Feature Idea: HBaseSessionService for low-latency, scalable session storage

[vaibhav-Singh9888]

Feature Idea: HBaseSessionService for low-latency, scalable session storage

Hello ADK Team and Community,

I'm a user of the google-adk-python framework and have been very impressed with its design, particularly the extensible SessionService architecture.

Motivation

For many real-time chat and agent applications, the latency of session storage is a critical performance factor. While the existing DatabaseSessionService is great for relational persistence and VertexAiSessionService is perfect for GCP-native deployments, there's an opportunity for a solution tailored to large-scale, self-hosted environments that demand extremely low-latency reads and writes.

Apache HBase is a distributed, non-relational database modeled after Google's Bigtable, designed for high-throughput, real-time access to massive datasets. Its architecture makes it an ideal backend for storing and retrieving conversation histories in high-traffic agent applications.

Proposal

I propose the addition of a new HBaseSessionService to the ADK. This would provide a first-class integration for developers who use or want to use an HBase cluster for managing session state and history.

I have already developed a working implementation of this for a personal project and believe it would be a valuable addition to the framework.

Here is how the new service would fit into the existing structure:

HBaseSessionService

• How it works: Connects to an Apache HBase cluster using a client library like happybase. Session data (id, state, events, etc.) is stored in an HBase table, with a row key designed for efficient lookups (e.g., a composite of app_name, user_id, and session_id). This structure leverages HBase's strengths in fast key-based retrieval.

• Persistence: Yes. Data is durably stored in the distributed HBase cluster and survives application restarts.

• Requires:

  • An accessible Apache HBase cluster.
  • The Python client library for HBase (e.g., happybase). This could be added as an optional dependency: pip install google-adk[hbase].

• Best for:

  • Production applications with very high transaction volumes.
  • Scenarios where microsecond-level latency for session retrieval and updates is critical.
  • Teams that already manage a Big Data ecosystem (e.g., Hadoop/HBase) and want to integrate their ADK agents with it.

Example Usage

The developer experience would be consistent with the other services:

# Requires: pip install google-adk[hbase] happybase
from google.adk.sessions import HBaseSessionService

# Connection details for the HBase Thrift server
HBASE_HOST = "your-hbase-host"
HBASE_PORT = 9090
HBASE_TABLE_NAME = "adk_sessions"

session_service = HBaseSessionService(
    host=HBASE_HOST,
    port=HBASE_PORT,
    table_name=HBASE_TABLE_NAME
)

# Usage would be identical to other services
# new_session = await session_service.create_session(app_name="my_app", ...)

Next Steps

I am opening this discussion to gauge the team's interest and get feedback on this proposal. If this aligns with the project's vision, I would be more than happy to clean up my existing code, add the necessary tests and documentation, and submit a pull request for review.

Thank you for your time and consideration!