roadmap
- 01
- 02
- 03
- 04
- 05
- 06
- 07
- 08
Prepare for Grafana k6 interview questions on scripting load tests in JavaScript. Covers scenarios and executors, checks and thresholds, the metric model, protocols and running k6 in CI.
161 of 161 questions in your plan: 0 known, 0 in progress, 0 due, 0 skipped, 161 new. 0 cut, outside your plan.
Two kinds of loop ask about k6, and they want different things from it. In QA and SDET interviews it is one test tool among several: you are handed a user journey and asked to script it, capture the auth token a login returns and reuse it on every request after it, give each virtual user its own account, and make the run fail when the service misbehaves rather than merely print a row of failed checks. Expect a comparison too: why a team would pick k6 over [JMeter](/roadmaps/jmeter), and what a test kept as JavaScript in Git changes for whoever maintains it. Performance engineers, SREs and platform teams ask about the numbers instead. Their questions test whether you know what a result can be trusted for: whether a fixed pool of virtual users can hold a target request rate once the service slows down, why one p95 across every URL hides the slow endpoint, what a tag carrying a raw ID does to the time-series database the run streams into, and where in a pipeline a load test can gate a release without becoming the flaky step everyone reruns. When load testing is only one round of the job you are after, k6 also sits inside two wider plans that carry the other rounds: the one for a [Java SDET](/roadmaps/java-sdet) and the one for a [QA Engineer](/roadmaps/qa-engineer).
Answers about k6 tend to break in a predictable order, and the drill should follow it. The [k6 primer](/topics/dev-k6) on the hub page shows how the pieces of a run connect. It is the page to reopen once your answers hold up one at a time but start to contradict each other. A junior candidate is expected to write a script that behaves as intended, so [script anatomy](/topics/dev-k6-script) comes first: what runs once per virtual user at startup, what runs on every iteration, why a file can only be opened in the init stage, and how `setup` hands its data to the iterations. Close behind it is [the pass or fail verdict](/topics/dev-k6-verdict). A `check` records a result and lets the run carry on; a threshold turns a result into the exit code a CI job acts on, and confusing the two is a frequent junior slip. From middle level up, [executor choice](/topics/dev-k6-executors) is where candidates separate. Looping virtual users form a closed model, in which a slowing service throttles its own load and the latency looks better than it was; an arrival-rate executor keeps starting iterations on schedule and reports the ones it could not start. A senior candidate raises that distinction unprompted when asked to reproduce a traffic spike, then says which numbers in the result that choice makes trustworthy. When the assignment is a load script written at home and sent in, script anatomy and the verdict are what the reviewer reads for; in the days before it, narrowing importance to must know is the one change to the page worth making.
Two kinds of loop ask about k6, and they want different things from it. In QA and SDET interviews it is one test tool among several: you are handed a user journey and asked to script it, capture the auth token a login returns and reuse it on every request after it, give each virtual user its own account, and make the run fail when the service misbehaves rather than merely print a row of failed checks. Expect a comparison too: why a team would pick k6 over JMeter, and what a test kept as JavaScript in Git changes for whoever maintains it.
Performance engineers, SREs and platform teams ask about the numbers instead. Their questions test whether you know what a result can be trusted for: whether a fixed pool of virtual users can hold a target request rate once the service slows down, why one p95 across every URL hides the slow endpoint, what a tag carrying a raw ID does to the time-series database the run streams into, and where in a pipeline a load test can gate a release without becoming the flaky step everyone reruns.
When load testing is only one round of the job you are after, k6 also sits inside two wider plans that carry the other rounds: the one for a Java SDET and the one for a QA Engineer.
Answers about k6 tend to break in a predictable order, and the drill should follow it. The k6 primer on the hub page shows how the pieces of a run connect. It is the page to reopen once your answers hold up one at a time but start to contradict each other.
A junior candidate is expected to write a script that behaves as intended, so script anatomy comes first: what runs once per virtual user at startup, what runs on every iteration, why a file can only be opened in the init stage, and how setup hands its data to the iterations. Close behind it is the pass or fail verdict. A check records a result and lets the run carry on; a threshold turns a result into the exit code a CI job acts on, and confusing the two is a frequent junior slip.
From middle level up, executor choice is where candidates separate. Looping virtual users form a closed model, in which a slowing service throttles its own load and the latency looks better than it was; an arrival-rate executor keeps starting iterations on schedule and reports the ones it could not start. A senior candidate raises that distinction unprompted when asked to reproduce a traffic spike, then says which numbers in the result that choice makes trustworthy.
When the assignment is a load script written at home and sent in, script anatomy and the verdict are what the reviewer reads for; in the days before it, narrowing importance to must know is the one change to the page worth making.