market-data-pipeline 0.8.11

Orchestration layer for market data pipelines

🚀 Market Data Pipeline

Enterprise-grade streaming DAG engine for real-time and batch market data with adaptive autoscaling

Features • Quick Start • Architecture • Documentation • Examples

---

📋 Table of Contents

• Overview
• Features
• Technology Stack
• Quick Start
• Architecture
• Installation
• Usage
• Configuration
• DAG Runtime
• KEDA Autoscaling
• Development
• Testing
• Production Deployment
• Monitoring
• Extending
• Contributing
• License

---

🎯 Overview

Market Data Pipeline is a production-grade orchestration system for real-time and batch market data processing. It provides a flexible, scalable platform for ingesting, transforming, and delivering market data from multiple sources (IBKR, replay, synthetic) to various sinks (TimescaleDB, Kafka, databases).

What It Does

Moves data from sources → through operators → into sinks with:

• ✅ Flow control & backpressure management
• ✅ Pacing compliance for API rate limits
• ✅ Event-time processing with watermarks
• ✅ Adaptive autoscaling via KEDA
• ✅ Comprehensive observability (Prometheus, Grafana)
• ✅ Production-ready deployment patterns

Use Cases

• Real-time market data ingestion from IBKR
• Historical data replay for backtesting
• High-frequency data aggregation (ticks → bars)
• Options chain processing with Greeks calculation
• Multi-provider data fusion and normalization
• Cost-optimized autoscaling deployments

---

✨ Features

Core Capabilities

• 🔄 Dual Runtime Modes

  • Classic Mode: Traditional pipeline orchestration
  • DAG Mode: Streaming directed acyclic graph engine

• 📊 Data Sources

  • IBKR: Live market data with reconnection handling
  • Replay: Historical data from Parquet files
  • Synthetic: Predictable data generation for testing

• ⚙️ Operators

  • Bar Aggregation: OHLCV bars with event-time correctness
  • Options Processing: Chain snapshots with Greeks
  • Windowing: Tumbling, sliding, and session windows
  • Partitioning: Hash-based data distribution
  • Custom: Extensible operator framework

• 📤 Sinks

  • Store: TimescaleDB via market_data_store
  • Kafka: High-throughput streaming
  • Database: Direct SQL writes with retries
  • Custom: Pluggable sink architecture

Advanced Features (Phase 5-6)

• 🎯 Streaming DAG Engine

  • Compose operators into directed acyclic graphs
  • Event-time windowing with watermarks
  • Hash partitioning for parallelism
  • Channel-based backpressure

• 📈 Adaptive Autoscaling

  • Dynamic rate adjustment based on downstream pressure
  • KEDA-based horizontal pod autoscaling
  • Prometheus metrics integration
  • Zero-touch operations in Kubernetes

• 🔧 Backpressure Feedback Loop

  • Closed-loop control from store to pipeline
  • Policy-based scaling (OK/SOFT/HARD)
  • Graceful degradation under load

Operational Excellence

• 🎛️ Flow Control: Bounded queues, drop policies, hybrid batching
• ⏱️ Pacing: Token-bucket rate limiting with provider-specific error codes
• 🔄 Retry Logic: Exponential backoff for transient failures
• 📊 Observability: 50+ Prometheus metrics, health checks, structured logs
• 🐳 Container-Native: Multi-stage Docker builds, Kubernetes manifests
• 🧪 Comprehensive Testing: 176 tests (unit + integration + load)

---

🛠️ Technology Stack

Core Technologies

Category	Technologies
Language	Python 3.11+
Async Runtime	asyncio, aiofiles, aiokafka
Web Framework	FastAPI, uvicorn
Data Processing	pandas, numpy, pyarrow
Configuration	Pydantic, pydantic-settings, YAML
Observability	Prometheus, OpenTelemetry, structlog
Message Queue	Kafka (aiokafka), MQTT
Database	TimescaleDB, PostgreSQL (via market_data_store)

Development Tools

Category	Technologies
Testing	pytest, pytest-asyncio, pytest-cov, freezegun
Code Quality	black, ruff, mypy
CI/CD	GitHub Actions, Docker, docker-compose
Container	Docker, Kubernetes, KEDA
Deployment	Helm (planned), docker-compose

Optional Dependencies

• DAG Runtime: mmh3 for fast hash partitioning
• IBKR: market_data_ibkr for live data
• Store: market_data_store for persistence
• Core: market_data_core for data models

---

🚀 Quick Start

Prerequisites

• Python 3.11 or higher
• Docker (optional, for full stack)
• Kubernetes cluster (optional, for KEDA autoscaling)

Local Development Setup

# Clone the repository
git clone https://github.com/mjdevaccount/market_data_pipeline.git
cd market_data_pipeline

# Create virtual environment
python -m venv .venv

# Activate (Windows PowerShell)
.\.venv\Scripts\Activate.ps1

# Activate (Linux/macOS)
source .venv/bin/activate

# Install with development dependencies
pip install -e ".[dev,dag]"

# Run tests
pytest tests/ -v

# Start the API server
mdp run --config configs/dag/bars.yaml

Docker Quick Start

# Start complete stack (pipeline + TimescaleDB)
docker compose up --build -d

# Run smoke test
./scripts/smoke_test.sh  # Linux/macOS
.\scripts\smoke_test.ps1  # Windows

# Check health
curl http://localhost:8083/health

# View metrics
curl http://localhost:8083/metrics

# Stop stack
docker compose down

Simple Python Example

from market_data_pipeline import create_pipeline

# Create a simple synthetic pipeline
pipeline = create_pipeline(
    tenant_id="demo",
    pipeline_id="quickstart",
    source="synthetic",
    symbols=["AAPL", "MSFT"],
    operator="bars",
    sink="store",
    duration_sec=30.0
)

# Run the pipeline
await pipeline.run()
await pipeline.close()

---

🏗️ Architecture

High-Level Architecture

┌──────────────────────────────────────────────────────────────────┐
│                    Market Data Pipeline                           │
└──────────────────────────────────────────────────────────────────┘
                              │
                              ▼
        ┌────────────────────────────────────┐
        │     Unified Runtime (v0.8.1)       │
        │  ┌──────────────┬──────────────┐   │
        │  │ Classic Mode │  DAG Mode    │   │
        │  └──────────────┴──────────────┘   │
        └────────────────────────────────────┘
                              │
        ┌─────────────────────┴─────────────────────┐
        ▼                     ▼                     ▼
┌───────────────┐     ┌───────────────┐    ┌───────────────┐
│   Sources     │     │   Operators   │    │    Sinks      │
│               │────▶│               │───▶│               │
│ • IBKR        │     │ • Bars        │    │ • Store       │
│ • Replay      │     │ • Options     │    │ • Kafka       │
│ • Synthetic   │     │ • Windows     │    │ • Database    │
│               │     │ • Partitions  │    │               │
└───────────────┘     └───────────────┘    └───────────────┘
        │                     │                     │
        └─────────────────────┴─────────────────────┘
                              │
                              ▼
                ┌──────────────────────────┐
                │   Observability Layer    │
                │                          │
                │  • Prometheus Metrics    │
                │  • Health Checks         │
                │  • Structured Logs       │
                └──────────────────────────┘
                              │
                              ▼
        ┌─────────────────────────────────────┐
        │   Kubernetes (KEDA Autoscaling)     │
        │                                     │
        │  ┌───────────┐    ┌──────────────┐ │
        │  │ Feedback  │───▶│ Rate         │ │
        │  │ Loop      │    │ Coordinator  │ │
        │  └───────────┘    └──────────────┘ │
        │         │                           │
        │         ▼                           │
        │  ┌────────────────────────────┐    │
        │  │ KEDA ScaledObject          │    │
        │  │ (Horizontal Autoscaling)   │    │
        │  └────────────────────────────┘    │
        └─────────────────────────────────────┘

Data Flow

Sources          Operators         Batchers         Sinks
──────────       ─────────────     ────────────     ─────────
IBKR API    ──▶  SecondBar    ──▶  Hybrid      ──▶  Store
Replay File ──▶  Aggregator   ──▶  Batcher     ──▶  (TimescaleDB)
Synthetic   ──▶  Options      ──▶  (Row/Byte/  ──▶  Kafka
                 Chain            Time limits)
                 
                 ↓ Event Time
                 ↓ Watermarks
                 ↓ Windowing

Phase 5: Streaming DAG Engine

┌────────────────────────────────────────────────────────────┐
│                    DAG Runtime (v0.8.0)                     │
├────────────────────────────────────────────────────────────┤
│                                                            │
│  Provider ──▶ Channel ──▶ Operator ──▶ Channel ──▶ Sink   │
│     │            │           │            │          │     │
│     │            │           │            │          │     │
│  (IBKR)      (bounded)   (transform)  (bounded)  (persist) │
│                 ▲                         ▲                │
│                 │                         │                │
│            Backpressure              Watermarks            │
│                                                            │
│  Features:                                                 │
│  • Event-time windowing                                    │
│  • Hash partitioning                                       │
│  • Contrib operators (OHLC, dedupe, throttle, router)     │
│  • Provider adapters (IBKR → DAG)                          │
│  • Unified CLI (mdp run --config)                          │
└────────────────────────────────────────────────────────────┘

Phase 6: Adaptive Autoscaling

┌────────────────────────────────────────────────────────────┐
│              Backpressure Feedback Loop (6.0A)              │
├────────────────────────────────────────────────────────────┤
│                                                            │
│  Store Queue  ──▶  FeedbackEvent  ──▶  FeedbackHandler    │
│  (9000/10000)      (level=HARD)        (scale=0.0)        │
│                                             │              │
│                                             ▼              │
│                                      RateCoordinator       │
│                                      (pause ingestion)     │
│                                                            │
└────────────────────────────────────────────────────────────┘

┌────────────────────────────────────────────────────────────┐
│                KEDA Autoscaling (6.0B)                      │
├────────────────────────────────────────────────────────────┤
│                                                            │
│  Metrics ──▶ Prometheus ──▶ KEDA ──▶ Kubernetes HPA       │
│  • queue_depth              Query     Scale: 1 → 8 pods   │
│  • backpressure_state       threshold                      │
│  • rate_scale_factor        (5000)                         │
│                                                            │
│  Result: Zero-touch horizontal scaling                     │
└────────────────────────────────────────────────────────────┘

Key Design Principles

1. Separation of Concerns: Clean abstractions between sources, operators, and sinks
2. Composability: Any source ↔ operator ↔ sink combination works
3. Resilience: Built-in retry, backpressure, and graceful degradation
4. Observability: Metrics at every stage, health endpoints, clear bottleneck detection
5. Performance: Async I/O, bounded queues, efficient batching
6. Type Safety: Pydantic models, mypy validation

---

📦 Installation

Standard Installation

# Basic installation
pip install -e .

# With development tools
pip install -e ".[dev]"

# With DAG runtime support
pip install -e ".[dag]"

# With all extras
pip install -e ".[dev,dag]"

Optional Dependencies

# IBKR live data (separate repository)
pip install market-data-ibkr

# TimescaleDB persistence (separate repository)
pip install market-data-store

# Core data models (separate repository)
pip install market-data-core

From Requirements

# Production dependencies (pinned versions)
pip install -r requirements.txt

# Development dependencies
pip install -r requirements.txt
pip install -e ".[dev,dag]"

Docker Installation

# Build image
docker build -t mdp-pipeline:latest .

# Or use docker-compose
docker compose up --build

---

💻 Usage

1. CLI (mdp command)

Unified Runtime

# Run DAG mode with YAML config
mdp run --config configs/dag/bars.yaml

# Run classic mode
mdp run --config configs/classic/bars.yaml

# List available commands
mdp --help

Legacy CLI

# Simple synthetic pipeline
mdp-legacy run --tenant T1 --symbols AAPL,MSFT --rate 100 --duration 30

# Using JSON spec
mdp-legacy runspec --spec examples/synthetic.json

# Start API server
mdp-legacy api --port 8000

2. Python API

Simple API

from market_data_pipeline import create_pipeline

# Quick start
pipeline = create_pipeline(
    tenant_id="demo",
    pipeline_id="simple",
    source="synthetic",
    symbols=["AAPL", "GOOGL"],
    operator="bars",
    sink="store",
    duration_sec=60.0
)

await pipeline.run()

Advanced Configuration

from market_data_pipeline import PipelineBuilder, PipelineSpec, PipelineOverrides

builder = PipelineBuilder()
spec = PipelineSpec(
    tenant_id="production",
    pipeline_id="hft_bars",
    source="ibkr",
    symbols=["SPY", "QQQ", "IWM"],
    operator="bars",
    sink="store",
    overrides=PipelineOverrides(
        ticks_per_sec=1000,
        pacing_max_per_sec=5000,
        batch_size=2000,
        sink_workers=8,
        bar_window_sec=5,
        bar_allowed_lateness_sec=2
    )
)

pipeline = builder.build(spec)
await pipeline.run()

Unified Runtime API

from market_data_pipeline.runtime import UnifiedRuntime
from market_data_pipeline.settings import UnifiedRuntimeSettings

# DAG mode
settings = UnifiedRuntimeSettings(
    mode="dag",
    dag={
        "graph": {
            "nodes": [
                {"id": "src", "type": "provider.ibkr.stream", "params": {...}},
                {"id": "op", "type": "operator.buffer", "params": {...}},
                {"id": "sink", "type": "operator.map", "params": {...}}
            ],
            "edges": [["src", "op"], ["op", "sink"]]
        }
    },
    feedback={"enable_feedback": True},
    metrics={"enable": True}
)

async with UnifiedRuntime(settings) as rt:
    await rt.run("my-job")

3. FastAPI Server

# Start server
uvicorn market_data_pipeline.runners.api:app --port 8083

# Health check
curl http://localhost:8083/health

# Create pipeline
curl -X POST http://localhost:8083/pipelines \
  -H "Content-Type: application/json" \
  -d '{
    "tenant_id": "T1",
    "pipeline_id": "test1",
    "source_type": "synthetic",
    "symbols": ["AAPL"],
    "rate": 100,
    "duration": 30
  }'

# Get metrics
curl http://localhost:8083/metrics

4. Docker Compose

# Start full stack
docker compose up -d

# View logs
docker compose logs -f pipeline

# Scale pipeline
docker compose up --scale pipeline=3

# Stop
docker compose down

---

⚙️ Configuration

Configuration Methods

1. Environment Variables (MDP_* prefix)
2. YAML Files (config.yaml, custom configs)
3. JSON Specs (examples/*.json)
4. Python Code (PipelineOverrides, settings objects)

Environment Variables

# Pipeline settings
export MDP_MODE=dag
export MDP_FB_ENABLE_FEEDBACK=true
export MDP_FB_PROVIDER_NAME=ibkr
export MDP_METRICS_ENABLE=true

# Database
export DATABASE_URL="postgresql://user:pass@localhost:5432/market_data"

# Kafka
export KAFKA_BOOTSTRAP_SERVERS="localhost:9092"
export KAFKA_TOPIC="market_data"

Example YAML Config

# configs/dag/bars.yaml
mode: dag
dag:
  nodes:
    - id: ibkr_source
      type: provider.ibkr.stream
      params:
        stream: "bars"
        symbols: ["AAPL", "MSFT"]
        resolution: "5s"
    - id: buffer
      type: operator.buffer
      params:
        max_items: 500
    - id: store_sink
      type: operator.map
      params:
        fn_name: "store_bars"
  edges:
    - [ibkr_source, buffer]
    - [buffer, store_sink]
feedback:
  enable_feedback: true
  provider_name: ibkr
metrics:
  enable: true

Configuration Overrides

PipelineOverrides(
    # Source
    ticks_per_sec=500,
    pacing_max_per_sec=2000,
    
    # Operator
    bar_window_sec=5,
    bar_allowed_lateness_sec=1,
    
    # Batcher
    batch_size=1000,
    max_bytes=1024_000,
    flush_ms=50,
    drop_policy="oldest",
    
    # Sink
    sink_workers=4,
    sink_queue_max=500
)

---

🎯 DAG Runtime (Phase 5)

The DAG runtime provides a powerful streaming engine for composing data pipelines.

Core Components

• Channels: Bounded queues with backpressure (high/low watermarks)
• Operators: Transform data (map, filter, buffer, window, partition)
• Registry: Component discovery and instantiation
• Builder: Graph construction from configuration

Example DAG

from market_data_pipeline.orchestration.dag import DagRuntime, Dag, Node, Edge

# Build DAG
dag = Dag()
dag.add_node(Node("source", meta={"type": "provider.ibkr.stream"}))
dag.add_node(Node("window", meta={"type": "operator.tumbling_window"}))
dag.add_node(Node("sink", meta={"type": "operator.map"}))
dag.add_edge(Edge("source", "window"))
dag.add_edge(Edge("window", "sink"))

# Run
runtime = DagRuntime(dag)
await runtime.run()

Windowing

from market_data_pipeline.orchestration.dag.windowing import (
    tumbling_window_event_time,
    TumblingWindowSpec
)

# 1-minute tumbling windows
async for frame in tumbling_window_event_time(
    source=data_stream,
    spec=TumblingWindowSpec(
        window_sec=60,
        watermark_lag_sec=5,
        allowed_lateness_sec=2
    )
):
    print(f"Window {frame.window_start} - {frame.window_end}: {len(frame.items)} items")

Partitioning

from market_data_pipeline.orchestration.dag.partitioning import hash_partition

# Partition by symbol
channels = await hash_partition(
    source=data_stream,
    key_fn=lambda item: item.symbol,
    num_partitions=4
)

# Process each partition
for i, channel in enumerate(channels):
    asyncio.create_task(process_partition(i, channel))

Contrib Operators

from market_data_pipeline.orchestration.dag.contrib.operators_contrib import (
    resample_ohlc,
    deduplicate,
    throttle,
    router
)

# OHLC resampling
bars = resample_ohlc(ticks, interval="1m")

# Deduplication
unique = deduplicate(stream, key_fn=lambda x: x.symbol, window_sec=10)

# Rate limiting
throttled = throttle(stream, max_per_sec=100)

# Routing
output_channels = router(stream, routes={
    "spy": lambda x: x.symbol == "SPY",
    "tech": lambda x: x.symbol in ["AAPL", "MSFT", "GOOGL"]
})

Complete DAG Example

See examples/run_dag_to_store.py for a full example with IBKR → DAG operators → Store sink.

---

📈 KEDA Autoscaling (Phase 6)

Phase 6.0A: Backpressure Feedback Loop

Dynamically adjusts pipeline ingestion rate based on downstream pressure.

# Configuration
feedback:
  enable_feedback: true
  provider_name: ibkr
  policy:
    ok: 1.0    # Full rate
    soft: 0.5  # Half rate
    hard: 0.0  # Paused

How It Works:

1. Store queue fills → FeedbackEvent published
2. FeedbackHandler receives event → adjusts RateCoordinator
3. Pipeline slows ingestion → queue drains
4. Feedback returns to OK → rate restored

Phase 6.0B: KEDA Horizontal Autoscaling

Automatically scales Kubernetes pods based on Prometheus metrics.

Metrics

# Queue depth (primary scaling metric)
pipeline_feedback_queue_depth{source="store_coordinator"}

# Backpressure state (0=ok, 1=soft, 2=hard)
pipeline_backpressure_state{provider="ibkr"}

# Rate scale factor (0.0-1.0)
pipeline_rate_scale_factor{provider="ibkr"}

Deployment

# Deploy to Kubernetes
kubectl apply -n market-data -f deploy/keda/deployment-pipeline.yaml
kubectl apply -n market-data -f deploy/keda/scaledobject-pipeline.yaml

# Watch scaling
kubectl get hpa -w
kubectl get pods -w

Scaling Behavior

Condition	Queue	Backpressure	Pods
Idle	< 2K	OK (0)	1
Moderate	5-7K	SOFT (1)	3-5
High	8-9K	SOFT (1)	6-8
Critical	9K+	HARD (2)	10 (max)

Documentation: See docs/PHASE_6.0B_KEDA_AUTOSCALING.md for complete guide.

---

🧑‍💻 Development

Setup Development Environment

# Clone repository
git clone https://github.com/mjdevaccount/market_data_pipeline.git
cd market_data_pipeline

# Create virtual environment
python -m venv .venv
source .venv/bin/activate  # Linux/macOS
.\.venv\Scripts\Activate.ps1  # Windows

# Install with dev dependencies
pip install -e ".[dev,dag]"

# Setup pre-commit (optional)
pre-commit install

Development Tools

# Code formatting
black src/ tests/
# or: ./scripts/dev.sh fmt

# Linting
ruff src/ tests/
# or: ./scripts/dev.sh lint

# Type checking
mypy src/

# Run tests
pytest tests/ -v
# or: ./scripts/dev.sh test

# Test coverage
pytest tests/ --cov=src --cov-report=html

# Clean caches
./scripts/dev.sh clean

Project Structure

market_data_pipeline/
├── src/market_data_pipeline/   # Source code
│   ├── source/                 # Data sources (IBKR, Synthetic, Replay)
│   ├── operator/               # Transformations (Bars, Options)
│   ├── batcher/                # Flow control (HybridBatcher)
│   ├── sink/                   # Outputs (Store, Kafka, Database)
│   ├── orchestration/          # DAG runtime, RateCoordinator
│   │   ├── dag/                # DAG engine (Phase 5)
│   │   ├── feedback/           # Backpressure feedback (Phase 6.0A)
│   │   └── coordinator.py      # Rate coordination
│   ├── runtime/                # Unified runtime (Phase 5.0.5)
│   ├── cli/                    # CLI commands
│   ├── runners/                # Service runners (API, CLI)
│   ├── settings/               # Configuration management
│   └── metrics.py              # Prometheus metrics
├── tests/                      # Test suite
│   ├── unit/                   # Unit tests
│   ├── integration/            # Integration tests
│   └── load/                   # Load tests
├── examples/                   # Usage examples
├── docs/                       # Documentation
├── deploy/                     # Deployment manifests
│   └── keda/                   # KEDA autoscaling (Phase 6.0B)
├── scripts/                    # Development scripts
└── configs/                    # Configuration files

---

🧪 Testing

Run Tests

# All tests
pytest tests/ -v

# Unit tests only
pytest tests/unit/ -v

# Integration tests (requires database)
export DATABASE_URL="postgresql://user:pass@localhost:5432/market_data"
pytest tests/integration/ -v

# Load tests
pytest tests/load/ -v

# With coverage
pytest tests/ --cov=src --cov-report=html
open htmlcov/index.html

# Contract tests (Core compatibility)
pytest tests/contracts/ -v

Test Categories

• Unit Tests (tests/unit/): 150+ tests covering individual components
• Integration Tests (tests/integration/): End-to-end workflows
• Load Tests (tests/load/): Performance benchmarks
• Contract Tests (tests/contracts/): Core v1.1.0 compatibility validation

Contract Tests

Contract tests verify compatibility with market-data-core protocol contracts:

# Run contract tests only
pytest tests/contracts/ -v

# Expected: 10 tests, ~4s duration

Purpose: These tests validate that Pipeline remains compatible with Core's data transfer objects (DTOs) and protocols. They are automatically triggered when Core publishes contract changes.

Categories:

• test_core_install.py: Core package imports and version compatibility
• test_feedback_flow.py: FeedbackEvent ↔ RateAdjustment transformations
• test_protocol_conformance.py: Protocol implementations (RateController, FeedbackPublisher)

Note: Contract tests are a minimal, fast subset of the full integration test suite. See tests/contracts/README.md for details.

Cross-Repo Testing: These tests are triggered automatically by market-data-core via GitHub Actions when Core's contract schemas change. See .github/workflows/README.md for workflow documentation

Pulse Integration (Phase 10.1)

The Pipeline subscribes to telemetry.feedback events from the Store via the Pulse event bus.

Configuration:

# Environment variables
PULSE_ENABLED=true                     # Enable Pulse (default: true)
EVENT_BUS_BACKEND=inmem                # inmem or redis (default: inmem)
REDIS_URL=redis://localhost:6379/0     # Redis connection (if backend=redis)
MD_NAMESPACE=mdp                       # Stream namespace
SCHEMA_TRACK=v1                        # Schema track (v1|v2)

Running Pulse Tests:

# Unit tests (inmem backend)
pytest tests/pulse/test_pulse_consumer.py -v

# Integration tests (redis backend, requires Redis)
EVENT_BUS_BACKEND=redis REDIS_URL=redis://localhost:6379/0 \
  pytest tests/pulse/test_redis_integration.py -v

Metrics:

• pulse_consume_total{stream,track,outcome} — Events consumed (success/error/duplicate)
• pulse_lag_ms{stream} — Consumer lag from event timestamp

Key Test Files

tests/unit/
├── orchestration/
│   ├── test_coordinator_feedback.py     # Phase 6.0A (13 tests)
│   ├── test_feedback_handler.py         # Phase 6.0A (12 tests)
├── metrics/
│   └── test_pipeline_metrics.py         # Phase 6.0B (8 tests)
├── dag/
│   ├── test_channel_backpressure.py     # Phase 5.0.1
│   ├── test_windowing_event_time.py     # Phase 5.0.2
│   └── test_contrib_operators.py        # Phase 5.0.3
└── ...

tests/integration/
├── test_feedback_integration.py         # Phase 6.0A (5 tests)
├── test_e2e_synthetic_store.py          # End-to-end
└── ...

Test Features

• ✅ Async Support: pytest-asyncio for async tests
• ✅ Time Control: freezegun for time-based testing
• ✅ Windows-Safe: Custom event loop fixtures
• ✅ Optional Deps: Graceful skipping when dependencies unavailable
• ✅ Real Code Paths: No mocks, tests exercise actual components

---

🚀 Production Deployment

Docker Deployment

# Build image
export REG="your-registry.example.com"
export TAG="0.8.1"
docker build -t $REG/mdp-pipeline:$TAG .
docker push $REG/mdp-pipeline:$TAG

Kubernetes Deployment

# Create namespace
kubectl create namespace market-data

# Deploy pipeline
kubectl apply -n market-data -f deploy/keda/deployment-pipeline.yaml

# Deploy KEDA autoscaling
kubectl apply -n market-data -f deploy/keda/scaledobject-pipeline.yaml

# Deploy Prometheus ServiceMonitor
kubectl apply -n market-data -f deploy/keda/prometheus-servicemonitor.yaml

# Verify
kubectl -n market-data get pods
kubectl -n market-data get scaledobject
kubectl -n market-data get hpa

Configuration

# Update deployment image
kubectl -n market-data set image deployment/mdp-pipeline \
  pipeline=$REG/mdp-pipeline:$TAG

# Update environment variables
kubectl -n market-data set env deployment/mdp-pipeline \
  MDP_FB_ENABLE_FEEDBACK=true \
  MDP_METRICS_ENABLE=true

Rollback

# Quick rollback
kubectl -n market-data rollout undo deployment/mdp-pipeline

# Rollback to specific version
kubectl -n market-data rollout undo deployment/mdp-pipeline --to-revision=2

Production Guides

• Deployment: See PHASE_6.0_PRODUCTION_ROLLOUT.md
• Quick Reference: See PHASE_6.0_SHIP_IT_CHECKLIST.md
• Verification: See PHASE_6.0_VERIFICATION_REPORT.md

---

📊 Monitoring & Observability

Prometheus Metrics

50+ metrics exposed at /metrics:

# Source metrics
mdp_source_items_total{source,tenant,symbol}
mdp_source_status{source,tenant}
mdp_source_pacing_blocked_total{source,tenant}

# Operator metrics
operator_bars_generated_total{tenant,pipeline,operator}

# Batcher metrics
batcher_items_added_total{tenant,pipeline,batcher}
batcher_queue_depth{tenant,pipeline,batcher}

# Sink metrics
mdp_sink_batches_written_total{sink,tenant}
mdp_sink_commit_seconds{sink,tenant}
mdp_sink_queue_depth{sink,tenant}

# Phase 6.0B KEDA metrics
pipeline_rate_scale_factor{provider}
pipeline_backpressure_state{provider}
pipeline_feedback_queue_depth{source}

Grafana Dashboard

# Import dashboard
kubectl -n monitoring apply -f monitoring/grafana-dashboard.json

Panels:

1. Queue Depth (time series)
2. Rate Scale Factor (gauge)
3. Backpressure State (stat)
4. Pod Count (stat)
5. Throughput (graph)
6. Latency (heatmap)

Health Checks

# Application health
curl http://localhost:8083/health
# {"status":"healthy","service":"market-data-pipeline"}

# Kubernetes liveness
kubectl -n market-data get pods
# Shows READY 1/1

# Detailed status
curl http://localhost:8083/status

Alerts

Recommended Prometheus alerts:

# High queue depth
- alert: PipelineQueueHigh
  expr: pipeline_feedback_queue_depth > 8000
  for: 5m

# Hard backpressure
- alert: PipelineHardBackpressure
  expr: pipeline_backpressure_state == 2
  for: 2m

# Low rate scale
- alert: PipelineRateScaleLow
  expr: pipeline_rate_scale_factor < 0.3
  for: 10m

---

🔧 Extending the Pipeline

Add a Custom Operator

from market_data_pipeline.operator.base import StatefulOperator, EventTimePolicy

class VWAPOperator(StatefulOperator):
    def __init__(self, window_sec: int = 60):
        policy = EventTimePolicy(window_sec=window_sec)
        super().__init__(policy=policy)
        self._volume_prices = {}
        self._volumes = {}
        
    def _on_tick(self, tick: Quote) -> None:
        symbol = tick.symbol
        if symbol not in self._volume_prices:
            self._volume_prices[symbol] = 0.0
            self._volumes[symbol] = 0
            
        self._volume_prices[symbol] += tick.price * tick.size
        self._volumes[symbol] += tick.size
        
        # Check if window complete (via EventTimePolicy)
        if self._should_flush(tick.timestamp):
            vwap = self._volume_prices[symbol] / self._volumes[symbol]
            self._ready.append({"symbol": symbol, "vwap": vwap})
            self._volume_prices[symbol] = 0.0
            self._volumes[symbol] = 0

Add a Custom Sink

from market_data_pipeline.sink.base import Sink
from market_data_pipeline.sink.capabilities import SinkCapabilities

class S3Sink(Sink):
    def __init__(self, bucket: str, prefix: str):
        self.bucket = bucket
        self.prefix = prefix
        self.s3_client = None
        
    @property
    def capabilities(self) -> SinkCapabilities:
        return SinkCapabilities.BATCH_WRITES
        
    async def start(self) -> None:
        import aioboto3
        self.session = aioboto3.Session()
        self.s3_client = await self.session.client('s3')
        
    async def write(self, batch: List[Bar]) -> None:
        key = f"{self.prefix}/{batch[0].timestamp.date()}.parquet"
        df = pd.DataFrame([b.model_dump() for b in batch])
        buffer = io.BytesIO()
        df.to_parquet(buffer, engine='pyarrow')
        buffer.seek(0)
        
        await self.s3_client.put_object(
            Bucket=self.bucket,
            Key=key,
            Body=buffer.getvalue()
        )
        
    async def close(self, drain: bool = True) -> None:
        if self.s3_client:
            await self.s3_client.close()

Register Components

from market_data_pipeline.orchestration.dag.registry import default_registry

registry = default_registry()
registry.register_operator("operator.vwap", VWAPOperator)
registry.register_sink("sink.s3", S3Sink)

---

🤝 Contributing

We welcome contributions! Please follow these guidelines:

Getting Started

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Make your changes
4. Run tests (pytest tests/ -v)
5. Run linters (black ., ruff ., mypy src/)
6. Commit changes (git commit -m 'Add amazing feature')
7. Push to branch (git push origin feature/amazing-feature)
8. Open a Pull Request

Code Standards

• Style: Follow black formatting (88 char line length)
• Linting: Pass ruff checks
• Type Hints: Full type annotations (mypy strict mode)
• Tests: Add tests for new features (maintain >95% coverage)
• Documentation: Update README and docstrings

Commit Messages

Follow conventional commits:

• feat: New feature
• fix: Bug fix
• docs: Documentation changes
• test: Test changes
• refactor: Code refactoring
• perf: Performance improvements

Pull Request Process

1. Update README.md with details of changes if applicable
2. Update documentation in docs/ if needed
3. Add tests for new functionality
4. Ensure all tests pass
5. Request review from maintainers

---

📚 Documentation

Core Documentation

• README.md - This file
• PIPELINE_BUILDER.md - Pipeline builder guide
• PRODUCTION.md - Production deployment guide

Phase Documentation

• Phase 5: Streaming DAG Engine

  • Phase 5 Overview
  • Phase 5.0.1: Core DAG Runtime
  • Phase 5.0.2: Windowing & Partitioning
  • Phase 5.0.3: Contrib Operators
  • Phase 5.0.4: Provider Integration
  • Phase 5.0.5: Unified Runtime

• Phase 6: Adaptive Autoscaling

  • Phase 6.0A: Backpressure Feedback
  • Phase 6.0B: KEDA Autoscaling
  • KEDA Deployment Guide
  • Production Rollout
  • Ship It Checklist

Examples

• examples/ - Code examples
  • run_dag_to_store.py - Complete DAG example
  • run_unified_runtime_basic.py - Unified runtime example
  • synthetic.json - JSON spec example
  • And 15+ more examples

---

🗺️ Roadmap

Completed ✅

• v0.1-0.2: Core abstractions, pipeline builder, JSON-spec
• v0.7-0.8: DAG runtime, unified runtime, adaptive autoscaling

Current Version: v0.9.0

• ✅ Streaming DAG engine
• ✅ Event-time windowing
• ✅ Hash partitioning
• ✅ Backpressure feedback loop
• ✅ KEDA autoscaling
• ✅ 176 tests passing

Future (v0.9+)

• v0.9: Enhanced IBKR integration

  • Reconnection handling
  • Pacing improvements
  • Error recovery

• v1.0: Multi-provider composite sources

  • Cross-exchange arbitrage
  • Provider failover
  • Data fusion

• v1.1: Advanced analytics

  • Volatility surfaces
  • Greeks aggregation
  • ML feature engineering

• v1.2: Horizontal scaling

  • Distributed DAG execution
  • Coordinator sharding
  • Cross-region deployment

---

🔗 Related Projects

• market_data_core - Core data models and provider SDKs
• market_data_store - TimescaleDB persistence layer
• market_data_ibkr - IBKR provider implementation

---

📜 License

MIT License - see LICENSE for details

---

👥 Authors

• Primary Developer - @mjdevaccount

---

🙏 Acknowledgments

• FastAPI - Modern web framework
• Pydantic - Data validation
• Prometheus - Metrics collection
• KEDA - Kubernetes autoscaling
• AsyncIO - Asynchronous I/O

---

📞 Support

• Issues: GitHub Issues
• Discussions: GitHub Discussions

---

Made with ❤️ for the trading community

⬆ Back to Top