Example Overview

The three Python examples demonstrate a real-time market data pipeline.

Architecture

flowchart TD
    BINANCE[Binance WebSocket<br/>btcusdt@depth@100ms]
    ALPACA[Alpaca WebSocket<br/>BTC/USD orderbooks]

    WSM[WebSocketManager]
    WM[WorkManager]
    API[API Gateway]
    DAPR[Dapr State + Pub/Sub]

    BINANCE -->|wss://| WSM
    ALPACA -->|wss:// + auth| WSM
    WSM -->|Publish depth| DAPR
    DAPR -->|State store| API
    API -->|btcusdt-orderbook| STATE[(State Store)]
    WM -->|Poll state| API
    WM -->|Calculate imbalance| RESULT[btcusdt-imbalance]

Data Flow

1. WebSocketManager connects to Binance/Alpaca and:
2. Publishes raw depth to Dapr pub/sub (btcusdt-depth, btcusd-orderbook)

3. Stores top-20 order book snapshot to state store

4. API Gateway provides:

5. State persistence (examples/btcusdt-orderbook)

6. Pub/sub topic management

7. WorkManager imbalance worker:

8. Monitors state store for new order book data
9. Calculates weighted bid/ask imbalance
10. Outputs pressure signal: SELLING / BUYING / NEUTRAL

Key Concepts

Concept	Implementation
WebSocket connection	WebSocketManagerClient.connect() with auto_reconnect=True
Order book management	Local OrderBook class with snapshot + diff updates
API authentication	Alpaca sends auth message over WebSocket
State store persistence	API Gateway REST endpoints (SaveState, GetState)
Reconnection	Exponential backoff: 1s, 2s, 4s, 8s, 16s, max 30s
Bid/ask imbalance	Σ ask_price × ask_qty − Σ bid_price × bid_qty for top N levels