cleanup duplicate agents

Author: Shreyas-Microsoft

src/App/WebApp.Dockerfile

@@ -36,4 +36,4 @@ COPY --from=frontend /home/node/app/static  /usr/src/app/static/
 WORKDIR /usr/src/app  
 EXPOSE 80  
 
-CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "80", "--log-level", "info", "--access-log"]
+CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "80", "--workers", "1", "--log-level", "info", "--access-log"]

src/App/app.py

@@ -76,21 +76,21 @@ def create_app():
     app.config["TEMPLATES_AUTO_RELOAD"] = True
 
     # Setup agent initialization and cleanup
-    # @app.before_serving
-    # async def startup():
-    #     app.wealth_advisor_agent = await AgentFactory.get_wealth_advisor_agent()
-    #     logging.info("Wealth Advisor Agent initialized during application startup")
-    #     app.search_agent = await AgentFactory.get_search_agent()
-    #     logging.info(
-    #         "Call Transcript Search Agent initialized during application startup"
-    #     )
-
-    # @app.after_serving
-    # async def shutdown():
-    #     await AgentFactory.delete_all_agent_instance()
-    #     app.wealth_advisor_agent = None
-    #     app.search_agent = None
-    #     logging.info("Agents cleaned up during application shutdown")
+    @app.before_serving
+    async def startup():
+        app.wealth_advisor_agent = await AgentFactory.get_wealth_advisor_agent()
+        logging.info("Wealth Advisor Agent initialized during application startup")
+        app.search_agent = await AgentFactory.get_search_agent()
+        logging.info(
+            "Call Transcript Search Agent initialized during application startup"
+        )
+
+    @app.after_serving
+    async def shutdown():
+        await AgentFactory.delete_all_agent_instance()
+        app.wealth_advisor_agent = None
+        app.search_agent = None
+        logging.info("Agents cleaned up during application shutdown")
 
     # app.secret_key = secrets.token_hex(16)
     # app.session_interface = SecureCookieSessionInterface()
@@ -1359,4 +1359,4 @@ def get_users():
         return str(e), 500
 
 
-app = create_app()
+app = create_app()

src/App/backend/services/chat_service.py

@@ -1,37 +1,22 @@
-import logging
 from quart import current_app
 from semantic_kernel.agents import AzureAIAgent, AzureAIAgentThread
 from semantic_kernel.contents.chat_message_content import ChatMessageContent
 from semantic_kernel.contents.utils.author_role import AuthorRole
 
-from backend.agents.agent_factory import AgentFactory
 from backend.common.config import config
 from backend.services.sqldb_service import get_client_name_from_db
 
-async def ensure_agents(app):
-    """Ensure both agents are initialized on demand."""
-    if not hasattr(app, "wealth_advisor_agent") or app.wealth_advisor_agent is None:
-        app.wealth_advisor_agent = await AgentFactory.get_wealth_advisor_agent()
-        logging.info("Wealth Advisor Agent initialized on demand")
-    if not hasattr(app, "search_agent") or app.search_agent is None:
-        app.search_agent = await AgentFactory.get_search_agent()
-        logging.info("Search Agent initialized on demand")
-
-async def delete_agents(app):
-    """Delete both agents after use."""
-    await AgentFactory.delete_all_agent_instance()
-    app.wealth_advisor_agent = None
-    app.search_agent = None
 
 async def stream_response_from_wealth_assistant(query: str, client_id: str):
     """
     Streams real-time chat response from the Wealth Assistant.
     Uses Semantic Kernel agent with SQL and Azure Cognitive Search based on the client ID.
     """
-    thread = None
     try:
         # Dynamically get the name from the database
-        selected_client_name = get_client_name_from_db(client_id)
+        selected_client_name = get_client_name_from_db(
+            client_id
+        )  # Optionally fetch from DB
 
         # Prepare fallback instructions with the single-line prompt
         additional_instructions = config.STREAM_TEXT_SYSTEM_PROMPT
@@ -52,35 +37,28 @@ async def stream_response_from_wealth_assistant(query: str, client_id: str):
             "{client_id}", client_id
         )
 
-        # Lazy agent initialization
-        await ensure_agents(current_app)
         agent: AzureAIAgent = current_app.wealth_advisor_agent
 
+        thread: AzureAIAgentThread = None
         message = ChatMessageContent(role=AuthorRole.USER, content=query)
         sk_response = agent.invoke_stream(
             messages=[message],
-            thread=None,
+            thread=thread,
             additional_instructions=additional_instructions,
         )
 
         async def generate():
-            nonlocal thread
             try:
                 # yields deltaText strings one-by-one
                 async for chunk in sk_response:
                     if not chunk or not chunk.content:
                         continue
-                    thread = chunk.thread if hasattr(chunk, "thread") else None
                     yield chunk.content  # just the deltaText
             finally:
-                if thread:
-                    await thread.delete()
-                # Delete agents after operation
-                await delete_agents(current_app)
+                thread = chunk.thread if chunk else None
+                await thread.delete() if thread else None
 
         return generate
     except Exception as e:
-        if thread:
-            await thread.delete()
-        await delete_agents(current_app)
-        raise e
+        await thread.delete() if thread else None
+        raise e