feat(telemetry): unify span creation paths for hierarchical trace tree (#3731 P3 Phase 1)

Replace disconnected withSpan/startSpanWithContext calls in runtime with
session-tracing typed helpers so LLM and tool spans become children of
the interaction span instead of siblings under the session root.

- Add toolContext ALS with runInToolSpanContext() for concurrent-safe
  tool span scoping (uses AsyncLocalStorage.run, not enterWith)
- Wire startLLMRequestSpan/endLLMRequestSpan in loggingContentGenerator
  for both streaming and non-streaming paths
- Wire startToolSpan/endToolSpan + startToolExecutionSpan/endToolExecutionSpan
  in coreToolScheduler with proper try/finally lifecycle
- Remove redundant withSpan('client.generateContent') wrapper from client.ts
- Fix endToolSpan to not override pre-set status when metadata is omitted
- Change startToolExecutionSpan to read parent from toolContext ALS
- Update tests for new span creation APIs and remove dead test infrastructure

Author: doudouOUC

docs/design/workflow-tracing-gaps.md

@@ -15,7 +15,6 @@ import {
 } from 'vitest';
 
 import type { Content, GenerateContentResponse, Part } from '@google/genai';
-import { SpanStatusCode } from '@opentelemetry/api';
 import { GeminiClient, SendMessageType } from './client.js';
 import { findCompressSplitPoint } from '../services/chatCompressionService.js';
 import {
@@ -161,29 +160,9 @@ const mockUiTelemetryService = vi.hoisted(() => ({
   reset: vi.fn(),
   addEvent: vi.fn(),
 }));
-const clientSpanCalls = vi.hoisted(
-  (): Array<{
-    name: string;
-    attributes: Record<string, string | number | boolean>;
-    statuses: Array<{ code: number; message?: string }>;
-  }> => [],
-);
-const mockWithSpan = vi.hoisted(() => vi.fn());
-
 vi.mock('../telemetry/tracer.js', () => ({
   API_CALL_ABORTED_SPAN_STATUS_MESSAGE: 'API call aborted',
   API_CALL_FAILED_SPAN_STATUS_MESSAGE: 'API call failed',
-  safeSetStatus: (
-    span: { setStatus: (status: { code: number; message?: string }) => void },
-    status: { code: number; message?: string },
-  ) => {
-    try {
-      span.setStatus(status);
-    } catch {
-      // Match production best-effort telemetry behavior.
-    }
-  },
-  withSpan: mockWithSpan,
 }));
 
 vi.mock('../telemetry/index.js', async (importOriginal) => {
@@ -329,32 +308,6 @@ describe('Gemini Client (client.ts)', () => {
   };
   beforeEach(async () => {
     vi.resetAllMocks();
-    clientSpanCalls.length = 0;
-    mockWithSpan.mockImplementation(
-      async (
-        name: string,
-        attributes: Record<string, string | number | boolean>,
-        fn: (span: {
-          setStatus: ReturnType<typeof vi.fn>;
-          setAttribute: ReturnType<typeof vi.fn>;
-          end: ReturnType<typeof vi.fn>;
-        }) => Promise<unknown>,
-      ) => {
-        const spanCall = {
-          name,
-          attributes,
-          statuses: [] as Array<{ code: number; message?: string }>,
-        };
-        clientSpanCalls.push(spanCall);
-        return fn({
-          setStatus: vi.fn((status: { code: number; message?: string }) => {
-            spanCall.statuses.push(status);
-          }),
-          setAttribute: vi.fn(),
-          end: vi.fn(),
-        });
-      },
-    );
     vi.mocked(uiTelemetryService.setLastPromptTokenCount).mockClear();
 
     // Default: createContentGenerator rejects (simulates test env without auth).
@@ -3250,15 +3203,6 @@ Other open files:
         }),
         'btw-prompt-id',
       );
-      expect(clientSpanCalls.at(-1)).toEqual(
-        expect.objectContaining({
-          name: 'client.generateContent',
-          attributes: {
-            model: DEFAULT_QWEN_FLASH_MODEL,
-            prompt_id: 'btw-prompt-id',
-          },
-        }),
-      );
     });
 
     it('should prefer an explicit prompt id override over the current context', async () => {
@@ -3286,15 +3230,6 @@ Other open files:
         }),
         'override-prompt-id',
       );
-      expect(clientSpanCalls.at(-1)).toEqual(
-        expect.objectContaining({
-          name: 'client.generateContent',
-          attributes: {
-            model: DEFAULT_QWEN_FLASH_MODEL,
-            prompt_id: 'override-prompt-id',
-          },
-        }),
-      );
     });
 
     it('should use config system prompt override when provided', async () => {
@@ -3398,7 +3333,7 @@ Other open files:
       );
     });
 
-    it('sets a generic span status when content generation fails', async () => {
+    it('propagates error when content generation fails', async () => {
       const contents = [{ role: 'user', parts: [{ text: 'hello' }] }];
       const abortSignal = new AbortController().signal;
       mockGenerateContentFn.mockRejectedValueOnce(
@@ -3413,15 +3348,9 @@ Other open files:
           DEFAULT_QWEN_FLASH_MODEL,
         ),
       ).rejects.toThrow('raw upstream 500 with sensitive details');
-
-      const spanCall = clientSpanCalls.at(-1);
-      expect(spanCall?.statuses).toEqual([
-        { code: SpanStatusCode.ERROR, message: 'API call failed' },
-      ]);
-      expect(JSON.stringify(spanCall?.statuses)).not.toContain('raw upstream');
     });
 
-    it('sets a generic aborted span status when content generation is aborted', async () => {
+    it('propagates error when content generation is aborted', async () => {
       const contents = [{ role: 'user', parts: [{ text: 'hello' }] }];
       const abortController = new AbortController();
       abortController.abort();
@@ -3437,12 +3366,6 @@ Other open files:
           DEFAULT_QWEN_FLASH_MODEL,
         ),
       ).rejects.toThrow('raw abort reason with sensitive details');
-
-      const spanCall = clientSpanCalls.at(-1);
-      expect(spanCall?.statuses).toEqual([
-        { code: SpanStatusCode.ERROR, message: 'API call aborted' },
-      ]);
-      expect(JSON.stringify(spanCall?.statuses)).not.toContain('raw abort');
     });
 
     // Note: there is currently no "fallback mode" model routing; the model used

packages/core/src/core/client.test.ts

@@ -12,7 +12,6 @@ import type {
   PartListUnion,
   Tool,
 } from '@google/genai';
-import { SpanStatusCode } from '@opentelemetry/api';
 
 // Config
 import { ApprovalMode, type Config } from '../config/config.js';
@@ -101,12 +100,6 @@ import { createHookOutput } from '../hooks/types.js';
 import { ideContextStore } from '../ide/ideContext.js';
 import { type File, type IdeContext } from '../ide/types.js';
 import type { StopHookOutput } from '../hooks/types.js';
-import {
-  API_CALL_ABORTED_SPAN_STATUS_MESSAGE,
-  API_CALL_FAILED_SPAN_STATUS_MESSAGE,
-  safeSetStatus,
-  withSpan,
-} from '../telemetry/tracer.js';
 
 const MAX_TURNS = 100;
 
@@ -1493,89 +1486,71 @@ export class GeminiClient {
     const promptId =
       promptIdOverride ?? promptIdContext.getStore() ?? this.lastPromptId!;
 
-    return withSpan(
-      'client.generateContent',
-      { model, prompt_id: promptId },
-      async (span) => {
-        let currentAttemptModel: string = model;
+    let currentAttemptModel: string = model;
 
-        try {
-          const userMemory = this.config.getUserMemory();
-          const finalSystemInstruction = generationConfig.systemInstruction
-            ? getCustomSystemPrompt(
-                generationConfig.systemInstruction,
-                userMemory,
-              )
-            : this.getMainSessionSystemInstruction();
-
-          const requestConfig: GenerateContentConfig = {
-            abortSignal,
-            ...generationConfig,
-            systemInstruction: finalSystemInstruction,
-          };
-
-          // When the requested model differs from the main model (e.g. fast model
-          // side queries for session recap / title / summary), resolve the target
-          // model's own ContentGeneratorConfig so that per-model settings like
-          // extra_body, samplingParams, and reasoning are not inherited from the
-          // main model's config. The retry authType is resolved alongside so that
-          // provider-specific checks (e.g. QWEN_OAUTH quota detection) reference
-          // the target model's provider.
-          const { contentGenerator, retryAuthType } = await this.config
-            .getBaseLlmClient()
-            .resolveForModel(model);
-
-          const apiCall = () => {
-            currentAttemptModel = model;
-
-            return contentGenerator.generateContent(
-              {
-                model,
-                config: requestConfig,
-                contents,
-              },
-              promptId,
-            );
-          };
-          const result = await retryWithBackoff(apiCall, {
-            authType: retryAuthType,
-            persistentMode: isUnattendedMode(),
-            signal: abortSignal,
-            heartbeatFn: (info) => {
-              process.stderr.write(
-                `[qwen-code] Waiting for API capacity... attempt ${info.attempt}, retry in ${Math.ceil(info.remainingMs / 1000)}s\n`,
-              );
-            },
-          });
-          return result;
-        } catch (error: unknown) {
-          if (abortSignal.aborted) {
-            safeSetStatus(span, {
-              code: SpanStatusCode.ERROR,
-              message: API_CALL_ABORTED_SPAN_STATUS_MESSAGE,
-            });
-            throw error;
-          }
+    try {
+      const userMemory = this.config.getUserMemory();
+      const finalSystemInstruction = generationConfig.systemInstruction
+        ? getCustomSystemPrompt(generationConfig.systemInstruction, userMemory)
+        : this.getMainSessionSystemInstruction();
+
+      const requestConfig: GenerateContentConfig = {
+        abortSignal,
+        ...generationConfig,
+        systemInstruction: finalSystemInstruction,
+      };
 
-          safeSetStatus(span, {
-            code: SpanStatusCode.ERROR,
-            message: API_CALL_FAILED_SPAN_STATUS_MESSAGE,
-          });
-          await reportError(
-            error,
-            `Error generating content via API with model ${currentAttemptModel}.`,
-            {
-              requestContents: contents,
-              requestConfig: generationConfig,
-            },
-            'generateContent-api',
-          );
-          throw new Error(
-            `Failed to generate content with model ${currentAttemptModel}: ${getErrorMessage(error)}`,
+      // When the requested model differs from the main model (e.g. fast model
+      // side queries for session recap / title / summary), resolve the target
+      // model's own ContentGeneratorConfig so that per-model settings like
+      // extra_body, samplingParams, and reasoning are not inherited from the
+      // main model's config. The retry authType is resolved alongside so that
+      // provider-specific checks (e.g. QWEN_OAUTH quota detection) reference
+      // the target model's provider.
+      const { contentGenerator, retryAuthType } = await this.config
+        .getBaseLlmClient()
+        .resolveForModel(model);
+
+      const apiCall = () => {
+        currentAttemptModel = model;
+
+        return contentGenerator.generateContent(
+          {
+            model,
+            config: requestConfig,
+            contents,
+          },
+          promptId,
+        );
+      };
+      const result = await retryWithBackoff(apiCall, {
+        authType: retryAuthType,
+        persistentMode: isUnattendedMode(),
+        signal: abortSignal,
+        heartbeatFn: (info) => {
+          process.stderr.write(
+            `[qwen-code] Waiting for API capacity... attempt ${info.attempt}, retry in ${Math.ceil(info.remainingMs / 1000)}s\n`,
           );
-        }
-      },
-    );
+        },
+      });
+      return result;
+    } catch (error: unknown) {
+      if (abortSignal.aborted) {
+        throw error;
+      }
+      await reportError(
+        error,
+        `Error generating content via API with model ${currentAttemptModel}.`,
+        {
+          requestContents: contents,
+          requestConfig: generationConfig,
+        },
+        'generateContent-api',
+      );
+      throw new Error(
+        `Failed to generate content with model ${currentAttemptModel}: ${getErrorMessage(error)}`,
+      );
+    }
   }
 
   /**