Merge pull request #28402 from wking/complete-update-for-abort-at-100

NO-JIRA: test/e2e/upgrade/monitor: --abort-at=100 requires a complete update

Author: openshift-merge-bot[bot]

pkg/cmd/openshift-tests/run-upgrade/command.go

@@ -29,7 +29,8 @@ func NewRunUpgradeCommand(streams genericclioptions.IOStreams) *cobra.Command {
 		Supported options:
 
 		* abort-at=NUMBER - Set to a number between 0 and 100 to control the percent of operators
-		at which to stop the current upgrade and roll back to the current version.
+		at which to stop the current upgrade and roll back to the current version (100 to require a
+		complete update).
 		* disrupt-reboot=POLICY - During upgrades, periodically reboot master nodes. If set to 'graceful'
 		the reboot will allow the node to shut down services in an orderly fashion. If set to 'force' the
 		machine will terminate immediately without clean shutdown.

test/e2e/upgrade/monitor.go

@@ -107,10 +107,23 @@ func (m *versionMonitor) ShouldReboot() []string {
 	return nil
 }
 
-func (m *versionMonitor) ShouldUpgradeAbort(abortAt int) bool {
+func (m *versionMonitor) ShouldUpgradeAbort(abortAt int, desired configv1.Update) bool {
 	if abortAt == 0 {
 		return false
 	}
+
+	if abortAt == 100 {
+		if m.lastCV == nil {
+			return false // wait for m.Check() to populate a ClusterVersion
+		}
+
+		if len(m.lastCV.Status.History) == 0 {
+			return false // wait for the cluster-version operator to populate history
+		}
+		history := m.lastCV.Status.History[0]
+		return history.Image == desired.Image && history.State == configv1.CompletedUpdate
+	}
+
 	coList, err := m.client.ConfigV1().ClusterOperators().List(context.Background(), metav1.ListOptions{})
 	if err != nil {
 		framework.Logf("Unable to retrieve cluster operators, cannot check completion percentage")

test/e2e/upgrade/upgrade.go

@@ -544,7 +544,7 @@ func clusterUpgrade(f *framework.Framework, c configv1client.Interface, dc dynam
 					return false, err
 				}
 
-				if !aborted && monitor.ShouldUpgradeAbort(abortAt) {
+				if !aborted && monitor.ShouldUpgradeAbort(abortAt, desired) {
 					framework.Logf("Instructing the cluster to return to %s / %s", original.Status.Desired.Version, original.Status.Desired.Image)
 					desired = configv1.Update{
 						Image: original.Status.Desired.Image,