I have tried this library with default configuration and identical k6 load test with a single api endpoint. Without load shedding: 100% responses are OK (200). With load shedding: 90% responses are OK (200); 10% responses are 503. Total number of requests is the same in both cases. So using this library I have reduced the number of RPS.
Hi, could you open an issue in the LoadShedding repository and provide more information? With that, we can understand what may be happening in this case. Thanks
Very good concept but without a configurable cpu/memory threshold for load shedding it will affect the auto-scaling that APIs have. There's a potential concern that load shedding could mask the true demand on your application from the auto-scaling mechanisms. If load shedding frequently drops or delays requests, auto-scaling might not trigger as expected because the perceived load is reduced.