feat(memory): add DatabaseMemoryService with SQLAlchemy async backend

[anmolg1997]

Summary

Adds DatabaseMemoryService, a durable, SQL-backed memory service that works with any SQLAlchemy-compatible async database (SQLite via aiosqlite, PostgreSQL via asyncpg, MySQL/MariaDB, etc.).

• Fills the gap between the volatile InMemoryMemoryService and the cloud-only Firestore/Vertex AI options. Gives self-hosted deployments a persistent memory backend with zero cloud dependencies
• Mirrors the keyword-extraction approach of FirestoreMemoryService and reuses the SQLAlchemy patterns from DatabaseSessionService
• Implements add_session_to_memory, add_events_to_memory (delta ingestion), and search_memory
• Includes lazy table creation with double-checked locking, async context manager support, and proper rollback-on-exception handling

Motivation

Currently, developers not using Vertex AI or Google Cloud Firestore have no persistent memory option. InMemoryMemoryService is explicitly volatile and loses all data on restart. This is a frequently requested feature:

• Closes Add support for additional Memory Bank services: DatabaseMemoryService/RedisMemoryService #2524, "Add support for additional Memory Bank services: DatabaseMemoryService"
• Closes SQLite-based memory service implementation #2976, "SQLite-based memory service implementation"

Design Decisions

Decision	Rationale
Placed in google.adk.memory (not integrations/)	Follows DatabaseSessionService placement in sessions/, and uses no external cloud SDK
Reuses DynamicJSON from sessions.schemas.shared	Consistent JSON/JSONB handling across SQLAlchemy services
Separate _memory_schemas.py for ORM models	Mirrors the sessions/schemas/ pattern, keeps concerns separated
Keyword search (not semantic/vector)	Matches FirestoreMemoryService approach. Semantic search can be added via a future SearchBackend abstraction
Lazy __getattr__ import in __init__.py	Matches DatabaseSessionService pattern, avoids hard SQLAlchemy dependency for users who don't need it

Changes

File	Description
src/google/adk/memory/database_memory_service.py	New: DatabaseMemoryService implementation
src/google/adk/memory/_memory_schemas.py	New: SQLAlchemy ORM model (StorageMemoryEntry)
src/google/adk/memory/__init__.py	Updated: lazy export of DatabaseMemoryService
tests/unittests/memory/test_database_memory_service.py	New: 19 tests covering CRUD, search, scoping, dedup, persistence

Usage

from google.adk.memory import DatabaseMemoryService

# SQLite (zero-config)
memory = DatabaseMemoryService("sqlite+aiosqlite:///memory.db")

# PostgreSQL
memory = DatabaseMemoryService("postgresql+asyncpg://user:pass@host/db")

async with memory:
    await memory.add_session_to_memory(session)
    result = await memory.search_memory(
        app_name="my_app", user_id="u1", query="agent tool usage"
    )

Test plan

• 19 unit tests passing (pytest tests/unittests/memory/test_database_memory_service.py)
• All 59 existing memory tests still pass (pytest tests/unittests/memory/)
• Code formatted with isort + pyink per CONTRIBUTING.md
• Tested with in-memory SQLite and file-based SQLite
• Context manager lifecycle (__aenter__/__aexit__) verified

[adk-bot]

Response from ADK Triaging Agent

Hello @anmolg1997, thank you for creating this PR!

Could you please provide logs or a screenshot of the passed pytest results?

This information will help reviewers to review your PR more efficiently. Thanks!

[anmolg1997]

Hi @adk-bot, thanks for the prompt. Full pytest results below.

New tests (19/19 passing):

tests/unittests/memory/test_database_memory_service.py::test_extract_keywords PASSED
tests/unittests/memory/test_database_memory_service.py::test_add_session_to_memory PASSED
tests/unittests/memory/test_database_memory_service.py::test_add_session_with_no_events PASSED
tests/unittests/memory/test_database_memory_service.py::test_add_session_to_memory_filters_no_content_events PASSED
tests/unittests/memory/test_database_memory_service.py::test_add_session_to_memory_skips_stop_words_only PASSED
tests/unittests/memory/test_database_memory_service.py::test_search_memory_empty_query PASSED
tests/unittests/memory/test_database_memory_service.py::test_search_memory_only_stop_words PASSED
tests/unittests/memory/test_database_memory_service.py::test_search_memory_simple_match PASSED
tests/unittests/memory/test_database_memory_service.py::test_search_memory_case_insensitive PASSED
tests/unittests/memory/test_database_memory_service.py::test_search_memory_multiple_matches PASSED
tests/unittests/memory/test_database_memory_service.py::test_search_memory_no_match PASSED
tests/unittests/memory/test_database_memory_service.py::test_search_memory_scoped_by_user PASSED
tests/unittests/memory/test_database_memory_service.py::test_search_memory_deduplication PASSED
tests/unittests/memory/test_database_memory_service.py::test_add_events_to_memory PASSED
tests/unittests/memory/test_database_memory_service.py::test_add_events_to_memory_without_session_id PASSED
tests/unittests/memory/test_database_memory_service.py::test_add_events_to_memory_skips_empty PASSED
tests/unittests/memory/test_database_memory_service.py::test_context_manager PASSED
tests/unittests/memory/test_database_memory_service.py::test_file_based_sqlite PASSED
tests/unittests/memory/test_database_memory_service.py::test_invalid_db_url PASSED
======================== 19 passed in 1.16s ========================

Full memory test suite (59/59 passing, no regressions):

tests/unittests/memory/test_database_memory_service.py: 19 passed
tests/unittests/memory/test_in_memory_memory_service.py: 11 passed
tests/unittests/memory/test_vertex_ai_memory_bank_service.py: 29 passed
======================= 59 passed in 16.37s ========================

All CI checks (CLA, header-check, agent-triage) are also passing.

[anmolg1997]

Hi @wuliang229, thanks for approving #5338. Would you be able to take a look at this one as well when you get a chance? It adds a SQL-backed DatabaseMemoryService (mirrors the existing DatabaseSessionService patterns) with 19 unit tests, all passing. Appreciate it.

[rohityan]

Hi @anmolg1997 , Thank you for your contribution! We appreciate you taking the time to submit this pull request.
Closing this PR here as it belongs to adk-python-community repo.
We highly recommend releasing the feature as a standalone package that we will then share through: https://google.github.io/adk-docs/integrations/