Add Tracing

.devcontainer/Dockerfile

@@ -17,7 +17,7 @@ RUN sudo apt-get update && sudo apt-get install -y \
 # Install keyring-related and IPython packages in the same layer to reduce the image size and build time
 COPY ./src/api/requirements.txt .
 RUN pip install -r requirements.txt \    
-&& pip install keyrings.alt dbus-python ipython ipykernel mkdocs-material
+    && pip install keyrings.alt dbus-python ipython ipykernel mkdocs-material
 
 # Configure the IPython kernel
 RUN ipython kernel install --name "python3" --user

.github/workflows/evaluations.yaml

@@ -24,6 +24,27 @@ jobs:
       AZURE_CREDENTIALS: ${{ secrets.AZURE_CREDENTIALS }}
       AZURE_OPENAI_ENDPOINT: ${{ vars.AZURE_OPENAI_ENDPOINT }}
       AZURE_RESOURCE_GROUP: ${{ vars.AZURE_RESOURCE_GROUP }}
+      APPINSIGHTS_CONNECTIONSTRING: ${{ vars.APPINSIGHTS_CONNECTIONSTRING }}
+      AZURE_CONTAINER_ENVIRONMENT_NAME: ${{ vars.AZURE_CONTAINER_ENVIRONMENT_NAME }}
+      AZURE_CONTAINER_REGISTRY_ENDPOINT: ${{ vars.AZURE_CONTAINER_REGISTRY_ENDPOINT }}
+      AZURE_CONTAINER_REGISTRY_NAME: ${{ vars.AZURE_CONTAINER_REGISTRY_NAME }}
+      AZURE_COSMOS_NAME: ${{ vars.AZURE_COSMOS_NAME }}
+      AZURE_EMBEDDING_NAME: ${{ vars.AZURE_EMBEDDING_NAME }}
+      AZURE_ENV_NAME: ${{ vars.AZURE_ENV_NAME }}
+      AZURE_LOCATION: ${{ vars.AZURE_LOCATION }}
+      AZURE_OPENAI_API_VERSION: ${{ vars.AZURE_OPENAI_API_VERSION }}
+      AZURE_OPENAI_CHAT_DEPLOYMENT: ${{ vars.AZURE_OPENAI_CHAT_DEPLOYMENT }}
+      AZURE_OPENAI_NAME: ${{ vars.AZURE_OPENAI_NAME }}
+      AZURE_OPENAI_RESOURCE_GROUP_LOCATION: ${{ vars.AZURE_OPENAI_RESOURCE_GROUP_LOCATION }}
+      AZURE_SEARCH_ENDPOINT: ${{ vars.AZURE_SEARCH_ENDPOINT }}
+      AZURE_SEARCH_NAME: ${{ vars.AZURE_SEARCH_NAME }}
+      COSMOS_CONTAINER: ${{ vars.COSMOS_CONTAINER }}
+      COSMOS_ENDPOINT: ${{ vars.COSMOS_ENDPOINT }}
+      OPENAI_TYPE: ${{ vars.OPENAI_TYPE }}
+      SERVICE_ACA_IMAGE_NAME: ${{ vars.SERVICE_ACA_IMAGE_NAME }}
+      SERVICE_ACA_NAME: ${{ vars.SERVICE_ACA_NAME }}
+      SERVICE_ACA_URI: ${{ vars.SERVICE_ACA_URI }}
+
     steps:
       - name: checkout repo content
         uses: actions/checkout@v4 # checkout the repository content

src/api/contoso_chat/chat_request.py

@@ -41,7 +41,6 @@ def get_response(customerId, question, chat_history):
     customer = get_customer(customerId)
     print("customer complete")
     context = product.find_products(question)
-    print(context)
     print("products complete")
     print("getting result...")
 
@@ -55,9 +54,11 @@ def get_response(customerId, question, chat_history):
         inputs={"question": question, "customer": customer, "documentation": context},
         configuration=model_config,
     )
-    print("result: ", result)
     return {"question": question, "answer": result, "context": context}
 
 if __name__ == "__main__":
+    from tracing import init_tracing
+
+    tracer = init_tracing(local_tracing=False)
     get_response(4, "What hiking jackets would you recommend?", [])
     #get_response(argv[1], argv[2], argv[3])

src/api/contoso_chat/product/product.py

@@ -30,7 +30,6 @@ def generate_embeddings(queries: List[str]) -> str:
         api_version=os.environ["AZURE_OPENAI_API_VERSION"],
         azure_ad_token_provider=token_provider
     )
-    print("client:", client)
     embeddings = client.embeddings.create(input=queries, model="text-embedding-ada-002")
     embs = [emb.embedding for emb in embeddings.data]
     items = [{"item": queries[i], "embedding": embs[i]} for i in range(len(queries))]
@@ -79,7 +78,6 @@ def retrieve_products(items: List[Dict[str, any]], index_name: str) -> str:
 
 def find_products(context: str) -> Dict[str, any]:
     # Get product queries
-    print("context:", context)
     model_config = {
         "azure_endpoint": os.environ["AZURE_OPENAI_ENDPOINT"],
         "api_version": os.environ["AZURE_OPENAI_API_VERSION"],
@@ -89,13 +87,11 @@ def find_products(context: str) -> Dict[str, any]:
         configuration=model_config,
         inputs={"context":context}
         )
-    print("queries:", queries)
     qs = json.loads(queries)
     # Generate embeddings
     items = generate_embeddings(qs)
     # Retrieve products
     products = retrieve_products(items, "contoso-products")
-    print("products:", products)
     return products
 
 

src/api/evaluate.py

@@ -8,12 +8,15 @@
 from evaluators.custom_evals.groundedness import groundedness_evaluation
 import jsonlines
 import pandas as pd
+from prompty.tracer import trace
+from tracing import init_tracing
 from contoso_chat.chat_request import get_response
 
 # %% [markdown]
 # ## Get output from data and save to results jsonl file
 
 # %%
+@trace
 def load_data():
     data_path = "./evaluators/data.jsonl"
 
@@ -22,7 +25,7 @@ def load_data():
     return df
 
 # %%
-
+@trace
 def create_response_data(df):
     results = []
 
@@ -49,6 +52,7 @@ def create_response_data(df):
     return results
 
 # %%
+@trace
 def evaluate():
     # Evaluate results from results file
     results_path = 'result.jsonl'
@@ -88,6 +92,7 @@ def evaluate():
     return df
 
 # %%
+@trace
 def create_summary(df):
     print("Evaluation summary:\n")
     print(df)
@@ -105,7 +110,7 @@ def create_summary(df):
 # %%
 # create main funciton for python script
 if __name__ == "__main__":
-
+   tracer = init_tracing(local_tracing=True)
    test_data_df = load_data()
    response_results = create_response_data(test_data_df)
    result_evaluated = evaluate()

src/api/main.py

@@ -7,12 +7,14 @@
 from fastapi.responses import StreamingResponse
 from fastapi.middleware.cors import CORSMiddleware
 from opentelemetry.instrumentation.fastapi import FastAPIInstrumentor
+from tracing import init_tracing
 
 from contoso_chat.chat_request import get_response
 
 base = Path(__file__).resolve().parent
 
 load_dotenv()
+tracer = init_tracing()
 
 app = FastAPI()
 
@@ -24,7 +26,7 @@
     origin_5173 = f"https://{code_space}-5173.app.github.dev"
     ingestion_endpoint = app_insights.split(';')[1].split('=')[1]
 
-    origins = [origin_8000, origin_5173, os.getenv("API_SERVICE_ACA_URI"), os.getenv("WEB_SERVICE_ACA_URI"), ingestion_endpoint]
+    origins = [origin_8000, origin_5173, os.getenv("SERVICE_ACA_URI")]
 else:
     origins = [
         o.strip()

src/api/tracing.py

@@ -0,0 +1,64 @@
+import os
+import json
+import logging
+import contextlib
+from typing import AsyncIterator, List
+from prompty.tracer import Tracer, PromptyTracer
+from opentelemetry import trace as oteltrace
+from opentelemetry import trace
+from opentelemetry.sdk.trace import TracerProvider
+from opentelemetry.sdk.trace.export import BatchSpanProcessor
+from opentelemetry.sdk.trace.export import SimpleSpanProcessor
+from opentelemetry.sdk.trace.sampling import ParentBasedTraceIdRatio
+from azure.monitor.opentelemetry.exporter import AzureMonitorTraceExporter
+
+_tracer = "prompty"
+
+@contextlib.contextmanager
+def trace_span(name: str):    
+    tracer = oteltrace.get_tracer(_tracer)    
+    with tracer.start_as_current_span(name) as span:        
+        def verbose_trace(key, value):            
+            if isinstance(value, dict):             
+                for k, v in value.items():                  
+                    verbose_trace(f"{key}.{k}", v)        
+            elif isinstance(value, (list, tuple)):
+                for index, item in enumerate(value):
+                    span.set_attribute(f"{index}", str(item))  
+            else:                
+                span.set_attribute(f"{key}", value)     
+        yield verbose_trace
+
+
+def init_tracing(local_tracing: bool = False):
+    """
+    Initialize tracing for the application
+    If local_tracing is True, use the PromptyTracer
+    If remote_tracing is True, use the OpenTelemetry tracer
+    If remote_tracing is not specified, defaults to using the OpenTelemetry tracer only if local_tracing is False
+    """
+
+    if local_tracing:
+        local_trace = PromptyTracer()
+        Tracer.add("PromptyTracer", local_trace.tracer)
+        # Tracer.add("ConsoleTracer", console_tracer)
+    else:
+        Tracer.add("OpenTelemetry", trace_span)
+
+        azmon_logger = logging.getLogger("azure")
+        azmon_logger.setLevel(logging.INFO)
+
+        # oteltrace.set_tracer_provider(TracerProvider())
+
+        # Configure Azure Monitor as the Exporter
+        app_insights = os.getenv("APPINSIGHTS_CONNECTIONSTRING")
+
+        # Add the Azure exporter to the tracer provider
+
+        oteltrace.set_tracer_provider(TracerProvider(sampler=ParentBasedTraceIdRatio(1.0)))
+        oteltrace.get_tracer_provider().add_span_processor(BatchSpanProcessor(AzureMonitorTraceExporter(connection_string=app_insights)))
+        # oteltrace.get_tracer_provider().add_span_processor(
+        #     SimpleSpanProcessor(trace_exporter)
+        # )
+
+        return oteltrace.get_tracer(_tracer)