[ws-manager] fix the possibility the volume restore time is incorrect

We might fall into the exponential backoff function, meaning that the
volume restore time would be calculated again even if we had already
calculated the volume restore time.

Signed-off-by: JenTing Hsiao <[email protected]>

Author: jenting

components/ws-manager/pkg/manager/manager.go

@@ -215,9 +215,9 @@ func (m *Manager) StartWorkspace(ctx context.Context, req *api.StartWorkspaceReq
 	span.LogKV("event", "pod description created")
 
 	var (
-		createPVC         bool
-		pvc               *corev1.PersistentVolumeClaim
-		volumeRestoreTime time.Time
+		createPVC          bool
+		pvc                *corev1.PersistentVolumeClaim
+		startTime, endTime time.Time
 	)
 	for _, feature := range startContext.Request.Spec.FeatureFlags {
 		if feature == api.WorkspaceFeatureFlag_PERSISTENT_VOLUME_CLAIM {
@@ -235,7 +235,7 @@ func (m *Manager) StartWorkspace(ctx context.Context, req *api.StartWorkspaceReq
 		if err != nil && !k8serr.IsAlreadyExists(err) {
 			return nil, xerrors.Errorf("cannot create pvc object for workspace pod: %w", err)
 		}
-		volumeRestoreTime = time.Now()
+		startTime = time.Now()
 	}
 
 	// create the Pod in the cluster and wait until is scheduled
@@ -276,8 +276,9 @@ func (m *Manager) StartWorkspace(ctx context.Context, req *api.StartWorkspaceReq
 			hist, err := m.metrics.volumeRestoreTimeHistVec.GetMetricWithLabelValues(wsType, req.Spec.Class)
 			if err != nil {
 				log.WithError(err).WithField("type", wsType).Warn("cannot get volume restore time histogram metric")
-			} else {
-				hist.Observe(time.Since(volumeRestoreTime).Seconds())
+			} else if endTime.IsZero() {
+				endTime = time.Now()
+				hist.Observe(endTime.Sub(startTime).Seconds())
 			}
 		}