I Built a Python Library to Measure Code's Carbon Footprint — Here's What I Learned

#python #sustainability #opensource #machinelearning

You've probably never asked this question while writing code:

"How much CO₂ did this function just emit?"

I hadn't either — until I started building EcoTrace.

The Problem With Existing Tools

CodeCarbon is great for getting started.
But after using it in production, I kept hitting the same walls:

No async support — modern Python is async, tracking shouldn't be sync-only
System-wide CPU measurement — in a multi-process environment, this is noise
No AMD/Intel GPU support — NVIDIA-only in 2025 is a real limitation
No side-by-side comparison — I wanted to know: is v1 or v2 of my function greener?

So I built EcoTrace.

How It Works in 5 Lines

from ecotrace import EcoTrace

eco = EcoTrace(region_code="TR")

@eco.track
def train_model():
    # your training code
    pass

train_model()
# [EcoTrace] Duration   : 12.4s
# [EcoTrace] Avg CPU    : 87.3%
# [EcoTrace] CO2 Emitted: 0.00312 gCO2

The Feature That Surprised Me Most

Process-scoped CPU measurement. Most tools measure system-wide CPU —
which means if you're running on a shared server, you're measuring
everyone's compute, not yours.

EcoTrace uses psutil.Process() to track only your process:

# This measures YOUR code, not the entire machine
self._current_process = psutil.Process()

In a Kubernetes pod running 8 services, this difference is enormous.

Async Support — Finally

@eco.track
async def fetch_and_process():
    data = await fetch_from_api()
    return process(data)

Same decorator. Works on sync and async automatically.
EcoTrace detects coroutines at decoration time and routes accordingly.

Comparing Two Implementations

This is my favorite feature:

def bert_inference():
    return model_bert.predict(text)

def distilbert_inference():
    return model_distilbert.predict(text)

result = eco.compare(bert_inference, distilbert_inference)
# BERT:      0.00042 gCO2 | 1.24s
# DistilBERT: 0.00018 gCO2 | 0.51s  ← 57% less carbon