VertexAiSessionService fails with ValidationError when using EventsCompactionConfig and ContextCacheConfig in ADK App

[bgautham-dev]

Description

I am trying to deploy an ADK agent in Agent Engine using VertexAiSessionService.
The setup was working correctly before introducing EventsCompactionConfig and ContextCacheConfig.

To enable conversation history compression and caching, I wrapped my agent using google.adk.apps.App and provided the compaction and cache configurations. After doing this, I started encountering a pydantic ValidationError during invocation.

My main goal is to control conversation history growth by compressing or filtering past events (to limit token usage).

---

Code Snippet

from vertexai import agent_engines
from google.adk.apps.app import EventsCompactionConfig
from google.adk.agents.context_cache_config import ContextCacheConfig
from google.adk.sessions import VertexAiSessionService
from agents.root_agent.agent import root_agent
from google.adk.apps import App

def session_service_builder(**kwargs):
    return VertexAiSessionService(location="asia-south1")

google_adk_app = App(
    root_agent=root_agent,
    name="root_agent",
    events_compaction_config=EventsCompactionConfig(
        compaction_interval=3,  # Trigger compaction every 3 new invocations
        overlap_size=1
    ),
    context_cache_config=ContextCacheConfig(
        min_tokens=2048,    # Minimum tokens to trigger caching
        ttl_seconds=600,    # Store for up to 10 minutes
        cache_intervals=5,  # Refresh after 5 uses
    )
)

app = agent_engines.AdkApp(
    agent=google_adk_app,
    enable_tracing=True,
    session_service_builder=session_service_builder
)

---

Error

Traceback (most recent call last):
  File "/code/app/api/factory/python_file_api_builder.py", line 473, in _async_stream_json_generator
    async for chunk in output:
  File "/home/appuser/.cache/pypoetry/virtualenvs/reasoning-engine-assembly-service-MATOk_fk-py3.12/lib/python3.12/site-packages/vertexai/agent_engines/templates/adk.py", line 1058, in async_stream_query
    async for event in events_async:
  File "/home/appuser/.cache/pypoetry/virtualenvs/reasoning-engine-assembly-service-MATOk_fk-py3.12/lib/python3.12/site-packages/google/adk/runners.py", line 561, in run_async
    async for event in agen:
  File "/home/appuser/.cache/pypoetry/virtualenvs/reasoning-engine-assembly-service-MATOk_fk-py3.12/lib/python3.12/site-packages/google/adk/runners.py", line 529, in _run_with_trace
    invocation_context = await self._setup_context_for_new_invocation(
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/appuser/.cache/pypoetry/virtualenvs/reasoning-engine-assembly-service-MATOk_fk-py3.12/lib/python3.12/site-packages/google/adk/runners.py", line 1275, in _setup_context_for_new_invocation
    invocation_context = self._new_invocation_context(
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/appuser/.cache/pypoetry/virtualenvs/reasoning-engine-assembly-service-MATOk_fk-py3.12/lib/python3.12/site-packages/google/adk/runners.py", line 1405, in _new_invocation_context
    return InvocationContext(
           ^^^^^^^^^^^^^^^^^^
  File "/home/appuser/.cache/pypoetry/virtualenvs/reasoning-engine-assembly-service-MATOk_fk-py3.12/lib/python3.12/site-packages/pydantic/main.py", line 253, in __init__
    validated_self = self.__pydantic_validator__.validate_python(data, self_instance=self)
                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
pydantic_core._pydantic_core.ValidationError: 1 validation error for InvocationContext

Full traceback indicates failure during _new_invocation_context inside google.adk.runners.

---

Questions

1. Does VertexAiSessionService currently support EventsCompactionConfig and ContextCacheConfig?

2. Is there any incompatibility between google.adk.apps.App wrapping and agent_engines.AdkApp?

3. If context caching/compaction is not supported with VertexAiSessionService, what is the recommended way to:

   • Compress conversation history?
   • Limit token growth?
   • Filter past events before sending to the model?

---

Expected Behavior

The agent should support:

• Automatic event compaction
• Context caching
• Controlled conversation history growth

while using VertexAiSessionService inside Agent Engine.

---