update

Author: aprilk-ms

infra/main.bicep

@@ -69,7 +69,7 @@ param storageAccountName string = ''
 @description('The log analytics workspace name. If ommited will be generated')
 param logAnalyticsWorkspaceName string = ''
 @description('Random seed to be used during generation of new resources suffixes.')
-param seed string = newGuid()
+param seed string = '' //newGuid()
 
 // Chat completion model
 @description('Format of the chat model to deploy')
@@ -418,6 +418,15 @@ module configStore 'core/config/configstore.bicep' = {
   }
 }
 
+module configStoreDataOwnerAccess 'core/security/role.bicep' = {
+  scope: rg
+  name: 'config-store-data-owner-role'
+  params: {
+    principalId: principalId
+    roleDefinitionId: '5ae67dd6-50cb-40e7-96ff-dc2bfa4b606b' // App Configuration Data Owner
+  }
+}
+
 // TODO: this will go away with auto-provisioning
 module onlineExperimentWorkspace 'core/config/onlineexperimentworkspace.bicep' = {
   name: 'online-experiment-workspace'

src/api/main.py

@@ -19,9 +19,6 @@
 from azure.identity import DefaultAzureCredential
 from azure.appconfiguration.provider import load
 
-from featuremanagement import FeatureManager
-from featuremanagement.azuremonitor import TargetingSpanProcessor,publish_telemetry
-
 from opentelemetry.instrumentation.fastapi import FastAPIInstrumentor
 
 from .search_index_manager import SearchIndexManager
@@ -64,10 +61,14 @@ async def lifespan(app: fastapi.FastAPI):
             exit()
         else:
             from azure.monitor.opentelemetry import configure_azure_monitor
+            from featuremanagement import FeatureManager
+            from featuremanagement.azuremonitor import publish_telemetry, TargetingSpanProcessor
+
             configure_azure_monitor(connection_string=application_insights_connection_string, span_processors=[TargetingSpanProcessor(targeting_context_accessor=get_targeting_context)])
 
-            # Inititalize the feature manager
+            # Inititalize the feature manager / TODO: Add null check
             app_config_conn_str = os.getenv("APP_CONFIGURATION_ENDPOINT") # this will become: project.experiments.get_connection_string()
+
             app_config = load(
                 endpoint=app_config_conn_str,
                 credential=DefaultAzureCredential(),
@@ -76,6 +77,7 @@ async def lifespan(app: fastapi.FastAPI):
                 refresh_interval=30,  # 30 seconds
             )
             feature_manager = FeatureManager(app_config, targeting_context_accessor=get_targeting_context, on_feature_evaluated=publish_telemetry)
+            app.state.feature_manager = feature_manager
 
     chat = await project.inference.get_chat_completions_client()
     embed = await project.inference.get_embeddings_client()
@@ -105,14 +107,16 @@ async def lifespan(app: fastapi.FastAPI):
     app.state.chat = chat
     app.state.search_index_manager = search_index_manager
     app.state.chat_model = os.environ["AZURE_AI_CHAT_DEPLOYMENT_NAME"]
-    app.state.feature_manager = feature_manager
+    
 
     yield
 
     await project.close()
-
     await chat.close()
 
+    if search_index_manager is not None:
+        await search_index_manager.close()
+
 # Below will be replaced by a helper function from App Config SDK
 
 # class TargetingSpanProcessor(SpanProcessor):
@@ -171,6 +175,7 @@ def create_app():
 
     app.include_router(routes.router)
 
-    FastAPIInstrumentor.instrument_app(app) #, server_request_hook=server_request_hook)
+    # TODO: do we need this?
+    #FastAPIInstrumentor.instrument_app(app) #, server_request_hook=server_request_hook)
 
     return app

src/api/routes.py

@@ -21,7 +21,7 @@
 from .util import get_logger, ChatRequest
 from .search_index_manager import SearchIndexManager
 from azure.core.exceptions import HttpResponseError
-
+import pydantic 
 
 logger = get_logger(
     name="azureaiapp_routes",
@@ -41,7 +41,6 @@
 router = fastapi.APIRouter()
 templates = Jinja2Templates(directory="api/templates")
 
-
 # Accessors to get app state
 def get_chat_client(request: Request) -> ChatCompletionsClient:
     return request.app.state.chat
@@ -79,15 +78,13 @@ async def chat_stream_handler(
     chat_client: ChatCompletionsClient = Depends(get_chat_client),
     model_deployment_name: str = Depends(get_chat_model),
     search_index_manager: SearchIndexManager = Depends(get_search_index_namager),
-    feature_manager: FeatureManager = Depends
+    feature_manager: FeatureManager = Depends(get_feature_manager)
 ) -> fastapi.responses.StreamingResponse:
     if chat_client is None:
         raise Exception("Chat client not initialized")
 
     async def response_stream():
         messages = [{"role": message.role, "content": message.content} for message in chat_request.messages]
-        model_deployment_name = globals["chat_model"]
-        feature_manager = globals["feature_manager"] 
 
         targeting_id = chat_request.sessionState.get('sessionId', str(uuid.uuid4()))
         attach(set_baggage("Microsoft.TargetingId", targeting_id))
@@ -107,23 +104,7 @@ async def response_stream():
         chat_coroutine = await chat_client.complete(
             model=model_deployment_name, messages=prompt_messages + messages, stream=True
         )
-        async for event in chat_coroutine:
-            if event.choices:
-                first_choice = event.choices[0]
-                yield (
-                    json.dumps(
-                        {
-                            "delta": {
-                                "content": first_choice.delta.content,
-                                "role": first_choice.delta.role,
-                            }
-                        },
-                        ensure_ascii=False,
-                    )
-                    + "\n"
-                )
 
-        prompt_messages = PromptTemplate.from_string('You are a helpful assistant').create_messages()
         # Use RAG model, only if we were provided index and we have found a context there.
         if search_index_manager is not None:
             context = await search_index_manager.search(chat_request)
@@ -135,6 +116,7 @@ async def response_stream():
                 logger.info(f"{prompt_messages=}")
             else:
                 logger.info("Unable to find the relevant information in the index for the request.")
+                
         try:
             chat_coroutine = await chat_client.complete(
                 model=model_deployment_name, messages=prompt_messages + messages, stream=True