QAAnywhere

Remote QA & testing jobs, worldwide.

Guides

Performance testing for QA engineers: JMeter, Gatling and k6 explained

By the QAAnywhere team · Published November 8, 2026

A skill that sets automation engineers apart

Performance testing answers a question functional tests never ask: does the system still work when many people use it at once? Few testers have done it properly, which is exactly why it stands out on a CV.

In remote QA listings it appears as an extra rather than a job of its own. On 8 November 2026, 9 of the 254 remote QA roles on QAAnywhere named a load testing tool: Gatling in 5, JMeter in 3, and k6 and LoadRunner in 2 each. All 9 also named Selenium, and 7 named Playwright. Only 4 roles had "performance" in the title. The message is clear: employers want automation engineers who can also design and run a meaningful load test, not dedicated performance specialists.

The kinds of performance test

Use the right words in interviews; they mean different things:

  • Load test. Expected peak traffic, for long enough to see stable behaviour. Answers "can we handle a normal busy day?"
  • Stress test. Push past expected load until something breaks. Answers "where is the limit, and how does it fail?"
  • Spike test. A sudden jump in traffic, such as a marketing email or a sale starting. Answers "do we recover, and how fast?"
  • Soak (endurance) test. Moderate load for hours. Finds memory leaks, connection pool exhaustion and slowly filling disks.
  • Capacity test. How many users or requests per second the system supports within its targets, used for planning.

The numbers that matter

A performance test produces a lot of numbers. These are the ones to report:

  • Response time percentiles. The 95th and 99th percentiles (p95, p99) tell you what slow users experience. Averages hide the problem: an average of 200 ms can sit on top of a p99 of 4 seconds.
  • Throughput. Requests or transactions per second the system completed.
  • Error rate. The share of requests that failed or timed out. A fast test with 5% errors is not a pass.
  • Resource use. CPU, memory, database connections and queue lengths on the system under test, which explain why the response times look the way they do.

Always agree targets before the test ("p95 under 500 ms and errors under 1% at 300 requests per second"). Without targets, a performance test is a set of interesting graphs, not a result.

JMeter: the long-standing standard

Apache JMeter has been the default open-source load tool for many years. Tests are built in a GUI as test plans (thread groups, samplers, assertions, listeners) and saved as XML .jmx files. It supports HTTP and many other protocols, and has a large plugin ecosystem.

Two practical rules: build and debug the plan in the GUI, but run real tests in non-GUI mode, which uses far less memory:

jmeter -n -t checkout.jmx -l results.jtl -e -o report/

JMeter's weakness is that XML test plans are hard to review in a pull request, which matters more in teams that treat tests as code.

Gatling: load tests as code on the JVM

Gatling defines scenarios in code, with DSLs for Java, Kotlin and Scala. It is efficient at generating high load from one machine and produces clear HTML reports. Because scenarios are code, they can be reviewed, versioned and reused like any other test. That fits teams already using Java tools such as Selenium, TestNG and REST Assured, which is the combination seen in the listings: the roles naming Gatling mostly also named those tools.

k6: JavaScript scripts with built-in pass/fail

k6 (from Grafana Labs) uses JavaScript for test scripts and is designed for developers and CI. Its thresholds turn a load test into a pass or fail result a pipeline can act on:

import http from 'k6/http';
import { check, sleep } from 'k6';

export const options = {
  vus: 50,
  duration: '5m',
  thresholds: {
    http_req_duration: ['p(95)<500'],
    http_req_failed: ['rate<0.01'],
  },
};

export default function () {
  const res = http.get('https://staging.example.test/api/products');
  check(res, { 'status is 200': (r) => r.status === 200 });
  sleep(1);
}

Fifty virtual users loop through the function for five minutes; the run fails if p95 response time goes above 500 ms or more than 1% of requests fail. k6 results are often sent to Grafana dashboards, which ties in with the monitoring many teams already use.

Which tool to learn

  • If your team writes Java tests, learn Gatling.
  • If your team writes JavaScript or TypeScript, or you want the easiest route to load tests in CI, learn k6.
  • If you are joining an enterprise with existing test plans, JMeter (or LoadRunner, the commercial veteran) is what you will meet.

The concepts transfer completely. An interviewer will care far more about how you designed a realistic test than which tool ran it.

Designing a realistic test

Most bad performance tests fail in design, not tooling:

  1. Model real behaviour. Use production analytics to choose the mix of requests (for example 70% browsing, 20% search, 10% checkout), with think time between steps.
  2. Use realistic data. A thousand users all searching for the same product only tests the cache.
  3. Test the right environment. A staging system with a tenth of production's resources gives results you cannot read directly. Know the differences and say so in your report.
  4. Warm up, then measure. Ramp up gradually and ignore the first minutes while caches and auto-scaling settle.
  5. Watch the system, not just the tool. Have the dashboards for the services and database open during the run.
  6. Do not load-test what you do not own. Exclude third-party services (payment providers, email) or stub them, and always get approval before generating heavy traffic against shared environments.

Reporting results to a remote team

Write the report so that someone who was not there can act on it: the goal and targets, the scenario and environment, a short table of the key numbers, the graphs that show the turning point, the bottleneck you found and your recommendation. One clear page beats a folder of exported charts.

Showing performance skills

  • Add a k6 or Gatling test to your public automation project, run it in CI with thresholds, and include a short written report.
  • On your CV, give a result, not just a tool: "Found a connection pool limit that capped checkout at 120 orders per minute; fixed before Black Friday."

The roles below are open now and name Gatling. Each listing shows where you can work from, so check that before applying.

Open remote QA roles that use Gatling

  1. Gatling Test Automation Engineer – API & UI IntegrationGdit · Any Location / Remote
  2. Gatling Test Automation Engineer - API & UI IntegrationGeneral Dynamics Information Technology · Remote
  3. Senior SDETYipitData · India Remote
  4. Senior QA Automation Engineer (Platform)Upgrade · United States (Remote)
  5. Senior QA Automation Engineer - HELOC (LATAM)Upgrade · Argentina (Remote)

More guides