Add function calling telemetry

Author: johnoliver

aiservices/openai/src/main/java/com/microsoft/semantickernel/aiservices/openai/chatcompletion/OpenAIChatCompletion.java

@@ -50,6 +50,7 @@
 import com.microsoft.semantickernel.hooks.PreChatCompletionEvent;
 import com.microsoft.semantickernel.hooks.PreToolCallEvent;
 import com.microsoft.semantickernel.implementation.CollectionUtil;
+import com.microsoft.semantickernel.implementation.telemetry.ChatCompletionSpan;
 import com.microsoft.semantickernel.implementation.telemetry.SemanticKernelTelemetry;
 import com.microsoft.semantickernel.orchestration.FunctionResult;
 import com.microsoft.semantickernel.orchestration.FunctionResultMetadata;
@@ -69,7 +70,6 @@
 import com.microsoft.semantickernel.services.chatcompletion.message.ChatMessageContentType;
 import com.microsoft.semantickernel.services.chatcompletion.message.ChatMessageImageContent;
 import com.microsoft.semantickernel.services.openai.OpenAiServiceBuilder;
-import io.opentelemetry.api.trace.Span;
 import java.io.IOException;
 import java.util.ArrayList;
 import java.util.Arrays;
@@ -419,26 +419,32 @@ private Mono<ChatMessages> internalChatMessageContentsAsync(
                     invocationContext)))
             .getOptions();
 
-        Span span = SemanticKernelTelemetry.startChatCompletionSpan(
-            getModelId(),
-            SemanticKernelTelemetry.OPEN_AI_PROVIDER,
-            options.getMaxTokens(),
-            options.getTemperature(),
-            options.getTopP());
-        return getClient()
-            .getChatCompletionsWithResponse(getDeploymentName(), options,
-                OpenAIRequestSettings.getRequestOptions())
-            .flatMap(completionsResult -> {
-                if (completionsResult.getStatusCode() >= 400) {
-                    SemanticKernelTelemetry.endSpanWithError(span);
-                    return Mono.error(new AIException(ErrorCodes.SERVICE_ERROR,
-                        "Request failed: " + completionsResult.getStatusCode()));
-                }
-                SemanticKernelTelemetry.endSpanWithUsage(span,
-                    completionsResult.getValue().getUsage());
+        return Mono.deferContextual(contextView -> {
+            ChatCompletionSpan span = ChatCompletionSpan.startChatCompletionSpan(
+                SemanticKernelTelemetry.getTelemetry(invocationContext),
+                contextView,
+                getModelId(),
+                SemanticKernelTelemetry.OPEN_AI_PROVIDER,
+                options.getMaxTokens(),
+                options.getTemperature(),
+                options.getTopP());
+
+            return getClient()
+                .getChatCompletionsWithResponse(getDeploymentName(), options,
+                    OpenAIRequestSettings.getRequestOptions())
+                .contextWrite(span.getReactorContextModifier())
+                .flatMap(completionsResult -> {
+                    if (completionsResult.getStatusCode() >= 400) {
+                        return Mono.error(new AIException(ErrorCodes.SERVICE_ERROR,
+                            "Request failed: " + completionsResult.getStatusCode()));
+                    }
 
-                return Mono.just(completionsResult.getValue());
-            })
+                    return Mono.just(completionsResult.getValue());
+                })
+                .doOnError(span::endSpanWithError)
+                .doOnSuccess(span::endSpanWithUsage)
+                .doOnTerminate(span::close);
+        })
             .flatMap(completions -> {
 
                 List<ChatResponseMessage> responseMessages = completions

aiservices/openai/src/main/java/com/microsoft/semantickernel/aiservices/openai/textcompletion/OpenAITextGenerationService.java

@@ -14,8 +14,6 @@
 import com.microsoft.semantickernel.services.StreamingTextContent;
 import com.microsoft.semantickernel.services.textcompletion.TextContent;
 import com.microsoft.semantickernel.services.textcompletion.TextGenerationService;
-import com.microsoft.semantickernel.implementation.telemetry.SemanticKernelTelemetry;
-import io.opentelemetry.api.trace.Span;
 import java.util.Collections;
 import java.util.HashMap;
 import java.util.List;
@@ -84,23 +82,14 @@ protected Mono<List<TextContent>> internalCompleteTextAsync(
 
         CompletionsOptions completionsOptions = getCompletionsOptions(text, requestSettings);
 
-        Span span = SemanticKernelTelemetry.startTextCompletionSpan(
-            getModelId(),
-            SemanticKernelTelemetry.OPEN_AI_PROVIDER,
-            completionsOptions.getMaxTokens(),
-            completionsOptions.getTemperature(),
-            completionsOptions.getTopP());
         return getClient()
             .getCompletionsWithResponse(getDeploymentName(), completionsOptions,
                 OpenAIRequestSettings.getRequestOptions())
             .flatMap(completionsResult -> {
                 if (completionsResult.getStatusCode() >= 400) {
-                    SemanticKernelTelemetry.endSpanWithError(span);
                     return Mono.error(new AIException(ErrorCodes.SERVICE_ERROR,
                         "Request failed: " + completionsResult.getStatusCode()));
                 }
-                SemanticKernelTelemetry.endSpanWithUsage(span,
-                    completionsResult.getValue().getUsage());
                 return Mono.just(completionsResult.getValue());
             })
             .map(completions -> {

aiservices/openai/src/test/java/com/microsoft/semantickernel/aiservices/openai/OtelCaptureTest.java

@@ -75,57 +75,6 @@ public static void shutdown() {
         otel.shutdown();
     }
 
-    @Test
-    public void otelTextCaptureTest() {
-
-        OpenAIAsyncClient openAIAsyncClient = Mockito.mock(OpenAIAsyncClient.class);
-
-        CompletionsUsage completionsUsage = Mockito.mock(CompletionsUsage.class);
-        Mockito.when(completionsUsage.getCompletionTokens()).thenReturn(22);
-        Mockito.when(completionsUsage.getPromptTokens()).thenReturn(55);
-
-        Completions completions = Mockito.mock(Completions.class);
-        Mockito.when(completions.getUsage()).thenReturn(completionsUsage);
-
-        Response<Completions> response = Mockito.mock(Response.class);
-        Mockito.when(response.getStatusCode()).thenReturn(200);
-        Mockito.when(response.getValue()).thenReturn(completions);
-
-        Mockito.when(openAIAsyncClient.getCompletionsWithResponse(
-            Mockito.any(),
-            Mockito.<CompletionsOptions>any(),
-            Mockito.any())).thenAnswer(invocation -> Mono.just(response));
-
-        TextGenerationService client = OpenAITextGenerationService.builder()
-            .withOpenAIAsyncClient(openAIAsyncClient)
-            .withModelId("a-model")
-            .build();
-
-        try {
-            client.getTextContentsAsync(
-                "foo",
-                null,
-                null).block();
-        } catch (Exception e) {
-            // Expect to fail
-        }
-
-        Assertions.assertFalse(spans.isEmpty());
-        Assertions.assertEquals("a-model",
-            spans.get(0).getAttributes().get(AttributeKey.stringKey("gen_ai.request.model")));
-        Assertions.assertEquals("text.completions",
-            spans.get(0).getAttributes().get(AttributeKey.stringKey("gen_ai.operation.name")));
-        Assertions.assertEquals("openai",
-            spans.get(0).getAttributes().get(AttributeKey.stringKey("gen_ai.system")));
-        Assertions.assertEquals(22,
-            spans.get(0).getAttributes()
-                .get(AttributeKey.longKey("gen_ai.response.completion_tokens")));
-        Assertions.assertEquals(55,
-            spans.get(0).getAttributes()
-                .get(AttributeKey.longKey("gen_ai.response.prompt_tokens")));
-
-    }
-
     @Test
     public void otelChatCaptureTest() {
         OpenAIAsyncClient openAIAsyncClient = Mockito.mock(OpenAIAsyncClient.class);

samples/semantickernel-concepts/semantickernel-syntax-examples/pom.xml

@@ -26,6 +26,11 @@
     </dependencyManagement>
 
     <dependencies>
+        <dependency>
+            <groupId>io.opentelemetry.instrumentation</groupId>
+            <artifactId>opentelemetry-reactor-3.1</artifactId>
+            <version>2.9.0-alpha</version>
+        </dependency>
         <dependency>
             <groupId>com.microsoft.semantic-kernel</groupId>
             <artifactId>semantickernel-api</artifactId>
@@ -165,6 +170,7 @@
                 </executions>
                 <configuration>
                     <mainClass>com.microsoft.semantickernel.samples.syntaxexamples.${sample}</mainClass>
+                    <cleanupDaemonThreads>false</cleanupDaemonThreads>
                 </configuration>
             </plugin>
         </plugins>