service_creation_latency: limit number of calls, exit on success

[cezarygerard]

/kind cleanup

[aojea]

/assign @aojea

[aojea]

I don't think that is necessary wrong to assert 10 times, maybe is indeed excessive , but the main problem here is that this doing an exec per command.

Another flaw in this code is that if no ipsare obtained , this will consider it as a success.

it seems the code in updateObject waits until an ingress.IP is created

Oh

[aojea]

oh, so this never returned before 🤦

[aojea]

oh, it does return, see lines 354-356

https://github.com/kubernetes/perf-tests/pull/2314/files#diff-89087237f4615aee550b7d157c898b60525acb6063585068ce42884480131616R354-R356

			if _, exists := p.creationTimes.Get(key, phaseName(reachabilityPhase, p.svc.Spec.Type)); exists {
				return
			}

[cezarygerard]

we should get rid for this lines

for the simple return here:

			if success == pingChecks {
				p.creationTimes.Set(key, phaseName(reachabilityPhase, p.svc.Spec.Type), time.Now())
				return
                         }

[aojea]

agree, is confusing to do the check at the beginning if we are explicitily setting it here ... unless there are multiple of these running in parlallel? do you know if that is even possible? the same service gets called several times?

[aojea]

please see suggestion

diff --git a/clusterloader2/pkg/measurement/common/service_creation_latency.go b/clusterloader2/pkg/measurement/common/service_creation_latency.go
index 6b69dedfb..8176851e0 100644
--- a/clusterloader2/pkg/measurement/common/service_creation_latency.go
+++ b/clusterloader2/pkg/measurement/common/service_creation_latency.go
@@ -345,7 +345,6 @@ func (p *pingChecker) run() {
                klog.Errorf("%s: meta key created error: %v", p.callerName, err)
                return
        }
-       success := 0
        for {
                select {
                case <-p.stopCh:
@@ -358,7 +357,6 @@ func (p *pingChecker) run() {
                        pod, err := execservice.GetPod()
                        if err != nil {
                                klog.Warningf("call to execservice.GetPod() ended with error: %v", err)
-                               success = 0
                                time.Sleep(pingBackoff)
                                continue
                        }
@@ -377,23 +375,23 @@ func (p *pingChecker) run() {
                                }
                                port = p.svc.Spec.Ports[0].Port
                        }
+                       if len(ips) == 0 {
+                               klog.Warningf("no ips found for service: %+v", p.svc)
+                               time.Sleep(pingBackoff)
+                               continue
+                       }
                        for _, ip := range ips {
-                               address := net.JoinHostPort(ip, fmt.Sprint(port))
-                               command := fmt.Sprintf("curl %s", address)
+                               command := fmt.Sprintf(`date; for i in $(seq 1 %d); do echo "$(date) Try: ${i}"; curl -g -q -s --max-time 15 --connect-timeout 1 %s; echo; done`, pingChecks, net.JoinHostPort(ip, fmt.Sprint(port)))
                                _, err = execservice.RunCommand(context.TODO(), pod, command)
                                if err != nil {
                                        break
                                }
                        }
                        if err != nil {
-                               success = 0
                                time.Sleep(pingBackoff)
                                continue
                        }
-                       success++
-                       if success == pingChecks {
-                               p.creationTimes.Set(key, phaseName(reachabilityPhase, p.svc.Spec.Type), time.Now())
-                       }
+                       p.creationTimes.Set(key, phaseName(reachabilityPhase, p.svc.Spec.Type), time.Now())
                }
        }
 }

[aojea]

/assign @wojtek-t

doing 3 checks instead of 10 does not sounds bad, specially if each check is a new kubectl exec call, that is very expensive, an alternative is to do the multiple checks in one call to the pod as in #2314 (comment)

The logic to return from the forever loop checks for the condition to exit at the beginning of the loop, I do not know if this plugins may execute the same service in parallel and this is needed to avoid overlapping #2314 (comment)

[wojtek-t]

/lgtm
/approve

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: cezarygerard, wojtek-t

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Needs approval from an approver in each of these files:

• clusterloader2/OWNERS [wojtek-t]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment