Performance / Load

Define each type and state when you would run it and what it catches.

Asked 'Load vs stress vs soak vs spike - define each and when you'd run it.' (asked almost every perf interview)

named in the job descriptions

Articulate that load testing is one type under the performance-testing umbrella (which also covers scalability, stability, resource use).

Asked 'How is load testing different from performance testing?'

asked in interviews, not named in any job description

Place performance inside the non-functional testing family (perf, reliability, scalability, resilience) and say what quality attribute each protects.

Asked 'What non-functional testing have you done?' / 'How do you treat performance as a quality attribute?'

named in the job descriptions

Report and reason about tail latency, explaining why percentiles beat the mean.

Asked 'Why report p95/p99 instead of average response time?' (the classic filter question)

asked in interviews, not named in any job description

Define throughput and read the RPS the system sustains before it plateaus.

Asked 'What metrics do you track in a perf test?' / 'What's a good throughput target and how do you set it?'

asked in interviews, not named in any job description

Track the fraction of failed requests and distinguish 'slow' from 'failing' as load rises.

Asked 'What metrics do you track?' / 'A test shows errors climbing at 500 VUs - what does that tell you?'

asked in interviews, not named in any job description

Name the server-side resource metrics you'd watch alongside client-side latency/throughput.

Asked 'What server metrics do you watch during a load test?'

named in the job descriptions

Install k6, write and run a .js/.ts load test locally, and explain the VU (virtual user) execution model.

Asked 'Walk me through how you'd load-test an API endpoint.' / 'Which tool have you used and why?'

named in the job descriptions

Recognize a JMeter test plan, its thread-group/sampler/listener structure, and know it's the legacy incumbent.

Asked 'Have you used JMeter?' / 'Our team is on JMeter - can you work with it?'

named in the job descriptions

Know Locust is the Python-native alternative and when a Python-first team would prefer it.

Asked 'Any experience with Locust?' / 'Why k6 over Locust?'

named in the job descriptions

Explain how each tool generates load and why VU != thread != request.

Asked 'How does your tool generate concurrency?' / 'How many users can you drive from one machine?'

asked in interviews, not named in any job description

Fire HTTP calls and assert on status and response time inside the script.

Asked part of 'walk me through a load-test script'

named in the job descriptions

Shape load over time with staged VU targets.

Asked 'How do you shape the load in a test?' / 'What's your ramp profile?'

asked in interviews, not named in any job description

Distinguish 'N concurrent users' (VU/ramping) from 'N requests/sec' (open-model arrival rate).

Asked 'Do you test by concurrent users or by request rate? When each?'

asked in interviews, not named in any job description

Feed distinct data per VU/iteration so you don't hammer one cached record.

Asked 'How do you avoid testing against a single cached record?' / 'How do you handle test data at load?'

named in the job descriptions

Insert realistic pauses and logically group requests so load models real users.

Asked 'How do you make load realistic?'

asked in interviews, not named in any job description

Encode pass/fail SLO criteria directly in the test so a breach = failed test.

Asked 'How do you set pass/fail on a perf test?' / 'What SLO would you set and why?'

asked in interviews, not named in any job description

Keep a baseline and compare each run to prove or catch a regression.

Asked 'How do you catch a perf regression before prod?'

named in the job descriptions

Wire the load test into CI so a threshold breach fails the build and results are published.

Asked 'How do you put a performance test in CI without making it flaky?' - THE differentiating answer

named in the job descriptions

Given a latency spike under load, reason through where the bottleneck lives (app vs DB vs network) and name the usual culprits.

Asked 'p95 latency spikes under load - how do you find the bottleneck?' / 'What causes performance bottlenecks?'

named in the job descriptions

Interpret the end-of-run summary to locate the tail, the throughput ceiling, and where failures begin.

Asked 'You ran the test - what do you look at in the results?'

asked in interviews, not named in any job description

Know the role APM plays: the load tool tells you WHAT slowed, APM tells you WHY.

Asked 'How do you correlate a load-test result with server behaviour?'

asked in interviews, not named in any job description

Know k6 natively load-tests gRPC and WebSocket and what changes for low-latency real-time services.

Asked 'Have you load-tested non-HTTP / real-time services?'

named in the job descriptions

Know that distributed/cloud load-gen exists and when you'd need it (beyond one machine / multi-region).

Asked 'How would you generate load beyond one machine?' (only in scale-heavy roles)

named in the job descriptions

Be current on what the modern tool offers, to pair perf with his 'modern stack / AI testing' narrative.

Asked 'Why did you pick k6?' / signalling you track the modern tooling

asked in interviews, not named in any job description

Load-test token-streaming LLM APIs and reason about metrics that differ from normal REST.

Asked 'How would you performance-test an AI/LLM feature?' - a talking point almost no competing SDET can give

asked in interviews, not named in any job description