The Rust Profiling Tweak That Cut Our CPU Usage by 73.9%

We thought our Rust service was already quick. It wasn’t — it was wasting CPU cycles like crazy.

The graphs told the story: CPU was climbing, traffic wasn’t. After profiling, we found the bottleneck, made a small change, and CPU usage dropped by 73.9%.

How We Found the Problem

This service handles HTTP requests, parses JSON payloads, and writes to PostgreSQL.

During a load test, CPU hit uncomfortable peaks. I suspected the database first — but I was wrong. We dropped in pprof and generated a flamegraph:

use pprof::protos::Message;
use std::{fs::File, io::Write};

fn main() -> Result<(), Box<dyn std::error::Error>> {
    let guard = pprof::ProfilerGuard::new(100)?; // 100 Hz sampling

    run_service()?; // Start your actual app here

    if let Ok(report) = guard.report().build() {
        let mut file = File::create("flamegraph.svg")?;
        report.flamegraph(&mut file)?;
    }

    Ok(())
}

The flamegraph didn’t lie — serde_json::from_str was chewing up CPU time.

The Root Cause

We were parsing JSON on every request, even if the payload was exactly the same as the previous one.