Description

When running multiple instances of an application using DatabaseSessionService (e.g., multiple Kubernetes pods, multiple containers, or multiple processes), a race condition occurs during table creation that causes instances to crash with a PostgreSQL constraint violation error.

Environment

• ADK Version: 1.3.0
• Python Version: 3.13
• Database: PostgreSQL 17
• Deployment: Kubernetes with 2+ replicas / Multiple Docker containers / Multiple processes

Error Message

sqlalchemy.exc.IntegrityError: (psycopg.errors.UniqueViolation) duplicate key value violates unique constraint "pg_type_typname_nsp_index"
DETAIL:  Key (typname, typnamespace)=(sessions, 16385) already exists.
[SQL: 
CREATE TABLE sessions (
    app_name VARCHAR(128) NOT NULL, 
    user_id VARCHAR(128) NOT NULL, 
    id VARCHAR(128) NOT NULL, 
    state JSONB NOT NULL, 
    create_time TIMESTAMP WITHOUT TIME ZONE NOT NULL, 
    update_time TIMESTAMP WITHOUT TIME ZONE NOT NULL, 
    PRIMARY KEY (app_name, user_id, id)
)
]

Root Cause

The issue occurs in database_session_service.py at line 343:

class DatabaseSessionService(BaseSessionService):
    def __init__(self, db_url: str, **kwargs: Any):
        # ... 
        # Line 343 - This creates tables without checking for concurrent access
        Base.metadata.create_all(self.db_engine)

When multiple instances start simultaneously:

1. Instance A checks if tables exist (they don't)
2. Instance B checks if tables exist (they don't)
3. Instance A starts creating tables
4. Instance B starts creating tables
5. PostgreSQL throws a constraint violation on its internal pg_type catalog

Steps to Reproduce

1. Create an application using DatabaseSessionService:

from google.adk.sessions import DatabaseSessionService

session_service = DatabaseSessionService(
    db_url="postgresql://user:pass@host/db",
    pool_size=10
)

2. Deploy with multiple replicas (Kubernetes example):

apiVersion: apps/v1
kind: Deployment
spec:
  replicas: 2  # Multiple replicas trigger the race condition
  template:
    spec:
      containers:
      - name: app
        image: myapp:latest

3. Both pods will attempt to start simultaneously, and one will crash with the pg_type error

Expected Behavior

DatabaseSessionService should handle concurrent initialization gracefully without crashes. Multiple instances should be able to start simultaneously without race conditions.

Suggested Fixes

1. Consider using checkfirst=True parameter.

Base.metadata.create_all(self.db_engine, checkfirst=True)