service_level = "priority" does not work with VertexAI #7263

Activity

1. llalitkumarrr

   self-assigned this

   on Sep 25, 2026
2. sanketpatil06

   added the

   Bug

   issue type on Sep 25, 2026
3. sanketpatil06

   added

   services[Component] This issue is related to runtime services, e.g. sessions, memory, artifacts, etc[Component] This issue is related to runtime services, e.g. sessions, memory, artifacts, etc

   on Sep 25, 2026
4. llalitkumarrr commented on Sep 28, 2026

   

   llalitkumarrr

   on Sep 28, 2026

   Collaborator

   More actions

   Hello @SaeeSaadat,

   We tried to reproduce this issue using version 2.9.2 but we did not get any errors. Could you please try this version and share your observations? For your reference the code we used for the reproduction is provided below.

   from google.adk import Agent
   from google.adk.models.google_llm import Gemini
   from google.adk import Runner
   from google.adk.sessions import InMemorySessionService
   import asyncio
   from google.genai import types
   from dotenv import load_dotenv
   load_dotenv(override=True)
   
   
   pim_agent = Agent(
       model=Gemini(
           model="gemini-3.8-flash"
       ),
       name="root_agent",
       description="Help full assistant",
       instruction="You are a helpful assistant that answers user queries.",
       generate_content_config=types.GenerateContentConfig(
           temperature=0.5,
           max_output_tokens=8192,
           safety_settings=[
               types.SafetySetting(
                category=types.HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT,
                threshold=types.HarmBlockThreshold.BLOCK_ONLY_HIGH,
               )
           ],
       ),
   )
   
   async def main():
       # Create a new session
       session_service = InMemorySessionService()
       session = await session_service.create_session(
           app_name="test-app",
           user_id="test-user",
           session_id="test-session",
   
       )
       print(f"Created session: {session.id}")
   
       runner = Runner(agent=pim_agent, session_service=session_service,app_name="test-app")
   
       new_message = types.Content(role="user", parts=[types.Part(text="What is 2 + 2?")])
       async for event in runner.run_async(
           new_message=new_message,
           user_id="test-user",
           session_id="test-session",
       ):
           print(f"\n\n\n\n{'='*100}\nresponse: {event.content.parts[0].text}\n\n\n\n{'='*100}")
   
   if __name__ == "__main__":
       asyncio.run(main())
   
   

   Response:

   ====================================================================================================
   response: 2 + 2 = 4.
   
   
   
   ====================================================================================================
   
5. llalitkumarrr

   added

   core[Component] This issue is related to the core interface and implementation[Component] This issue is related to the core interface and implementation

   request clarification[Status] The maintainer need clarification or more information from the author[Status] The maintainer need clarification or more information from the author

   and removed

   services[Component] This issue is related to runtime services, e.g. sessions, memory, artifacts, etc[Component] This issue is related to runtime services, e.g. sessions, memory, artifacts, etc

   on Sep 28, 2026
6. SaeeSaadat commented on Sep 28, 2026

   

   SaeeSaadat

   on Sep 28, 2026

   Author

   More actions

   @llalitkumarrr Hi. did you use VertexAI authentication? Or API KEY? It works fine with API KEY but not with VertexAI.
7. llalitkumarrr commented on Sep 28, 2026

   

   llalitkumarrr

   on Sep 28, 2026

   Collaborator

   More actions

   Hi @SaeeSaadat, We used VertexAI for this reproduction.
8. GWeale

   self-assigned this

   on Oct 26, 2026
9. GWeale commented on Oct 26, 2026

   

   GWeale

   on Oct 26, 2026

   Collaborator

   More actions

   hi @SaeeSaadat, thanks for sticking with this, and sorry about the earlier confusion: our first repro didn't set service_tier, which is why we missed it. vertex reads the tier from request headers and rejects the body value that google-genai sends. since adk just forwards service_tier straight to google-genai, the fix needs to happen over in googleapis/python-genai#2847. in the meantime, you can work around it by setting X-Vertex-AI-LLM-Request-Type: shared and X-Vertex-AI-LLM-Shared-Request-Type: priority in generate_content_config.http_options.headers. i'll close this issue so we can keep the tracking in one place.
10. GWeale

    closed this as not plannedon Oct 26, 2026