returned session id and used it in evals.

src/google/adk/cli/cli_eval.py

@@ -230,7 +230,7 @@ async def run_evals(
         print(f"Running Eval: {eval_set_id}:{eval_name}")
         session_id = f"{EVAL_SESSION_ID_PREFIX}{str(uuid.uuid4())}"
 
-        inference_result = (
+        inference_result, session_id = (
             await EvaluationGenerator._generate_inferences_from_root_agent(
                 invocations=eval_case.conversation,
                 root_agent=root_agent,

src/google/adk/evaluation/evaluation_generator.py

@@ -28,6 +28,10 @@
 from ..memory.in_memory_memory_service import InMemoryMemoryService
 from ..runners import Runner
 from ..sessions.base_session_service import BaseSessionService
+try: 
+  from ..sessions.vertex_ai_session_service import VertexAiSessionService
+except ImportError:
+  VertexAiSessionService = None
 from ..sessions.in_memory_session_service import InMemorySessionService
 from ..sessions.session import Session
 from ..utils.context_utils import Aclosing
@@ -132,9 +136,10 @@ async def _process_query(
       agent_to_evaluate = root_agent.find_agent(agent_name)
       assert agent_to_evaluate, f"Sub-Agent `{agent_name}` not found."
 
-    return await EvaluationGenerator._generate_inferences_from_root_agent(
+    response_invocations, _ = await EvaluationGenerator._generate_inferences_from_root_agent(
         invocations, agent_to_evaluate, reset_func, initial_session
     )
+    return response_invocations
 
   @staticmethod
   async def _generate_inferences_from_root_agent(
@@ -146,7 +151,7 @@ async def _generate_inferences_from_root_agent(
       session_service: Optional[BaseSessionService] = None,
       artifact_service: Optional[BaseArtifactService] = None,
       memory_service: Optional[BaseMemoryService] = None,
-  ) -> list[Invocation]:
+  ) -> tuple[list[Invocation], str]:
     """Scrapes the root agent given the list of Invocations."""
     if not session_service:
       session_service = InMemorySessionService()
@@ -158,14 +163,22 @@ async def _generate_inferences_from_root_agent(
         initial_session.app_name if initial_session else "EvaluationGenerator"
     )
     user_id = initial_session.user_id if initial_session else "test_user_id"
-    session_id = session_id if session_id else str(uuid.uuid4())
 
-    _ = await session_service.create_session(
+    if VertexAiSessionService and isinstance(session_service, VertexAiSessionService):
+      vertex_session = await session_service.create_session(
+          app_name=app_name,
+          user_id=user_id,
+          state=initial_session.state if initial_session else {}
+      )
+      session_id = vertex_session.id
+    else:
+      session_id = session_id if session_id else str(uuid.uuid4())
+      _ = await session_service.create_session(
         app_name=app_name,
         user_id=user_id,
         state=initial_session.state if initial_session else {},
         session_id=session_id,
-    )
+      )
 
     if not artifact_service:
       artifact_service = InMemoryArtifactService()
@@ -219,7 +232,7 @@ async def _generate_inferences_from_root_agent(
           )
       )
 
-    return response_invocations
+    return response_invocations, session_id
 
   @staticmethod
   def _process_query_with_session(session_data, data):

src/google/adk/evaluation/local_eval_service.py

@@ -359,7 +359,7 @@ async def _perform_inference_sigle_eval_item(
     )
 
     try:
-      inferences = (
+      inferences, session_id = (
           await EvaluationGenerator._generate_inferences_from_root_agent(
               invocations=eval_case.conversation,
               root_agent=root_agent,
@@ -371,6 +371,7 @@ async def _perform_inference_sigle_eval_item(
       )
 
       inference_result.inferences = inferences
+      inference_result.session_id = session_id  # Relevant for Vertex AI Session Service and other services that use ad-hoc session id.
       inference_result.status = InferenceStatus.SUCCESS
 
       return inference_result