a2a-redis

Redis integrations for the Agent-to-Agent (A2A) Python SDK.

This package provides Redis-backed implementations of core A2A components for persistent task storage, reliable event queue management, and push notification configuration using Redis.

Features

• RedisTaskStore & RedisJSONTaskStore: Redis-backed task storage using hashes or JSON
• RedisStreamsQueueManager & RedisStreamsEventQueue: Persistent, reliable event queues with consumer groups
• RedisPubSubQueueManager & RedisPubSubEventQueue: Real-time, low-latency event broadcasting
• RedisPushNotificationConfigStore: Task-based push notification configuration storage
• Consumer Group Strategies for Streams: Flexible load balancing and instance isolation patterns

Installation

pip install a2a-redis

Quick Start

from a2a_redis import RedisTaskStore, RedisStreamsQueueManager, RedisPushNotificationConfigStore
from a2a_redis.utils import create_redis_client
from a2a.server.request_handlers import DefaultRequestHandler
from a2a.server.apps import A2AStarletteApplication

# Create Redis client with connection management
redis_client = create_redis_client(url="redis://localhost:6379/0", max_connections=50)

# Initialize Redis components
task_store = RedisTaskStore(redis_client, prefix="myapp:tasks:")
queue_manager = RedisStreamsQueueManager(redis_client, prefix="myapp:queues:")
push_config_store = RedisPushNotificationConfigStore(redis_client, prefix="myapp:push:")

# Use with A2A request handler
request_handler = DefaultRequestHandler(
    agent_executor=YourAgentExecutor(),
    task_store=task_store,
    queue_manager=queue_manager,
    push_config_store=push_config_store,
)

# Create A2A server
server = A2AStarletteApplication(
    agent_card=your_agent_card,
    http_handler=request_handler
)

Queue Components

The package provides both high-level queue managers and direct queue implementations:

Queue Managers

• RedisStreamsQueueManager - Manages Redis Streams-based queues
• RedisPubSubQueueManager - Manages Redis Pub/Sub-based queues
• Both implement the A2A SDK's QueueManager interface

Event Queues

• RedisStreamsEventQueue - Direct Redis Streams queue implementation
• RedisPubSubEventQueue - Direct Redis Pub/Sub queue implementation
• Both implement the EventQueue interface through protocol compliance

Queue Manager Types: Streams vs Pub/Sub

RedisStreamsQueueManager

Key Features:

• Persistent storage: Events remain in streams until explicitly trimmed
• Guaranteed delivery: Consumer groups with acknowledgments prevent message loss
• Load balancing: Multiple consumers can share work via consumer groups
• Failure recovery: Unacknowledged messages can be reclaimed by other consumers
• Event replay: Historical events can be re-read from any point in time
• Ordering: Maintains strict insertion order with unique message IDs

Use Cases:

• Task event queues requiring reliability
• Audit trails and event history
• Work distribution systems
• Systems requiring failure recovery
• Multi-consumer load balancing

Trade-offs:

• Higher memory usage (events persist)
• More complex setup (consumer groups)
• Slightly higher latency than pub/sub

RedisPubSubQueueManager

Key Features:

• Real-time delivery: Events delivered immediately to active subscribers
• No persistence: Events not stored, only delivered to active consumers
• Fire-and-forget: No acknowledgments or delivery guarantees
• Broadcasting: All subscribers receive all events
• Low latency: Minimal overhead for immediate delivery
• Minimal memory usage: No storage of events

Use Cases:

• Live status updates and notifications
• Real-time dashboard updates
• System event broadcasting
• Non-critical event distribution
• Low-latency requirements
• Simple fan-out scenarios

Not suitable for:

• Critical event processing requiring guarantees
• Systems requiring event replay or audit trails
• Offline-capable applications
• Work queues requiring load balancing

Components

Task Storage

RedisTaskStore

Stores task data in Redis using hashes with JSON serialization. Works with any Redis server.

from a2a_redis import RedisTaskStore

task_store = RedisTaskStore(redis_client, prefix="mytasks:")

# A2A TaskStore interface methods
await task_store.save("task123", {"status": "pending", "data": {"key": "value"}})
task = await task_store.get("task123")
success = await task_store.delete("task123")

# List all task IDs (utility method)
task_ids = await task_store.list_task_ids()

RedisJSONTaskStore

Stores task data using Redis's JSON module for native JSON operations and complex nested data.

from a2a_redis import RedisJSONTaskStore

# Requires Redis 8 or RedisJSON module
json_task_store = RedisJSONTaskStore(redis_client, prefix="mytasks:")

# Same interface as RedisTaskStore but with native JSON support
await json_task_store.save("task123", {"complex": {"nested": {"data": "value"}}})

Queue Managers

Both queue managers implement the A2A QueueManager interface with full async support:

import asyncio
from a2a_redis import RedisStreamsQueueManager, RedisPubSubQueueManager
from a2a_redis.streams_consumer_strategy import ConsumerGroupConfig, ConsumerGroupStrategy

# Choose based on your requirements:

# For reliable, persistent processing
streams_manager = RedisStreamsQueueManager(redis_client, prefix="myapp:streams:")

# For real-time, low-latency broadcasting
pubsub_manager = RedisPubSubQueueManager(redis_client, prefix="myapp:pubsub:")

# With custom consumer group configuration (streams only)
config = ConsumerGroupConfig(strategy=ConsumerGroupStrategy.SHARED_LOAD_BALANCING)
streams_manager = RedisStreamsQueueManager(redis_client, consumer_config=config)

async def main():
    # Same interface for both managers
    queue = await streams_manager.create_or_tap("task123")

    # Enqueue events
    await queue.enqueue_event({"type": "progress", "message": "Task started"})
    await queue.enqueue_event({"type": "progress", "message": "50% complete"})

    # Dequeue events
    try:
        event = await queue.dequeue_event(no_wait=True)  # Non-blocking
        print(f"Got event: {event}")
        await queue.task_done()  # Acknowledge the message (streams only)
    except RuntimeError:
        print("No events available")

    # Close queue when done
    await queue.close()

asyncio.run(main())

Consumer Group Strategies

The Streams queue manager supports different consumer group strategies:

from a2a_redis.streams_consumer_strategy import ConsumerGroupStrategy, ConsumerGroupConfig

# Multiple instances share work across a single consumer group
config = ConsumerGroupConfig(strategy=ConsumerGroupStrategy.SHARED_LOAD_BALANCING)

# Each instance gets its own consumer group
config = ConsumerGroupConfig(strategy=ConsumerGroupStrategy.INSTANCE_ISOLATED)

# Custom consumer group name
config = ConsumerGroupConfig(strategy=ConsumerGroupStrategy.CUSTOM, group_name="my_group")

streams_manager = RedisStreamsQueueManager(redis_client, consumer_config=config)

RedisPushNotificationConfigStore

Stores push notification configurations per task. Implements the A2A PushNotificationConfigStore interface.

from a2a_redis import RedisPushNotificationConfigStore
from a2a.types import PushNotificationConfig

config_store = RedisPushNotificationConfigStore(redis_client, prefix="myapp:push:")

# Create push notification config
config = PushNotificationConfig(
    url="https://webhook.example.com/notify",
    token="secret_token",
    id="webhook_1"
)

# A2A interface methods
await config_store.set_info("task123", config)

# Get all configs for a task
configs = await config_store.get_info("task123")
for config in configs:
    print(f"Config {config.id}: {config.url}")

# Delete specific config or all configs for a task
await config_store.delete_info("task123", "webhook_1")  # Delete specific
await config_store.delete_info("task123")  # Delete all

Requirements

• Python 3.11+
• redis >= 4.0.0
• a2a-sdk >= 0.2.16 (Agent-to-Agent Python SDK)
• uvicorn >= 0.35.0

Optional Dependencies

• RedisJSON module for RedisJSONTaskStore (enhanced nested data support)
• Redis Stack or Redis with modules for full feature support

Development

# Clone the repository
git clone https://github.com/a2aproject/a2a-redis.git
cd a2a-redis

# Create virtual environment and install dependencies
uv venv
source .venv/bin/activate  # or .venv\Scripts\activate on Windows
uv sync --dev

# Run tests with coverage
uv run pytest --cov=a2a_redis --cov-report=term-missing

# Run linting and formatting
uv run ruff check src/ tests/
uv run ruff format src/ tests/
uv run pyright src/

# Install pre-commit hooks
uv run pre-commit install

# Run examples
uv run python examples/basic_usage.py
uv run python examples/redis_travel_agent.py

Testing

Tests use Redis database 15 for isolation and include both mock and real Redis integration tests:

# Run all tests
uv run pytest

# Run specific test file
uv run pytest tests/test_streams_queue_manager.py -v

# Run with coverage
uv run pytest --cov=a2a_redis --cov-report=term-missing

License

MIT License