Performance Test Engineer interview questions and answers

10 of 10 questions shown

1. juniorBasicsDefine load, stress, soak and spike tests.

   Model answer. Load: expected peak traffic. Stress: beyond capacity to find the breaking point and how the system recovers. Soak: sustained moderate load for hours to find leaks and degradation. Spike: a sudden surge, for example a marketing push, to see how fast the system scales and recovers.

   I can answer this
2. juniorMetricsWhy report percentiles instead of the average response time?

   Model answer. Averages hide the tail: a 200 ms mean can contain 5% of requests taking 3 seconds. Percentiles (p95, p99) show what slower users experience. Never average percentiles across instances; merge histograms instead.

   I can answer this
3. midWorkloadHow do you build a realistic workload model?

   Model answer. Start from production data: peak requests per second per endpoint, the mix of user journeys, think times and data variety. Derive concurrency from Little's Law, N = throughput times time in system, and model ramp patterns that match real traffic rather than a flat wall of users.

   I can answer this
4. midCalculationEach iteration takes 2 seconds including think time and you need 100 iterations per second. How many virtual users?

   Model answer. Little's Law: N = 100 per second times 2 seconds = 200 concurrent virtual users. If the system slows down and iterations take 4 seconds, a closed model with 200 users only delivers 50 per second, which is why arrival-rate models matter.

   Tools: k6

   I can answer this
5. midMethodWhat is coordinated omission and why does it matter?

   Model answer. A closed-model load generator waits for each response before sending the next request. When the server stalls, the tool stops sending and records few slow samples, so latency looks better than it was. Use open-model executors (k6 constant-arrival-rate, Gatling open injection) or correct the histogram.

   Tools: k6, Gatling

   I can answer this
6. midJMeterWhich JMeter mistakes invalidate results?

   Model answer. Running load in GUI mode, leaving View Results Tree listeners on, not correlating dynamic values such as tokens, saturating a single load generator and ignoring ramp-up. Run in CLI (non-GUI) mode, use lightweight reporting, check generator CPU and distribute load when needed.

   Tools: JMeter

   I can answer this
7. midDiagnosisResponse times climb sharply at 400 users while app-server CPU is at 30%. What do you check?

   Model answer. The system is not CPU-bound, so look for queuing: database connection pool exhaustion or lock contention, thread pool limits, a slow downstream dependency, GC pauses and load balancer limits. Use APM traces and pool metrics, and confirm the load generator itself is not the bottleneck.

   I can answer this
8. seniorPipelineWhere do performance tests belong in a delivery pipeline?

   Model answer. A short smoke test with thresholds on every pull request, full load tests nightly on a production-like environment, capacity and soak tests before major releases, and production SLO monitoring as the final truth. Store baselines so trends, not single runs, drive decisions.

   Tools: k6, JMeter

   I can answer this
9. seniorCapacityYour test environment is a quarter of production. How do you extrapolate results?

   Model answer. Do not assume linear scaling. Measure per-instance throughput, then validate with 1, 2 and 4 nodes to find where scaling bends, use a production-sized data set for the database, and state the confidence level. Confirm with a limited production canary where possible.

   I can answer this
10. seniorCachingHow do you make results meaningful when caches are involved?

    Model answer. Test cold and warm states separately and use a realistic key distribution instead of the same IDs repeatedly. Measure the hit ratio, and include a flush scenario to expose stampedes when many requests miss at once.

    I can answer this

No question matches those filters. Clear the search or choose All.