mirror of https://github.com/prometheus/prometheus
Browse Source
resolvePodRef is in a hot path: ``` ROUTINE ======================== github.com/prometheus/prometheus/discovery/kubernetes.(*Endpoints).resolvePodRef in discovery/kubernetes/endpoints.go 2.50TB 2.66TB (flat, cum) 22.28% of Total . . 447:func (e *Endpoints) resolvePodRef(ref *apiv1.ObjectReference) *apiv1.Pod { . . 448: if ref == nil || ref.Kind != "Pod" { . . 449: return nil . . 450: } 2.50TB 2.50TB 451: p := &apiv1.Pod{} . . 452: p.Namespace = ref.Namespace . . 453: p.Name = ref.Name . . 454: . 156.31GB 455: obj, exists, err := e.podStore.Get(p) . . 456: if err != nil { . . 457: level.Error(e.logger).Log("msg", "resolving pod ref failed", "err", err) . . 458: return nil . . 459: } . . 460: if !exists { ``` This is some low hanging fruit that we can easily optimize. The key of an object has format "namespace/name" so generate that inside of Prometheus itself and use pooling. ``` goos: linux goarch: amd64 pkg: github.com/prometheus/prometheus/discovery/kubernetes cpu: Intel(R) Core(TM) i9-10885H CPU @ 2.40GHz │ olddisc │ newdisc │ │ sec/op │ sec/op vs base │ ResolvePodRef-16 516.3n ± 17% 289.5n ± 7% -43.92% (p=0.000 n=10) │ olddisc │ newdisc │ │ B/op │ B/op vs base │ ResolvePodRef-16 1168.00 ± 0% 24.00 ± 0% -97.95% (p=0.000 n=10) │ olddisc │ newdisc │ │ allocs/op │ allocs/op vs base │ ResolvePodRef-16 2.000 ± 0% 2.000 ± 0% ~ (p=1.000 n=10) ¹ ¹ all samples are equal ``` Signed-off-by: Giedrius Statkevičius <giedrius.statkevicius@vinted.com>pull/15230/head
Giedrius Statkevičius
4 weeks ago
2 changed files with 52 additions and 1 deletions
Loading…
Reference in new issue