_name: str, embeddings: Embeddings, index_params: dict, connection_args: Optional[dict] = None) -> Milvus:
         """

{crewplus-0.2.70.dist-info → crewplus-0.2.72.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: crewplus
-Version: 0.2.70
+Version: 0.2.72
 Summary: Base services for CrewPlus AI applications
 Author-Email: Tim Liu <tim@opsmateai.com>
 License: MIT

{crewplus-0.2.70.dist-info → crewplus-0.2.72.dist-info}/RECORD

@@ -1,7 +1,7 @@
-crewplus-0.2.70.dist-info/METADATA,sha256=M2xMdWJbR5BsBLfIGxner83D2vLfE5ui-hmEmhcVMUQ,5424
-crewplus-0.2.70.dist-info/WHEEL,sha256=9P2ygRxDrTJz3gsagc0Z96ukrxjr-LFBGOgv3AuKlCA,90
-crewplus-0.2.70.dist-info/entry_points.txt,sha256=6OYgBcLyFCUgeqLgnvMyOJxPCWzgy7se4rLPKtNonMs,34
-crewplus-0.2.70.dist-info/licenses/LICENSE,sha256=2_NHSHRTKB_cTcT_GXgcenOCtIZku8j343mOgAguTfc,1087
+crewplus-0.2.72.dist-info/METADATA,sha256=RYFh6zu4R5RZg4Ojlp2jIKOV61A7DNvwu_4s1SDu1X0,5424
+crewplus-0.2.72.dist-info/WHEEL,sha256=9P2ygRxDrTJz3gsagc0Z96ukrxjr-LFBGOgv3AuKlCA,90
+crewplus-0.2.72.dist-info/entry_points.txt,sha256=6OYgBcLyFCUgeqLgnvMyOJxPCWzgy7se4rLPKtNonMs,34
+crewplus-0.2.72.dist-info/licenses/LICENSE,sha256=2_NHSHRTKB_cTcT_GXgcenOCtIZku8j343mOgAguTfc,1087
 crewplus/__init__.py,sha256=m46HkZL1Y4toD619NL47Sn2Qe084WFFSFD7e6VoYKZc,284
 crewplus/callbacks/__init__.py,sha256=YG7ieeb91qEjp1zF0-inEN7mjZ7yT_D2yzdWFT8Z1Ws,63
 crewplus/callbacks/async_langfuse_handler.py,sha256=A4uFeLpvOUdc58M7sZoE65_C1V98u0QCvx5jUquM0pM,7006
@@ -10,16 +10,17 @@ crewplus/services/azure_chat_model.py,sha256=iWzJ2GQFSNmwJx-2O5_xKPSB6VVc-7T6bcf
 crewplus/services/gemini_chat_model.py,sha256=DYqz01H2TIHiCDQesSozVfOsMigno6QGwOtIweg7UHk,40103
 crewplus/services/init_services.py,sha256=tc1ti8Yufo2ixlJpwg8uH0KmoyQ4EqxCOe4uTEWnlRM,2413
 crewplus/services/model_load_balancer.py,sha256=Q9Gx3GrbKworU-Ytxeqp0ggHSgZ1Q6brtTk-nCl4sak,12095
-crewplus/services/tracing_manager.py,sha256=_C4zYj6o_k5mDWY7vM8UeaVaXp8SqrkvYOb0Jj-y3sY,7566
+crewplus/services/tracing_manager.py,sha256=pwNFeA77vnoZMh_AUOnK5TvAaPOOLg5oDnVOe1yUa9A,8502
 crewplus/utils/__init__.py,sha256=2Gk1n5srFJQnFfBuYTxktdtKOVZyNrFcNaZKhXk35Pw,142
 crewplus/utils/schema_action.py,sha256=GDaBoVFQD1rXqrLVSMTfXYW1xcUu7eDcHsn57XBSnIg,422
 crewplus/utils/schema_document_updater.py,sha256=frvffxn2vbi71fHFPoGb9hq7gH2azmmdq17p-Fumnvg,7322
+crewplus/utils/tracing_util.py,sha256=39F9ydMjXSy5UbQIjYDKc1yiBVnibuWt0cGpYM55k7Q,2281
 crewplus/vectorstores/milvus/__init__.py,sha256=OeYv2rdyG7tcREIjBJPyt2TbE54NvyeRoWMe7LwopRE,245
 crewplus/vectorstores/milvus/milvus_schema_manager.py,sha256=-QRav-hzu-XWeJ_yKUMolal_EyMUspSg-nvh5sqlrlQ,11442
 crewplus/vectorstores/milvus/schema_milvus.py,sha256=wwNpfqsKS0xeozZES40IvB0iNwUtpCall_7Hkg0dL1g,27223
-crewplus/vectorstores/milvus/vdb_service.py,sha256=Jo2GWzdEuDSPlADCWA7wTgDbYQ65QiYWgFGbtkG9vG8,37789
+crewplus/vectorstores/milvus/vdb_service.py,sha256=U8I6IUYZK0gCe1R9rTnVezvZfEcUS9UEbKEoeJPX8kY,37528
 docs/GeminiChatModel.md,sha256=zZYyl6RmjZTUsKxxMiC9O4yV70MC4TD-IGUmWhIDBKA,8677
 docs/ModelLoadBalancer.md,sha256=aGHES1dcXPz4c7Y8kB5-vsCNJjriH2SWmjBkSGoYKiI,4398
 docs/VDBService.md,sha256=Dw286Rrf_fsi13jyD3Bo4Sy7nZ_G7tYm7d8MZ2j9hxk,9375
 docs/index.md,sha256=3tlc15uR8lzFNM5WjdoZLw0Y9o1P1gwgbEnOdIBspqc,1643
-crewplus-0.2.70.dist-info/RECORD,,
+crewplus-0.2.72.dist-info/RECORD,,