testPodSandboxCreation: skip sandbox errors for pods which were not d…

…eleted during network update

"pods should successfully create sandboxes" test should mark pod events as flakes if network is being updated. These pods eventually get created, so the test is too strict

pkg/synthetictests/networking.go

@@ -5,6 +5,7 @@ import (
 	"strings"
 	"time"
 
+	"github.com/openshift/origin/pkg/monitor/intervalcreation"
 	"github.com/openshift/origin/pkg/monitor/monitorapi"
 
 	"github.com/openshift/origin/pkg/test/ginkgo"
@@ -44,16 +45,30 @@ func testPodSandboxCreation(events monitorapi.Intervals) []*ginkgo.JUnitTestCase
 		}
 		deletionTime := getPodDeletionTime(eventsForPods[event.Locator], event.Locator)
 		if deletionTime == nil {
+			// mark sandboxes errors as flakes if networking is being updated
+			// these pods eventually get created
+			operatorsProgressing := intervalcreation.IntervalsFromEvents_OperatorProgressing(events, event.From, event.To)
+			networkProgressing := false
+			for _, ev := range operatorsProgressing {
+				if strings.Contains(ev.Message, "clusteroperator/network") || strings.Contains(ev.Message, "clusteroperator/machine-config") {
+					networkProgressing = true
+					break
+				}
+			}
 			// this indicates a failure to create the sandbox that should not happen
-			failures = append(failures, fmt.Sprintf("%v - never deleted - %v", event.Locator, event.Message))
+			if networkProgressing {
+				flakes = append(flakes, fmt.Sprintf("%v - never deleted - network rollout - %v", event.Locator, event.Message))
+			} else {
+				failures = append(failures, fmt.Sprintf("%v - never deleted - %v", event.Locator, event.Message))
+			}
 		} else {
 			timeBetweenDeleteAndFailure := event.From.Sub(*deletionTime)
 			switch {
 			case timeBetweenDeleteAndFailure < 1*time.Second:
 				// nothing here, one second is close enough to be ok, the kubelet and CNI just didn't know
 			case timeBetweenDeleteAndFailure < 5*time.Second:
 				// withing five seconds, it ought to be long enough to know, but it's close enough to flake and not fail
-				flakes = append(failures, fmt.Sprintf("%v - %0.2f seconds after deletion - %v", event.Locator, timeBetweenDeleteAndFailure.Seconds(), event.Message))
+				flakes = append(flakes, fmt.Sprintf("%v - %0.2f seconds after deletion - %v", event.Locator, timeBetweenDeleteAndFailure.Seconds(), event.Message))
 			case deletionTime.Before(event.From):
 				// something went wrong.  More than five seconds after the pod ws deleted, the CNI is trying to set up pod sandboxes and can't
 				failures = append(failures, fmt.Sprintf("%v - %0.2f seconds after deletion - %v", event.Locator, timeBetweenDeleteAndFailure.Seconds(), event.Message))