`DatabaseSessionService.append_event` crashes with `PydanticSerializationError: Unable to serialize unknown type: <class 'function'>`

[varun-official]

Description

When using DatabaseSessionService (or any persistent session service), append_event crashes with PydanticSerializationError because the ADK Runner attaches non-serializable function objects to events before calling session_service.append_event().

The issue is in Runner._append_new_message_to_session ([runners.py:880](http://runners.py:880/)) which passes events containing function references to session_service.append_event(). The DatabaseSessionService then calls event.model_dump(exclude_none=True, mode="json") in StorageEvent.from_event() (schemas/[v1.py:194](http://v1.py:194/)), which fails on the non-serializable function objects.

This only affects persistent session services. InMemorySessionService works because it never serializes events.

Steps to Reproduce

1. Create an agent with tools (e.g., MCP tools via McpToolset)
2. Use DatabaseSessionService with PostgreSQL:

from google.adk.sessions.database_session_service import DatabaseSessionService
from google.adk.runners import Runner

runner = Runner(
 app_name="my-agent",
 agent=my_agent,
 session_service=DatabaseSessionService(
 db_url="postgresql+asyncpg://localhost:5432/mydb"
 ),
 artifact_service=InMemoryArtifactService(),
 memory_service=InMemoryMemoryService(),
 credential_service=InMemoryCredentialService(),
)

3. Send a message that triggers tool execution
4. Agent crashes on the first append_event call

Error Traceback

pydantic_core._pydantic_core.PydanticSerializationError: Unable to serialize unknown type: <class 'function'>

 File "google/adk/[runners.py](http://runners.py/)", line 880, in _append_new_message_to_session
 await self.session_service.append_event(session=session, event=event)

 File "google/adk/sessions/database_session_service.py", line 476, in append_event
 sql_session.add(schema.StorageEvent.from_event(session, event))

 File "google/adk/sessions/schemas/[v1.py](http://v1.py/)", line 194, in from_event
 event_data=event.model_dump(exclude_none=True, mode="json"),

pydantic_core._pydantic_core.PydanticSerializationError:
 Unable to serialize unknown type: <class 'function'>

Root Cause

StorageEvent.from_event() in schemas/[v1.py:194](http://v1.py:194/) calls:

event_data=event.model_dump(exclude_none=True, mode="json")

This does not use a fallback parameter, so any non-serializable type (like function objects attached by the Runner during tool resolution) causes a hard crash. Pydantic v2 supports a fallback parameter in model_dump() / model_dump_json() that can gracefully handle unknown types.

Environment

• google-adk: 1.22.1 (also confirmed on latest main branch / 1.26.0)
• Python: 3.13
• Pydantic: 2.12.5
• Database: PostgreSQL via asyncpg
• Tools: MCP tools via McpToolset / streamablehttp_client

Impact

• Severity: High - Any agent using persistent session services with tools will crash
• Scope: Affects DatabaseSessionService, and any custom persistent session service implementation
• Affected versions: All versions (1.0.0 through 1.26.0)

Related Issues

• Tool execution succeeds but agent crashes due to serialization of unused tool variable #1056 - Tool execution succeeds but agent crashes due to serialization of unused tool variable

• Unable to serialize Async generator type for streaming tool  #960 - Unable to serialize Async generator type for streaming tool

• EventCompaction deserialized as dict instead of Pydantic model in DatabaseSessionService #3633 - EventCompaction deserialized as dict instead of Pydantic model in DatabaseSessionService
  Model Information:

• Are you using LiteLLM: Yes

• Which model is being used: azure/gpt-5.1-chat

[llalitkumarrr]

Hello @varun-official ,

We have just released a new version i.e. google-adk==1.27.2. Could you please test it and let us know your observations?

[adk-bot]

This issue has been automatically marked as stale because it has not had recent activity for 7 days after a maintainer requested clarification. It will be closed if no further activity occurs within 7 days.