Configure Load Testing Infrastructure with Locust

version: "3.8"
services:
  master:
    image: locustio/locust:2.24.0
    ports:
      - "8089:8089"
    volumes:
      - ./:/mnt/locust
    command: -f /mnt/locust/locustfile.py --master
  worker:
    image: locustio/locust:2.24.0
    volumes:
      - ./:/mnt/locust
    command: -f /mnt/locust/locustfile.py --worker --master-host master
    deploy:
      replicas: 4
    depends_on:
      - master

deploy.replicas: 4 — four workers, each on a separate CPU core. The number of workers can be increased up to 50 or more. For Kubernetes load, Locust can be deployed on a cluster using Helm charts for massive scalability.

How to integrate Locust into CI and analyze metrics?

Headless mode with assertions via --exit-code-on-error and CSV parsing:

---
- name: Run Locust
  run: |
    locust -f locfile.py --headless \
      -u 500 -r 50 --run-time 5m \
      --csv=results --html=report.html \
      --exit-code-on-error 1
- name: Check p95 latency
  run: |
    p95=$(awk -F',' 'NR>1 && $1=="Aggregated" {print $14}' results_stats.csv)
    if (( $(echo "$p95 > 500" | bc -l) )); then
      echo "p95 latency ${p95}ms exceeds 500ms threshold"
      exit 1
    fi

Locust exports results_stats.csv with fields: request_count, failure_count, median_response_time, average_response_time, p50, p95, p99, and other percentiles. Average response time of 250ms at 2000 requests per second is a typical result for a well-tuned API. Our service includes a DevOps audit to optimize load testing pipelines.

Typical issues when setting up

• GIL and CPU limitations. Locust uses gevent for coroutines — this bypasses GIL for I/O, but CPU-bound operations (heavy response processing) block. If the test has many json.loads() on large payloads (e.g., 500KB), capacity is eaten up. Solution: offload processing to separate threads or profile.
• Realistic wait time. between(0, 0.1) is not realistic; it's a DDoS. between(1, 5) with wait_time is closer to a real user.
• Data management. For unique users, use LoadTestShape or a CSV feeder via itertools.cycle. Repeated login under one account from thousands of workers breaks sessions and gives false results.

What's included in turnkey load testing setup

1. Audit and design – analyze test scenarios and infrastructure.
2. Script development – write Python scenarios reflecting business logic.
3. Distributed execution – set up Docker Compose or Kubernetes.
4. CI/CD integration – plug into GitHub Actions or GitLab CI.
5. Metrics visualization – optional Grafana + InfluxDB.
6. Documentation and training – reproduce tests and train your team.

Our expertise in load testing

We have been doing load testing for over 5 years, completing 50+ projects for e-commerce, fintech, and SaaS. Our engineers are AWS and Kubernetes certified, ensuring correct infrastructure operation even under high load. Time savings on debugging scenarios — up to 50% compared to self-setup. 90% of our clients continue using Locust after the initial setup. Clients note a 30% reduction in testing infrastructure cost after implementing distributed Locust. Average response time improvements of 40% after tuning.

Get a consultation on Locust setup for your project — we'll assess your infrastructure and propose the optimal solution. Order an audit of your current load testing scenarios to ensure your tests produce relevant results and maximize API performance.