Merge pull request #135 from getjavelin/feature/openai-register

feat: Refactor JavelinClient initialization, improve tracing with OpenTelemetry status handling, and simplify OpenAI base URL logic

Author: ankumar

examples/openai_client.py

@@ -12,19 +12,19 @@
 javelin_api_key = os.getenv('JAVELIN_API_KEY')
 gemini_api_key = os.getenv("GEMINI_API_KEY")
 
-# Initialize Javelin Client
+# Global JavelinClient, used for everything
 config = JavelinConfig(
     base_url="https://api-dev.javelin.live",
     # base_url="http://localhost:8000",
     javelin_api_key=javelin_api_key,
 )
-client = JavelinClient(config)
+client = JavelinClient(config) # Global JavelinClient
 
 # Initialize Javelin Client
 def initialize_javelin_client():
-    javelin_api_key = os.getenv('JAVELIN_API_KEY')
     config = JavelinConfig(
         base_url="https://api-dev.javelin.live",
+        # base_url="http://localhost:8000",
         javelin_api_key=javelin_api_key,
     )
     return JavelinClient(config)
@@ -288,8 +288,13 @@ def main_sync():
     openai_completions()
     openai_embeddings()
     openai_streaming_chat()
+    
+    openai_client = create_azureopenai_client() # same global client
+    register_azureopenai(client, openai_client)
+    azure_openai_chat_completions(openai_client)
 
-    client = initialize_javelin_client()
+    # Pending: model specs, uncomment after model is available
+    '''
     openai_client = create_gemini_client()
     register_gemini(client, openai_client)
     gemini_chat_completions(openai_client)
@@ -298,20 +303,15 @@ def main_sync():
     gemini_image_understanding(openai_client)
     gemini_structured_output(openai_client)
     gemini_embeddings(openai_client)
-    
-    client = initialize_javelin_client()
-    openai_client = create_azureopenai_client()
-    register_azureopenai(client, openai_client)
-    azure_openai_chat_completions(openai_client)
 
-    client = initialize_javelin_client()
     openai_client = create_deepseek_client()
     register_deepseek(client, openai_client)
     deepseek_chat_completions(openai_client)
 
     # deepseek_reasoning_model()
 
     mistral_chat_completions()
+    '''
 
 async def main_async():
     await async_openai_chat_completions()

javelin_sdk/client.py

@@ -17,8 +17,8 @@
 from javelin_sdk.tracing_setup import configure_span_exporter
 import inspect
 from opentelemetry.trace import SpanKind
+from opentelemetry.trace import Status, StatusCode
 from opentelemetry.semconv._incubating.attributes import gen_ai_attributes
-import openai
 
 API_BASEURL = "https://api-dev.javelin.live"
 API_BASE_PATH = "/v1"
@@ -158,13 +158,13 @@ def register_provider(self,
 
         client_id = id(openai_client)
         if client_id in self.patched_clients:
+            print (f"Client {client_id} already patched")
             return openai_client  # Skip if already patched
 
         self.patched_clients.add(client_id)  # Mark as patched
 
         # Store the OpenAI base URL
-        if self.openai_base_url is None:
-            self.openai_base_url = openai_client.base_url
+        self.openai_base_url = openai_client.base_url
 
         # Point the OpenAI client to Javelin's base URL
         openai_client.base_url = f"{self.base_url}/{provider_name}"
@@ -244,8 +244,8 @@ def _sync_execution(span):
                         else:
                             return _sync_execution(span)
                     except Exception as e:
-                        span.set_attribute("error", True)
-                        span.set_attribute("error.message", str(e))
+                        span.set_status(Status(StatusCode.ERROR, str(e)))
+                        span.set_attribute("is_exception", True)
                         raise
             else:
                 # Tracing is disabled
@@ -259,6 +259,15 @@ def _capture_response_details(span, response, kwargs, system_name):
             if hasattr(response, "to_json"):
                 response_data = response.to_dict()
 
+                # Set status code based on response
+                status_code = response_data.get("status_code", 200)
+                status_message = response_data.get("status_message", "OK")
+
+                if status_code >= 400:
+                    span.set_status(Status(StatusCode.ERROR, status_message))
+                else:
+                    span.set_status(Status(StatusCode.OK, status_message))
+
                 # Set basic response attributes
                 JavelinClient.set_span_attribute_if_not_none(span, gen_ai_attributes.GEN_AI_RESPONSE_MODEL, response_data.get('model'))
                 JavelinClient.set_span_attribute_if_not_none(span, gen_ai_attributes.GEN_AI_RESPONSE_ID, response_data.get('id'))