fix(olm): Fix the issue with missing events due to rate limit

Event recorder has a rate limit (defaultSpamBurst) 25 events per
10 minutes. After the limit is reached, refill rate reduces to
1 event per 5 minutes which causes some events to get lost.
Currently implementation allows CSV to be stuck in Pending phase
for a long time if there are some missing requirements. As a result,
during sync cycle, CSV continues to spam `CSVReasonRequirementsNotMet`
events. Now, the code is fixed to not send any identical events that
match current state of CSV.

Signed-off-by: Vu Dinh <[email protected]>

Author: dinhxuanvu

pkg/controller/operators/olm/operator.go

@@ -884,23 +884,23 @@ func (a *Operator) transitionCSVState(in v1alpha1.ClusterServiceVersion) (out *v
 
 		if !met {
 			logger.Info("requirements were not met")
-			out.SetPhaseWithEvent(v1alpha1.CSVPhasePending, v1alpha1.CSVReasonRequirementsNotMet, "one or more requirements couldn't be found", now, a.recorder)
+			out.SetPhaseWithEventIfChanged(v1alpha1.CSVPhasePending, v1alpha1.CSVReasonRequirementsNotMet, "one or more requirements couldn't be found", now, a.recorder)
 			syncError = ErrRequirementsNotMet
 			return
 		}
 
 		// Check for CRD ownership conflicts
 		if syncError = a.crdOwnerConflicts(out, a.csvSet(out.GetNamespace(), v1alpha1.CSVPhaseAny)); syncError != nil {
 			if syncError == ErrCRDOwnerConflict {
-				out.SetPhaseWithEventIfChanged(v1alpha1.CSVPhaseFailed, v1alpha1.CSVReasonOwnerConflict, syncError.Error(), now, a.recorder)
+				out.SetPhaseWithEvent(v1alpha1.CSVPhaseFailed, v1alpha1.CSVReasonOwnerConflict, syncError.Error(), now, a.recorder)
 			}
 			return
 		}
 
 		// Check for APIServices ownership conflicts
 		if syncError = a.apiServiceOwnerConflicts(out); syncError != nil {
 			if syncError == ErrAPIServiceOwnerConflict {
-				out.SetPhaseWithEventIfChanged(v1alpha1.CSVPhaseFailed, v1alpha1.CSVReasonOwnerConflict, syncError.Error(), now, a.recorder)
+				out.SetPhaseWithEvent(v1alpha1.CSVPhaseFailed, v1alpha1.CSVReasonOwnerConflict, syncError.Error(), now, a.recorder)
 			}
 			return
 		}
@@ -943,7 +943,7 @@ func (a *Operator) transitionCSVState(in v1alpha1.ClusterServiceVersion) (out *v
 		} else {
 			// Set phase to failed if it's been a long time since the last transition (5 minutes)
 			if metav1.Now().Sub(out.Status.LastTransitionTime.Time) >= 5*time.Minute {
-				out.SetPhaseWithEventIfChanged(v1alpha1.CSVPhaseFailed, v1alpha1.CSVReasonInstallCheckFailed, fmt.Sprintf("install timeout"), now, a.recorder)
+				out.SetPhaseWithEvent(v1alpha1.CSVPhaseFailed, v1alpha1.CSVReasonInstallCheckFailed, fmt.Sprintf("install timeout"), now, a.recorder)
 			}
 		}