How We Built a Sub-10ms Latency Monitoring Engine in Python & Docker

Measuring network metrics accurately across global edges requires more than just calling basic HTTP client methods. When microservices depend on real-time execution, small spikes in DNS resolution or TLS handshake times can degrade system telemetry.

In this guide, we'll break down how to structure a lightweight, asynchronous network performance testing suite in Python, containerized with Docker for multi-region deployment.

1. Deconstructing the Latency Timeline

When executing web requests, total round-trip time (RTT) is split into distinct network phases:

1. DNS Resolution: Resolving domain names to IP addresses.
2. TCP Connection: Completing the 3-way handshake.
3. TLS Negotiation: Certificate exchange and encryption setup (for HTTPS).
4. Time to First Byte (TTFB): Server processing and initial response transmission.

To get precise metrics, your monitoring engine must measure each segment independently rather than relying on standard wall-clock execution time.

2. Asynchronous Inspection in Python

Using Python’s aiohttp combined with high-resolution system timers (time.perf_counter()), we can isolate network stages:

python
import asyncio
import time
import aiohttp

async def inspect_endpoint(url: str):
    metrics = {}

    t0 = time.perf_counter()
    async with aiohttp.ClientSession() as session:
        t_connect = time.perf_counter()
        async with session.get(url) as response:
            t_first_byte = time.perf_counter()
            content = await response.read()
            t_done = time.perf_counter()

            metrics['connection_setup'] = round((t_connect - t0) * 1000, 2)
            metrics['ttfb'] = round((t_first_byte - t_connect) * 1000, 2)
            metrics['download_time'] = round((t_done - t_first_byte) * 1000, 2)
            metrics['total_ms'] = round((t_done - t0) * 1000, 2)
            metrics['status'] = response.status

    return metrics

# Run standard check
if __name__ == '__main__':
    result = asyncio.run(inspect_endpoint('[https://api.github.com](https://api.github.com)'))
    print("Telemetry Data:", result)