04/18/2023, 6:35 PM
Hi, guys! We are going to use SpiceDB in a production self-hosted environment (k8s on GKE), but before we decided to benchmark. We did a benchmark on 2000 QPS on the Check gRPC method. We have two pods. After a few benchmarking attempts we've discovered, that only one node usually serves all the requests (see image). We've tried to use gRPC
strategy to distribute traffic between two pods. We tested with the same request body, that's why we usually end-up with loaded only one node, but we don't understand why these "waves" happens. When traffic stops on one node and starts on the second. Thanks a lot for any advice!