ClusterOperators should not go Progressing only for a node reboot

This is to cover the node rebooting case from the rule [1] that
is introduced recently:

```
Operators should not report Progressing only because DaemonSets
owned by them are adjusting to a new node from cluster scaleup or
a node rebooting from cluster upgrade.
```

The test fails if

- `co/machine-config` never became Progressing=True during a
  cluster upgrade, or
- some CO left Progressing=False during the upgrade after
  `machine-config` became Progressing=True. This should not
  have taken place as `machine-config` was rebooting the nodes
  which was the only thing ongoing to the cluster during that
  time.

[1]. https://github.com/openshift/api/blob/61248d910ff74aef020492922d14e6dadaba598b/config/v1/types_cluster_operator.go#L163-L164

Author: hongkailiu

pkg/monitortests/clusterversionoperator/legacycvomonitortests/monitortest.go

@@ -45,6 +45,7 @@ func (w *legacyMonitorTests) EvaluateTestsFromConstructedIntervals(ctx context.C
 	isUpgrade := platformidentification.DidUpgradeHappenDuringCollection(finalIntervals, time.Time{}, time.Time{})
 	if isUpgrade {
 		junits = append(junits, testUpgradeOperatorStateTransitions(finalIntervals, w.adminRESTConfig)...)
+		junits = append(junits, clusterOperatorIsNotProgressingWhenMachineConfigIs(finalIntervals)...)
 	} else {
 		junits = append(junits, testStableSystemOperatorStateTransitions(finalIntervals, w.adminRESTConfig)...)
 	}

pkg/monitortests/clusterversionoperator/legacycvomonitortests/operators.go

@@ -6,21 +6,21 @@ import (
 	"strings"
 	"time"
 
-	"github.com/openshift/origin/pkg/monitortestlibrary/utility"
+	configv1 "github.com/openshift/api/config/v1"
+	clientconfigv1 "github.com/openshift/client-go/config/clientset/versioned/typed/config/v1"
+	"github.com/sirupsen/logrus"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/apimachinery/pkg/util/sets"
 	"k8s.io/client-go/kubernetes"
+	"k8s.io/client-go/rest"
 
-	"github.com/openshift/origin/pkg/monitortests/clusterversionoperator/operatorstateanalyzer"
-	"github.com/sirupsen/logrus"
-
-	configv1 "github.com/openshift/api/config/v1"
-	clientconfigv1 "github.com/openshift/client-go/config/clientset/versioned/typed/config/v1"
 	"github.com/openshift/origin/pkg/monitor/monitorapi"
 	"github.com/openshift/origin/pkg/monitortestlibrary/platformidentification"
 	platformidentification2 "github.com/openshift/origin/pkg/monitortestlibrary/platformidentification"
+	"github.com/openshift/origin/pkg/monitortestlibrary/utility"
+	"github.com/openshift/origin/pkg/monitortests/clusterversionoperator/operatorstateanalyzer"
 	"github.com/openshift/origin/pkg/test/ginkgo/junitapi"
 	exutil "github.com/openshift/origin/test/extended/util"
-	"k8s.io/client-go/rest"
 )
 
 // exceptionCallback consumes a suspicious condition and returns an
@@ -599,6 +599,134 @@ func testOperatorStateTransitions(events monitorapi.Intervals, conditionTypes []
 	return ret
 }
 
+func clusterOperatorIsNotProgressingWhenMachineConfigIs(events monitorapi.Intervals) []*junitapi.JUnitTestCase {
+	var ret []*junitapi.JUnitTestCase
+	upgradeWindows := getUpgradeWindows(events)
+
+	var machineConfigProgressing time.Time
+	var eventsInUpgradeWindows monitorapi.Intervals
+
+	var start, stop time.Time
+	for _, event := range events {
+		if !isInUpgradeWindow(upgradeWindows, event) {
+			continue
+		}
+		eventsInUpgradeWindows = append(eventsInUpgradeWindows, event)
+		if start.IsZero() || event.From.Before(start) {
+			start = event.From
+		}
+		if stop.IsZero() || event.To.After(stop) {
+			stop = event.To
+		}
+	}
+	duration := stop.Sub(start).Seconds()
+
+	eventsByOperator := getEventsByOperator(eventsInUpgradeWindows)
+	for _, mcEvent := range eventsByOperator["machine-config"] {
+		condition := monitorapi.GetOperatorConditionStatus(mcEvent)
+		if condition == nil {
+			continue // ignore non-condition intervals
+		}
+		if condition.Type == configv1.OperatorProgressing && condition.Status == configv1.ConditionTrue {
+			machineConfigProgressing = mcEvent.To
+		}
+	}
+
+	mcTestCase := &junitapi.JUnitTestCase{
+		Name:     fmt.Sprintf("[bz-Machine Config Operator] clusteroperator/machine-config must go Progressing=True during an upgrade test"),
+		Duration: duration,
+	}
+	if machineConfigProgressing.IsZero() {
+		mcTestCase.FailureOutput = &junitapi.FailureOutput{
+			Output: fmt.Sprintf("machine-config was never Progressing=True during the upgrade window from %s to %s", start.Format(time.RFC3339), stop.Format(time.RFC3339)),
+		}
+		return []*junitapi.JUnitTestCase{mcTestCase}
+	} else {
+		mcTestCase.SystemOut = fmt.Sprintf("machine-config became Progressing=True at %s during the upgrade window from %s to %s", machineConfigProgressing.Format(time.RFC3339), start.Format(time.RFC3339), stop.Format(time.RFC3339))
+	}
+	ret = append(ret, mcTestCase)
+
+	for _, operatorName := range platformidentification.KnownOperators.Difference(sets.NewString("machine-config")).List() {
+		bzComponent := platformidentification.GetBugzillaComponentForOperator(operatorName)
+		if bzComponent == "Unknown" {
+			bzComponent = operatorName
+		}
+		testName := fmt.Sprintf("[bz-%v] clusteroperator/%v should stay Progressing=False while MCO is Progressing=True", bzComponent, operatorName)
+		operatorEvents := eventsByOperator[operatorName]
+		if len(operatorEvents) == 0 {
+			ret = append(ret, &junitapi.JUnitTestCase{
+				Name:     testName,
+				Duration: duration,
+			})
+			continue
+		}
+
+		except := func(co string, condition *configv1.ClusterOperatorStatusCondition) string {
+			return ""
+		}
+
+		var excepted, fatal []string
+		for _, operatorEvent := range operatorEvents {
+			if operatorEvent.From.Before(machineConfigProgressing) {
+				continue
+			}
+			condition := monitorapi.GetOperatorConditionStatus(operatorEvent)
+			if condition == nil {
+				continue // ignore non-condition intervals
+			}
+			if condition.Type == "" {
+				fatal = append(fatal, fmt.Sprintf("failed to convert %v into a condition with a type", operatorEvent))
+				continue
+			}
+
+			if condition.Type != configv1.OperatorProgressing || condition.Status == configv1.ConditionFalse {
+				continue
+			}
+
+			// if there was any switch, it was wrong/unexpected at some point
+			failure := fmt.Sprintf("%v", operatorEvent)
+
+			exception := except(operatorName, condition)
+			if exception == "" {
+				fatal = append(fatal, failure)
+			} else {
+				excepted = append(excepted, fmt.Sprintf("%s (exception: %s)", failure, exception))
+			}
+		}
+
+		output := fmt.Sprintf("%d unexpected clusteroperator state transitions while machine-config is progressing during the upgrade window from %s to %s", len(fatal), start.Format(time.RFC3339), stop.Format(time.RFC3339))
+		if len(fatal) > 0 {
+			output = fmt.Sprintf("%s.  These did not match any known exceptions, so they cause this test-case to fail:\n\n%v\n", output, strings.Join(fatal, "\n"))
+		} else {
+			output = fmt.Sprintf("%s, as desired.", output)
+		}
+		output = fmt.Sprintf("%s\n%d unwelcome but acceptable clusteroperator state transitions while machine-config is progressing during the upgrade window from %s to %s", output, len(excepted), start.Format(time.RFC3339), stop.Format(time.RFC3339))
+		if len(excepted) > 0 {
+			output = fmt.Sprintf("%s.  These should not happen, but because they are tied to exceptions, the fact that they did happen is not sufficient to cause this test-case to fail:\n\n%v\n", output, strings.Join(excepted, "\n"))
+		} else {
+			output = fmt.Sprintf("%s, as desired.", output)
+		}
+
+		if len(fatal) > 0 || len(excepted) > 0 {
+			ret = append(ret, &junitapi.JUnitTestCase{
+				Name:      testName,
+				Duration:  duration,
+				SystemOut: output,
+				FailureOutput: &junitapi.FailureOutput{
+					Output: output,
+				},
+			})
+		}
+
+		if len(fatal) == 0 {
+			// add a success so we flake (or pass) and don't fail
+			ret = append(ret, &junitapi.JUnitTestCase{Name: testName, Duration: duration, SystemOut: output})
+		}
+	}
+
+	return ret
+}
+
 type startedStaged struct {
 	// OSUpdateStarted is the event Reason emitted by the machine config operator when a node begins extracting
 	// it's OS content.