healthz

healthz

Interactive HTTP health monitoring tool with real-time metrics visualization and per-second granularity.

Monitor your APIs, detect blocking in async services, and find performance bottlenecks with a beautiful terminal UI inspired by btop.

Features

• 🚀 Constant-rate load testing - Send requests at exact intervals (no batching)
• 📊 Per-second metrics - True temporal granularity for p95/p99 latency
• 📈 Real-time graphs - Symmetrical butterfly chart showing p99/p95 trends
• ⌨️ Interactive controls - Adjust rate and window on-the-fly without restart
• 🎯 Blocking detection - Track pending requests and timeouts to detect event loop saturation
• 💾 Export data - Save per-second aggregated metrics to CSV
• 🎨 Nord-themed UI - Beautiful color scheme optimized for terminals

Installation

# Install from PyPI
pip install healthz

# Or install from source
git clone https://github.com/Delos-Intelligence/healthz.git
cd healthz
pip install -e .

Usage

Basic

# Monitor a health endpoint at 30 requests/second
healthz http://localhost:8000/healthz --rate 30

Advanced

# POST request with custom headers
healthz http://api.example.com/endpoint \
  --rate 50 \
  --method POST \
  --header "Authorization: Bearer token" \
  --data '{"check": true}'

Interactive Controls

While the TUI is running:

Key	Action
↑ / +	Increase request rate (+10 req/s)
↓ / -	Decrease request rate (-10 req/s)
w	Increase time window (+5s)
s	Decrease time window (-5s)
p	Pause/resume sending requests
r	Reset all statistics
e	Export per-second metrics to CSV
q	Quit

Display

The TUI shows four panels:

• Metrics Panel (top): Request counts, success rate, pending requests with health status indicator
• Response Time Stats (left): Min, Avg, p50, p95, p99, Max latency with color coding
• Distribution Histogram (right): Response time distribution across buckets
• Butterfly Chart (bottom): Symmetrical p99 (top) / p95 (bottom) time-series graph

Health Status Colors

• 🟢 Green (Healthy): Pending < rate
• 🟡 Yellow (Degraded): Pending < rate × 3
• 🔴 Red (Blocked): Pending ≥ rate × 3

Use Cases

Detect Blocking in Async Services

Monitor /healthz while load testing other endpoints to detect when your async event loop gets saturated:

# Terminal 1: Monitor health check
healthz http://localhost:8000/healthz --rate 30

# Terminal 2: Load your API
while true; do
  curl -X POST http://localhost:8000/api/endpoint \
    -H "Content-Type: application/json" \
    -d '{"data": "test"}'
done

If the health check starts showing high pending counts or timeouts, your service is blocking!

Find Resource Limits

Gradually increase the rate to find when your service starts degrading:

• Start at low rate: healthz URL --rate 10
• Press ↑ repeatedly to increase rate
• Watch for pending count to rise or p95/p99 to spike
• That's your service's limit!

Compare Before/After

Export baseline metrics, make code changes, run again and compare:

# Before optimization
healthz http://localhost:8000/healthz --rate 50
# Press 'e' to export to healthz_metrics_YYYYMMDD_HHMMSS.csv

# After optimization
healthz http://localhost:8000/healthz --rate 50
# Press 'e' to export again and compare

Architecture

healthz uses a per-second bucket architecture for efficient memory usage:

• Each second gets a SecondBucket storing aggregated stats (success/timeout/error counts, response times)
• When a response arrives, it updates the bucket for the second the request was sent (not completed)
• All displays (metrics, histogram, graph) read from these buckets
• Individual request objects are never stored - only aggregated data
• Old buckets (>5 minutes) are automatically cleaned up

This means you get accurate per-second granularity while using minimal memory.

Development

# Clone and install in development mode
git clone https://github.com/Delos-Intelligence/healthz.git
cd healthz
pip install -e ".[dev]"

# Run tests
pytest

# Format code
black healthz/
ruff check healthz/

Requirements

• Python 3.10+
• aiohttp
• textual
• textual-plotext
• click

License

MIT License - see LICENSE file for details.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.