k6-tests

K6 load testing via GitHub Actions. Workflow runs a k6 tests, and the test results will be available on external Grafana or as an HTML \ TXT artifact of Workflow.

The ramping-arrival-rate was chosen as the executor for testing since the weather application API most often receives single requests from users. Users usually visit weather forecast services to see the forecast for a specific city. For the stress test limit set to 1000rp/s, k6 adjusts the number of users automatically. The test also includes 1rp/s PUT to simulate weather updates.

Normal load test is limited to 500rp/s

---

SLO/SLA:

• request latency p(95) < 250ms
• failed http requests < 1%

---

Tests:

• Load test for normal load conditions:

  • 

• Stress test for simulate above average load:

  • 

• Breakpoint test for finding maximum load to handle SLO/SLA:

  • always passing 😃

    • System's found maximums after several tests (only read):

      VUS	RPS
      345	1.9k
      426	1.6k

Conclusion

The bottleneck seems to be the in the k8 infrastructure pod CPU. After running the breakpoint test, the degradation of latency is noted due to k8s pods bumping to internal CPU limits. At the same time, the database cluster is not peaking to the CPU and I/O limits.

Graphs and stats