desc:=fmt.Sprintf("latency/resource should be within limit when create %d pods with %v interval [Benchmark]",itArg.podsNr,itArg.interval)
desc:=fmt.Sprintf("latency/resource should be within limit when create %d pods with %v interval [Benchmark][NodeSpecialFeature:Benchmark]",itArg.podsNr,itArg.interval)
desc:=fmt.Sprintf("latency/resource should be within limit when create %d pods with %v interval (QPS %d) [Benchmark]",itArg.podsNr,itArg.interval,itArg.APIQPSLimit)
desc:=fmt.Sprintf("latency/resource should be within limit when create %d pods with %v interval (QPS %d) [Benchmark][NodeSpecialFeature:Benchmark]",itArg.podsNr,itArg.interval,itArg.APIQPSLimit)
// The latency caused by API QPS limit takes a large portion (up to ~33%) of e2e latency.
// It makes the pod startup latency of Kubelet (creation throughput as well) under-estimated.
// Here we set API QPS limit from default 5 to 60 in order to test real Kubelet performance.
desc:=fmt.Sprintf("latency/resource should be within limit when create %d pods with %d background pods [Benchmark]",itArg.podsNr,itArg.bgPodsNr)
desc:=fmt.Sprintf("latency/resource should be within limit when create %d pods with %d background pods [Benchmark][NodeSpeicalFeature:Benchmark]",itArg.podsNr,itArg.bgPodsNr)