A tiny Spring Boot app whose /api/quote/{symbol} endpoint calls a simulated slow dependency with limited capacity (8 workers x 50 ms = about 160 requests/s, queue of 100, then HTTP 503), plus Gatling Java DSL simulations that find the knee, and the same test written for k6.

Tool	Version	Source
Spring Boot	4.1.2	`maven-metadata.xml` (latest GA; 4.2.0-RC2 is not GA)
Gatling (`gatling-charts-highcharts`)	3.16.0	`maven-metadata.xml`; the report itself says released 2026-09-29
`gatling-maven-plugin`	4.21.12	`maven-metadata.xml`
k6	v2.3.0	`dl.k6.io` deb package, unpacked without installing
JDK	25

Run

./scripts/run-all.sh            # about 6 minutes; regenerates everything in output/
K6=/path/to/k6 ./scripts/run-all.sh   # include the k6 steps (skipped when k6 is not found)

Load numbers depend on the machine: treat them as indicative. Single runs: mvn test-compile gatling:test -Dgatling.simulationClass=lab.gatling.KneeStepsSimulation (start the app first: mvn package -DskipTests && java -jar target/gatling-spring-boot-lab-1.0.0.jar).

Files

src/main/java/lab/gatling/ the app: DownstreamPool (the bounded pool), QuoteController
src/test/java/lab/gatling/ simulations BasicSimulation, KneeStepsSimulation, KneeRampSimulation, ClosedModelSimulation, FailingAssertionSimulation, and SaturationTest (JUnit)
k6/knee.js, k6/threshold-fail.js the k6 equivalents
scripts/run-all.sh, scripts/analyze.py run everything, turn reports into tables and the chart
output/ captured real output of the last run