Add support for OTLP exported metrics and tracing

app/backend/app.py

@@ -413,8 +413,7 @@ def create_app():
     app = Quart(__name__)
     app.register_blueprint(bp)
 
-    if os.getenv("APPLICATIONINSIGHTS_CONNECTION_STRING"):
-        configure_azure_monitor()
+    def instrument_app():
         # This tracks HTTP requests made by aiohttp:
         AioHttpClientInstrumentor().instrument()
         # This tracks HTTP requests made by httpx:
@@ -424,6 +423,19 @@ def create_app():
         # This middleware tracks app route requests:
         app.asgi_app = OpenTelemetryMiddleware(app.asgi_app)  # type: ignore[method-assign]
 
+    if os.getenv("APPLICATIONINSIGHTS_CONNECTION_STRING"):
+        configure_azure_monitor()
+        instrument_app()
+    elif os.getenv("OTLP_GRPC_ENDPOINT"):
+        from app.backend.otlp_tracing import configure_oltp_grpc_tracing
+        configure_oltp_grpc_tracing(endpoint=os.getenv("OTLP_GRPC_ENDPOINT"))
+        instrument_app()
+    elif os.getenv("OTLP_HTTP_ENDPOINT"):
+        from app.backend.otlp_tracing import configure_oltp_http_tracing
+        configure_oltp_http_tracing(endpoint=os.getenv("OTLP_HTTP_ENDPOINT"))
+        instrument_app()
+
+
     # Level should be one of https://docs.python.org/3/library/logging.html#logging-levels
     default_level = "INFO"  # In development, log more verbosely
     if os.getenv("WEBSITE_HOSTNAME"):  # In production, don't log as heavily

app/backend/approaches/chatreadretrieveread.py

@@ -14,6 +14,8 @@
 from core.authentication import AuthenticationHelper
 from core.modelhelper import get_token_limit
 
+from opentelemetry import metrics
+
 
 class ChatReadRetrieveReadApproach(ChatApproach):
     """
@@ -49,6 +51,8 @@ def __init__(
         self.query_language = query_language
         self.query_speller = query_speller
         self.chatgpt_token_limit = get_token_limit(chatgpt_model)
+        self.meter_ratelimit_remaining_tokens = metrics.get_meter(__name__).create_gauge("openai.ratelimit-remaining-tokens", description="OpenAI tokens remaining in limit")
+        self.meter_ratelimit_remaining_requests = metrics.get_meter(__name__).create_gauge("openai.ratelimit-remaining-requests", description="OpenAI remaining requests")
 
     @property
     def system_message_chat_conversation(self):
@@ -128,7 +132,7 @@ async def run_until_final_call(
             few_shots=self.query_prompt_few_shots,
         )
 
-        chat_completion: ChatCompletion = await self.openai_client.chat.completions.create(
+        chat_completion_response = await self.openai_client.chat.completions.with_raw_response.create(
             messages=query_messages,  # type: ignore
             # Azure OpenAI takes the deployment name as the model name
             model=self.chatgpt_deployment if self.chatgpt_deployment else self.chatgpt_model,
@@ -138,6 +142,9 @@ async def run_until_final_call(
             tools=tools,
             tool_choice="auto",
         )
+        self.meter_ratelimit_remaining_tokens.set(int(chat_completion_response.headers.get("x-ratelimit-remaining-tokens", 0)))
+        self.meter_ratelimit_remaining_requests.set(int(chat_completion_response.headers.get("x-ratelimit-remaining-requests", 0)))
+        chat_completion = chat_completion_response.parse()
 
         query_text = self.get_search_query(chat_completion, original_user_query)
 

app/backend/otlp_tracing.py

@@ -0,0 +1,51 @@
+from opentelemetry.sdk.resources import SERVICE_NAME, Resource
+
+from opentelemetry import trace
+from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter as OTLPHTTPSpanExporter
+from opentelemetry.exporter.otlp.proto.grpc.trace_exporter import OTLPSpanExporter as OTLPGRPCSpanExporter
+from opentelemetry.sdk.trace import TracerProvider
+from opentelemetry.sdk.trace.export import BatchSpanProcessor
+
+from opentelemetry import metrics
+from opentelemetry.exporter.otlp.proto.http.metric_exporter import OTLPMetricExporter as OTLPHTTPMetricExporter
+from opentelemetry.exporter.otlp.proto.grpc.metric_exporter import OTLPMetricExporter as OTLPGRPCMetricExporter
+from opentelemetry.sdk.metrics import MeterProvider
+from opentelemetry.sdk.metrics.export import PeriodicExportingMetricReader
+
+
+def configure_oltp_http_tracing(service_name: str = "azure-search-openai-demo", endpoint: str = "http://localhost:4318"):
+    # Service name is required for most backends
+    resource = Resource(attributes={
+        SERVICE_NAME: service_name
+    })
+
+    traceProvider = TracerProvider(resource=resource)
+    # Trick OLTP into thinking the httpx client is a requests session so that HTTP2 works
+    processor = BatchSpanProcessor(OTLPHTTPSpanExporter(endpoint=f"{endpoint}/v1/traces",))
+    traceProvider.add_span_processor(processor)
+    trace.set_tracer_provider(traceProvider)
+
+    reader = PeriodicExportingMetricReader(
+        OTLPHTTPMetricExporter(endpoint=f"{endpoint}/v1/metrics")
+    )
+    meterProvider = MeterProvider(resource=resource, metric_readers=[reader])
+    metrics.set_meter_provider(meterProvider)
+
+
+def configure_oltp_grpc_tracing(service_name: str = "azure-search-openai-demo", endpoint: str = "http://localhost:4317"):
+    # Service name is required for most backends
+    resource = Resource(attributes={
+        SERVICE_NAME: service_name
+    })
+
+    traceProvider = TracerProvider(resource=resource)
+    # Trick OLTP into thinking the httpx client is a requests session so that HTTP2 works
+    processor = BatchSpanProcessor(OTLPGRPCSpanExporter(endpoint=endpoint, insecure=True))
+    traceProvider.add_span_processor(processor)
+    trace.set_tracer_provider(traceProvider)
+
+    reader = PeriodicExportingMetricReader(
+        OTLPGRPCMetricExporter(endpoint=endpoint, insecure=True)
+    )
+    meterProvider = MeterProvider(resource=resource, metric_readers=[reader])
+    metrics.set_meter_provider(meterProvider)

app/backend/requirements.in

@@ -14,6 +14,9 @@ opentelemetry-instrumentation-httpx
 opentelemetry-instrumentation-requests
 opentelemetry-instrumentation-aiohttp-client
 opentelemetry-instrumentation-openai
+opentelemetry-exporter-otlp-proto-grpc
+opentelemetry-exporter-otlp-proto-http
+
 msal
 azure-keyvault-secrets
 cryptography