From Bash to CLI: When a Quick Benchmark Gets Serious

Green Algeria Map’s benchmark pipeline started as a simple comparison between two backends. A bash script, a k6 file, some results.

Then I added Go as a baseline to see how NestJS and Spring Boot compared. Then sequential runs, shared Postgres, a compare script. The scripts grew because the benchmark kept growing.

The First Script

The original run.sh took a backend name and ran k6 against it:

#!/usr/bin/env bash
set -euo pipefail

BACKEND="${1:-nestjs}"
if [ "$BACKEND" != "nestjs" ] && [ "$BACKEND" != "springboot" ]; then
  echo "Usage: $0 {nestjs|springboot}"
  exit 1
fi

if [ "$BACKEND" = "nestjs" ]; then
  BASE_URL="http://localhost:8080"
else
  BASE_URL="http://localhost:8081"
fi

OUTDIR="results/$(date +%Y%m%d-%H%M)-$BACKEND"
mkdir -p "$OUTDIR"

for SCENARIO in auth zones mix; do
  k6 run \
    --out json="$OUTDIR/$SCENARIO.json" \
    --summary-export="$OUTDIR/$SCENARIO-summary.json" \
    -e BASE_URL="$BASE_URL" \
    "benchmark/$SCENARIO.js"
done

Positional args, hardcoded ports, a for loop. It worked.

The Pipeline Script

The pipeline.sh had to orchestrate everything: start Docker, run migrations, seed data, wait for health checks, run benchmarks, clean up.

wait_for() {
  local url="$1" label="$2" max=60
  for i in $(seq 1 $max); do
    if curl -sf "$url" >/dev/null 2>&1; then
      return
    fi
    sleep 2
  done
  exit 1
}

run_nestjs() {
  docker compose --profile nestjs up -d postgres rustfs
  wait_for "http://localhost:5432" "PostgreSQL"
  cd backend-nestjs
  node scripts/create-bucket.mjs
  pnpm migration:run
  pnpm seed
  cd ..
  docker compose --profile nestjs up -d nestjs-app
  wait_for "http://localhost:8080/api/health/live" "NestJS"
  ./benchmark/run.sh nestjs
  docker compose --profile nestjs down -v
}

Two functions (run_nestjs, run_springboot), same pattern, different ports. No error recovery.

The Compare Script

The compare script used inline Python to dig into k6’s JSON output:

nest_avg=$(python3 -c "
import json
d = json.load(open('$nest_json'))
print(d['metrics']['http_req_duration']['avg'])
")

Four of these per backend per scenario, wrapped in a function. A winner-takes-all ranking built from bash + Python strings.

What the Growth Looked Like

The pipeline script had three backends, each with its own copy of the same orchestration logic. A compare script that ranked results using inline Python.

The rewrite came when the benchmark outgrew the script format entirely.

The Breaking Point

The Go backend was the third backend. The script kept growing. Multiple backends, shared Postgres, sequential runs, a compare script.