[Agent Engine][Adk] app_name mismatch between AdkApp._tmpl_attrs and Runner causes GCS artifact lookup failures in Agent Engine
Nobody has claimed this yet.
Assessment
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Newbie friendliness
- 55/100
Research direction
Start in vertexai.agent_engines.templates.adk and inspect AdkApp.set_up(), then compare its template app_name with Runner._validate_runner_params and _convert_response_events. Reproduce the artifact save/load path using the described Agent Engine environment and confirm that the wrapped ADK App name is used consistently without relying on the numerical GOOGLE_CLOUD_AGENT_ENGINE_ID.
Written by the indexing model from the issue text.
Description
Environment details
- OS type and version: Linux (Vertex AI Agent Engine environment)
- Python version:
Python 3.11+ google-cloud-aiplatformversion:1.143.0
Steps to reproduce
- Create an agent via Google ADK exposing tools that generate or return visual artifacts (images/videos) using
tool_context.save_artifact(). - Instantiate and wrap the ADK agent into a Vertex AI SDK template using
AdkApp(fromvertexai.agent_engines.templates.adk). - Deploy this agent to Agent Engine (Vertex AI platform container mapping to Gemini Enterprise).
- Trigger any tool that saved an artifact from the UI chat.
Root Cause Analysis
The issue occurs because the application name used by the container falls back onto the ID assigned by the Cloud platform if not explicitly provided, while the Runner defaults to the name defined in the ADK App instance.
In AdkApp.set_up(), the fallback logic is as follows:
# AdkApp.set_up() fallback logic
if not self._tmpl_attrs.get("app_name"):
if "GOOGLE_CLOUD_AGENT_ENGINE_ID" in os.environ:
self._tmpl_attrs["app_name"] = os.environ.get(
"GOOGLE_CLOUD_AGENT_ENGINE_ID",
)
else:
self._tmpl_attrs["app_name"] = _DEFAULT_APP_NAME
However, the Runner created later defaults to app.name (the static default "app" in ADK if not overridden):
# Runner._validate_runner_params defaults to app.name
app_name = app_name or app.name # resolves to "app" instead of the ID
This creates a mismatch failure during session artifact reads in _convert_response_events.
Actual Behavior
-
Save Path evaluates using
app.name(e.g.,"app"or a custom name like"mova"):
GCS_BUCKET/app/{user_id}/{session_id}/artifact.png -
Load Path evaluates using
AdkApp._tmpl_attrs["app_name"](e.g., the numerical ID assigned by Google Cloud):
GCS_BUCKET/8269664996770709504/{user_id}/{session_id}/artifact.png
Because strings like "app" and "8269664996770709504" on bucket paths do not map identically, files aren't found on state loads, and the Gemini Enterprise UI draws empty boxes or fails to render the media.
Moreover, if the user tries to use GOOGLE_CLOUD_AGENT_ENGINE_ID directly to solve the mismatch, the purely numerical ID causes Pydantic validation failures on boot because it is not a valid Python identifier (starts with a digit).
Expected Behavior
The AdkApp should ensure that the app_name in its template attributes matches the name of the ADK App being wrapped, or provide a clean way to synchronize them, rather than falling back to the numerical GOOGLE_CLOUD_AGENT_ENGINE_ID.
Workaround Applied
To fix this manually in user code, one can override _tmpl_attrs["app_name"] after calling super().set_up() in the custom AdkApp class:
class AgentEngineApp(AdkApp):
def set_up(self) -> None:
super().set_up()
# Enforce that AdkApp uses the same name as the ADK App
self._tmpl_attrs["app_name"] = adk_app.name
- Dominant language
- Python
- Stars
- 907
- Forks
- 467
- Avg merge
- 1d 8h
- Merged PRs (30d)
- 40
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from googleapis/python-aiplatform
-
api: vertex-ai
Difficulty 1/5 Under an hour Newbie friendliness 85/100
googleapis/python-aiplatform#7132 ·
-
api: vertex-ai
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
googleapis/python-aiplatform#7097 ·
-
CustomContainerTrainingJob.run drops max_wait_duration=0 instead of requesting indefinite DWS wait Openapi: vertex-ai
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
googleapis/python-aiplatform#7067 · 1 comment ·
-
api: vertex-ai
Difficulty 2/5 1-3 hours Newbie friendliness 68/100
googleapis/python-aiplatform#6877 ·
-
api: vertex-ai
Difficulty 2/5 1-3 hours Newbie friendliness 68/100
googleapis/python-aiplatform#6865 · 1 comment ·
All issues in googleapis/python-aiplatform
Similar issues
-
bug confirmed issue
Difficulty 2/5 1-3 hours Newbie friendliness 75/100
open-webui/open-webui#30750 · 1 comment ·
-
Difficulty 2/5 1-3 hours Newbie friendliness 75/100
-
enhancement
Difficulty 2/5 1-3 hours Newbie friendliness 75/100
OpenwaterHealth/openmotion-bloodflow-app#604 · 1 comment ·
-
Difficulty 2/5 1-3 hours Newbie friendliness 70/100
-
good first issue
Difficulty 1/5 Under an hour Newbie friendliness 90/100