should propagation time include action

[lahabana]

At the moment we wait for pods to be available (or policy to be applied) before computing the propagation time.

mesh-perf/test/k8s/simple_test.go

Lines 214 to 230 in 286a0cd

	err := k8s.RunKubectlE(
	cluster.GetTesting(),
	cluster.GetKubectlOptions(TestNamespace),
	"scale", "statefulset", observable, fmt.Sprintf("--replicas=%d", replicas),
	)
	Expect(err).ToNot(HaveOccurred())
	err = cluster.Install(WaitNumPods(TestNamespace, replicas, observable))
	Expect(err).ToNot(HaveOccurred())
	propagationStart := time.Now()
	Eventually(func(g Gomega) {
	membership, err := admin.GetStats(fmt.Sprintf("cluster.%s_kuma-test_svc_80.membership_total", observable))
	g.Expect(err).ToNot(HaveOccurred())
	g.Expect(membership.Stats[0].Value).To(BeNumerically("==", replicas))
	}, "60s", "1s").Should(Succeed())
	AddReportEntry("endpoint_propagation_duration", time.Since(propagationStart).Milliseconds())

Should we include that time?
If we do we'll be sensitive of the time it takes to start a pod so platform dependent. However, this is a stat that's more useful to users than what we have today.

Another option is to watch for change to endpointSlice and count from there. This would sum up as "how much longer does it take for an endpoint to be added compared to core k8s" though that wouldn't even be exact...

[lobkovilya]

Triage: regardless of the approach we pick using WaitNumPods is not accurate enough as it's using a loop with sleeps. Probably we should watch an endpointSlice and start the timer when the endpoint is added.

[github-actions]

This issue was inactive for 90 days. It will be reviewed in the next triage meeting and might be closed.
If you think this issue is still relevant, please comment on it or attend the next triage meeting.