fix: aclose all async generators to fix OTel tracing context

See https://github.com/google/adk-python/issues/1670#issuecomment-3115891100

PiperOrigin-RevId: 794659547

src/google/adk/evaluation/agent_evaluator.py

@@ -32,6 +32,7 @@ from pydantic import BaseModel
 from pydantic import ValidationError
 
 from ..agents.base_agent import BaseAgent
+from ..utils.context_utils import Aclosing
 from .constants import MISSING_EVAL_DEPENDENCIES_MESSAGE
 from .eval_case import IntermediateData
 from .eval_case import Invocation
@@ -538,10 +539,11 @@ class AgentEvaluator:
     # Generate inferences
     inference_results = []
     for inference_request in inference_requests:
-      async for inference_result in eval_service.perform_inference(
-          inference_request=inference_request
-      ):
-        inference_results.append(inference_result)
+      async with Aclosing(
+          eval_service.perform_inference(inference_request=inference_request)
+      ) as agen:
+        async for inference_result in agen:
+          inference_results.append(inference_result)
 
     # Evaluate metrics
     # As we perform more than one run for an eval case, we collect eval results
@@ -551,14 +553,15 @@ class AgentEvaluator:
         inference_results=inference_results,
         evaluate_config=EvaluateConfig(eval_metrics=eval_metrics),
     )
-    async for eval_result in eval_service.evaluate(
-        evaluate_request=evaluate_request
-    ):
-      eval_id = eval_result.eval_id
-      if eval_id not in eval_results_by_eval_id:
-        eval_results_by_eval_id[eval_id] = []
+    async with Aclosing(
+        eval_service.evaluate(evaluate_request=evaluate_request)
+    ) as agen:
+      async for eval_result in agen:
+        eval_id = eval_result.eval_id
+        if eval_id not in eval_results_by_eval_id:
+          eval_results_by_eval_id[eval_id] = []
 
-      eval_results_by_eval_id[eval_id].append(eval_result)
+        eval_results_by_eval_id[eval_id].append(eval_result)
 
     return eval_results_by_eval_id
 

src/google/adk/evaluation/evaluation_generator.py

@@ -30,6 +30,7 @@ from ..runners import Runner
 from ..sessions.base_session_service import BaseSessionService
 from ..sessions.in_memory_session_service import InMemorySessionService
 from ..sessions.session import Session
+from ..utils.context_utils import Aclosing
 from .eval_case import EvalCase
 from .eval_case import IntermediateData
 from .eval_case import Invocation
@@ -189,18 +190,25 @@ class EvaluationGenerator:
       tool_uses = []
       invocation_id = ""
 
-      async for event in runner.run_async(
-          user_id=user_id, session_id=session_id, new_message=user_content
-      ):
-        invocation_id = (
-            event.invocation_id if not invocation_id else invocation_id
-        )
+      async with Aclosing(
+          runner.run_async(
+              user_id=user_id, session_id=session_id, new_message=user_content
+          )
+      ) as agen:
+        async for event in agen:
+          invocation_id = (
+              event.invocation_id if not invocation_id else invocation_id
+          )
 
-        if event.is_final_response() and event.content and event.content.parts:
-          final_response = event.content
-        elif event.get_function_calls():
-          for call in event.get_function_calls():
-            tool_uses.append(call)
+          if (
+              event.is_final_response()
+              and event.content
+              and event.content.parts
+          ):
+            final_response = event.content
+          elif event.get_function_calls():
+            for call in event.get_function_calls():
+              tool_uses.append(call)
 
       response_invocations.append(
           Invocation(

src/google/adk/evaluation/llm_as_judge.py

@@ -24,6 +24,7 @@ from ..models.base_llm import BaseLlm
 from ..models.llm_request import LlmRequest
 from ..models.llm_response import LlmResponse
 from ..models.registry import LLMRegistry
+from ..utils.context_utils import Aclosing
 from .eval_case import Invocation
 from .eval_metrics import EvalMetric
 from .evaluator import EvaluationResult
@@ -109,21 +110,22 @@ class LlmAsJudge(Evaluator):
       num_samples = self._judge_model_options.num_samples
       invocation_result_samples = []
       for _ in range(num_samples):
-        async for llm_response in self._judge_model.generate_content_async(
-            llm_request
-        ):
-          # Non-streaming call, so there is only one response content.
-          score = self.convert_auto_rater_response_to_score(llm_response)
-          invocation_result_samples.append(
-              PerInvocationResult(
-                  actual_invocation=actual,
-                  expected_invocation=expected,
-                  score=score,
-                  eval_status=get_eval_status(
-                      score, self._eval_metric.threshold
-                  ),
-              )
-          )
+        async with Aclosing(
+            self._judge_model.generate_content_async(llm_request)
+        ) as agen:
+          async for llm_response in agen:
+            # Non-streaming call, so there is only one response content.
+            score = self.convert_auto_rater_response_to_score(llm_response)
+            invocation_result_samples.append(
+                PerInvocationResult(
+                    actual_invocation=actual,
+                    expected_invocation=expected,
+                    score=score,
+                    eval_status=get_eval_status(
+                        score, self._eval_metric.threshold
+                    ),
+                )
+            )
       if not invocation_result_samples:
         continue
       per_invocation_results.append(